Download?Doi=10.1.1.8.8094.Pdf

STOCK MARKET DECLINES AND MARKET FUNDAMENTALS

HO YEW JOEMalaya of

FACULTY OF ECONOMICS AND ADMINISTRATION UNIVERSITY OF MALAYA KUALA LUMPUR University 2017 STOCK MARKET DECLINES AND MARKET FUNDAMENTALS

Malaya HO YEWof JOE

THESIS SUBMITTED AS A FULFILLMENT FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

FACULTYUniversity OF ECONOMICS AND ADMINISTRATION 2017 UNIVERSITI MALAYA

ORIGINAL LITERARY WORK DECLARATION

Name of Candidate: Ho Yew Joe (I.C./Passport No.:

Registration/Matrix No.: EHA 100002

Name of Degree: Doctor of Philosophy

Title of Project Paper/Research Report/Dissertation/Thesis (“this Work”):

Field of Study:

I do solemnly and sincerely declare that:

(1) I am the sole author/writer of this Work; (2) This work is original; (3) Any use of any work in which copyright exists was done by way of fair dealing and for permitted purposes and any excerpt or extract from, or reference to or reproduction of any copyright work has been disclosed expressly and sufficiently and the title of the Work and its authorship haveMalaya been acknowledged in this Work; (4) I do not have any actual knowledge nor do I ought reasonably to know that the making of this work constitutes an infringement of any copyright work; (5) I hereby assign all and every rights in theof copyright to this Work to the University of Malaya (“UM”), who henceforth shall be owner of the copyright in this Work and that any reproduction or use in any form or by any means whatsoever is prohibited without the written consent of UM having been first had and obtained; (6) I am fully aware that if in the course of making this Work I have infringed any copyright whether intentionally or otherwise, I may be subject to legal action or any other action as may be determined by UM.

Candidate’sUniversity Signature Date Subscribed and solemnly declared before,

Witness’s Signature Date Name: Designation:

ii ABSTRACT

This research investigates if market fundamentals are significant in predicting stock market declines in an ex-ante fashion. The dates of stock market declines are determined via a list of ex-post models which comprise the approaches of parametric, non-parametric, semi- parametric and scale-invariant. This includes the Markov-switching model (with various extractions of probabilities for predictability tests), naïve moving average, Brys-Boschan algorithm (the Lunde & Timmermann variant and Candelon, Piplak & Straetman variant), and the integrated identifications with the JLS model and JLS "negative bubble" model.

This research also improvises the existing methodologies and introduces the semi- parametric model of “naïve moving average negative return” in identifying bear market and in using the “dichotomised smoothed probabilities” whichMalaya is transformed from the output of the Markov-switching model for predictabilityof tests. Market fundamentals such as dividend growth, change in the cyclically adjusted price earnings ratio, change in

Commerce Department composite index of 11 leading indicators, term spreads (both 3M-

10Y and 5M-10Y), Chicago Fed National Activity Index and ISM manufacturing survey

(inventories index) are found to be among the most consistent best predictors for stock market declines. The significance of these market fundamentals varies according to different forecasting horizons. University

iii

ABSTRAK

Kajian ini menyelidik sama ada asas-asas pasaran adalah berkesan dalam meramal penurunan pasaran saham secara yang ex-ante. Tempoh-tempoh penurunan pasaran saham adalah ditentukan melalui satu senarai model ex-post yang merangkumi pendekatan parametrik, bukan parametrik, semi-parametrik dan skala tak berubah. Ini termasuk model

Markov-switching (dengan pelbagai pengekstrakan kebarangkalian untuk ujian ramalan), purata bergerak naif, algoritma Brys-Boschan (varian Lunde & Timmermann dan varian

Candelon, Piplak & Straetman), dan pengenalpastian bersepadu dengan model JLS & model JLS "gelembung negatif". Kajian ini juga menambahbaikan metodologi yang sedia ada dan memperkenalkan model separuh parametrikMalaya "purata bergerak naif pulangan negatif" untuk mengenalpasti pasaran merundum dan menggunakan "kebarangkalian dichotomized terlancar" yang diubahsuai daripadaof output model Markov-switching untuk ujian ramalan. Asas-asas pasaran seperti pertumbuhan dividen, perubahan dalam nisbah pendapatan harga diselaraskan secara kitaran, perubahan dalam indeks komposit 11 penunjuk utama Jabatan Perdagangan, “spread” tempoh (bagi kedua-dua 3M-10y dan 5M-

10y), Indeks Aktiviti Negara Chicago Fed dan kaji selidik pengeluar ISM (index inventori) adalah didapati antara peramal yang terbaik yang paling konsisten untuk meramal penurunan pasaran saham. Keberkesanan asas-asas pasaran ini berubah mengikut jangka masaUniversity ramalan yang berbeza.

AKNOWLEDGEMENTS

“Boundless misery, fleeting triumph and a lifetime of fulfilment” - this phrase pretty much encapsulates how I feel after putting the last word in my thesis. It has been a long and lonely journey in this pitch-dark tunnel of which I am glad I am finally out of. Throughout the journey, there was almost not a single day that I did not think of my late father, Edward

Ho Ah Hah @ Hoh Chee Wai and Wow Wow, the only dog I ever had. The both of you pulled me through it. Pa, I miss you and I hope I have made you proud. I wish the both of you eternal happiness, wherever you are.

I am grateful to my supervisor. Dr. Mario Arturo Ruiz Estrada who never stopped believing in me. Not to forget my co-supervisor too, MalayaAssoc. Prof. Dr. Yap Su Fei for her support. I am also indebted to Prof. Dr. Didier Sornette for his sharing of literatures with data that are critical to this thesis. I am alsoof extremely thankful to Prof. Dr. Chen Shiu-

Sheng for sharing with me a crucial journal on nested models. It was a privilege to engage in numerous insightful discussions with Dr. Lim Kian Ping and Dr. Quah Chee Heong. I am very thankful to Nicholas Teong Khan Vun for teaching me programming using Matlab.

Also, I would like to thank my sister Dr. Ho Yuen Yee and her partner Grant Boland for their kind assistance in proofreading my work in the first round. University I am also thankful to my wife Ooi Hui Qing for her patience and love, and to my family as well as her family. Last but not least, I am extremely grateful to my colleague and close friend Nurhamiza Mumin who was working on her Ph.D too, for the endless hours she endured listening to my agony in finishing this thesis.

TABLE OF CONTENTS

TITLE ...... i

ORIGINAL LITERARY WORK DECLARATION ...... ii

ABSTRACT ...... iii

ABSTRAK ...... iv

ACKNOWLEDGEMENTS ...... v

TABLE OF CONTENTS ...... vi

LIST OF FIGURES ...... xiii

LIST OF TABLES ...... xvi LIST OF ABBREVIATIONS ...... Malaya xviii LIST OF APPENDICES ...... xxi CHAPTER 1: INTRODUCTION ...... of 1 1.0 Introduction ...... 1 1.1 Background of Research ...... 2 1.2 Problem Statements ...... 4 1.3 Research Questions ...... 10 1.4 Research Objectives ...... 11 1.5 Scientific Paradigm ...... 12 1.6 Scope of Research ...... 13 1.7 Definitions of Key Terms...... 13 1.7.1 Definitions of Bull and Bear Markets ...... 14 University1.7.2 Definitions of Stock Market Crashes ...... 15 1.8 Significance of Research ...... 16 1.9 Outline of Research ...... 18 1.10 Concluding Remark ...... 20

CHAPTER 2: LITERATURE REVIEW ...... 22 2.0 Introduction ...... 22 2.1 History and Conventional Worldviews of Research ...... 23 2.1.1 Etymology of Stock Market Declines ...... 24 2.1.2 Origin of Stock Market Prediction ...... 29 2.1.3 Random Walk Theory ...... 35 2.1.4 Efficient Market Hypothesis ...... 38 2.1.5 Divergence of Theories and Practices ...... 41 2.1.6 Stock Market Bubbles ...... 43 2.1.7 Business Cycle and Stock Market ...... 48 2.1.8 Market Fundamentals and Anomalies ...... 50 2.2 Advancement and Alternative Worldviews of Research ...... 52 2.2.1 Departure from Conventional Worldviews ...... 52 2.2.2 Adaptive Market Hypothesis ...... 54 2.2.3 Introduction to the Complex Systems TheoryMalaya ...... 55 2.2.4 Etymology of the Complex Systems Theory ...... 59 2.2.5 Measurement without Theoryof ...... 63 2.2.6 Complexity of Economic ...... 64 2.2.7 Evolution of the Complex Systems Theory in Economics ...... 67 2.2.8 Complexity of Financial Markets and Stock Market Crashes ...... 69 2.3 Classification and Evaluation of Literature...... 72 2.3.1 Post-mortem Reviews and Theoretical Studies ...... 73 2.3.2 Empirical Studies I: Defining Stock Market Regimes ...... 89 2.3.3 Empirical Studies II: Predicting Stock Market Returns / Regimes ...... 94 2.4 Theoretical Framework ...... 102 2.5 Concluding Remark ...... 107 University

CHAPTER 3: METHODOLOGY ...... 113 3.0 Introduction ...... 113 3.1 Empirical Framework...... 113 3.2 Data and Variables ...... 119 vii

3.2.1 Shiller’s Financial Variables ...... 121 3.2.2 Estrella & Miskin’s Financial Variables ...... 123 3.2.3 Innovation to Data and Variables and Justifications ...... 124 3.3 Description of Test Variables...... 128 3.3.1 Dividends [Market Aggregate] ...... 129 3.3.2 Earnings [Market Aggregate] ...... 130 3.3.3 Real Dividends [Market Aggregate] ...... 131 3.3.4 Real Earnings [Market Aggregate] ...... 131 3.3.5 Cyclically Adjusted Price Earnings Ratio ...... 132 3.3.6 3-month Treasury Bill ...... 132 3.3.7 5-year Treasury Note ...... 133 3.3.8 10-year Treasury Note ...... 134 3.3.9 5-year – 3-month Term Spreads ...... 134 3.3.10 10-year – 3-month Term Spreads ...... 136 3.3.11 Money Supply M1 [Seasonally Adjusted]Malaya ...... 136 3.3.12 Money Supply M2 [Seasonally Adjusted] ...... 138 3.3.13 Real Money Supply M1 [Seasonallyof Adjusted] ...... 138 3.3.14 Real Money Supply M2 [Seasonally Adjusted] ...... 139 3.3.15 Consumer Price Index ...... 139 3.3.16 Growth in Real GDP [Lagged 1 Quarter] ...... 141 3.3.17 ISM Manufacturer Survey (Inventories Index) ...... 143 3.3.18 Contracts and Orders for Plant and Equipment [Seasonally Adjusted] ...... 145 3.3.19 New Private Housing Permits [Seasonally Adjusted] ...... 147 3.3.20 Change in Manufacturers’ Unfilled Durable Goods Orders [Seasonally Adjusted] ...... 148 University3.3.21 Purchasing Managers’ Index ...... 149 3.3.22 University of Michigan Consumer Sentiment Index ...... 151 3.3.23 Commerce Department Composite Index of 11 Leading Indicators [Seasonally Adjusted] ...... 153 3.3.24 Chicago Fed National Activity Index ...... 155 3.4 Specifications for Bear Markets and Bull Markets ...... 157

viii

3.4.1 Parametric Approach: Markov-switching (Filtered Probabilities & Smoothed Probabilities) ...... 158 a. Transition Probabilities ...... 159 b. Probabilities for Parameters Estimation ...... 160 3.4.2 Semi-Parametric Approach: Markov-switching (Dichotomized Smoothed Probabilities) ...... 163 3.4.3 Semi-Parametric Approach: Naïve Moving Average ...... 163 3.4.4 Semi-parametric Approach: Naïve Moving Average Negative Return ...... 164 3.4.5 Non-parametric Approach: Lunde & Timmermann’s B-B Algorithm (2004) Variant ...... 166 3.4.6 Non-parametric Approach: Candelon, Piplak & Straetmans’ B-B Algorithm (2008) Variant ...... 169 3.5 Specification of Crashes and Rebounds ...... 169 3.5.1 Scale-invariant Approach: JLS Model and JLS “Negative Bubbles” Model ...... Malaya 170 3.6 Predictive Regression Model for In-sample and Out-of-sample Tests for Parametric Approaches ...... 171 3.7 Probit Model for In-sample and Out-of-sampleof Tests for Semi-parametric, Non-parametric and Econophysics Approaches ...... 174 3.8 Concluding Remark ...... 175

CHAPTER 4: RESULTS ...... 178 4.0 Introduction ...... 178

4.1 Descriptive Statistics ...... 180 4.2 University Unit Root Tests for Stationarity and Data Transformation ...... 181 4.3 Findings of Models for Stock Market Declines ...... 185

4.3.1 Markov-switching Summary Outputs ...... 185 4.3.2 Markov-switching Filtered Probabilities Result ...... 188 4.3.3 Markov-switching Smoothed Probabilities Result ...... 189 4.3.4 Markov-switching Dichotomized Smoothed Probabilities Result ...... 190 4.3.5 Naïve Moving Average Result ...... 191 ix

4.3.6 Naïve Moving Average Negative Return Result ...... 192 4.3.7 Lunde & Timmermann’s B-B Algorithm Result ...... 193 4.3.8 Candelon, Piplak & Straetmans’ B-B Algorithm Result ...... 194 4.3.9 JLS’ Crashes Result ...... 195 4.3.10 JLS’ Negative Bubbles Rebounds Result ...... 196 4.4 Observations for Out-of-sample Tests ...... 198

4.5 In-sample Predictability Test Results ...... 199

4.5.1 Markov-switching Filtered Probabilities In-sample Predictability Test Results ...... 200 4.5.2 Markov-switching Smoothed Probabilities In-sample Predictability Test Results ...... 204 4.5.3 Markov-switching Dichotomized Smoothed Probabilities In-sample Predictability Test Results ...... 208 4.5.4 Naïve Moving Average In-sample Predictability Test Results ...... 213 4.5.5 Naïve Moving Average Negative ReturnMalaya In-sample Predictability Test Results ...... 218 4.5.6 Lunde & Timmermann’s B-B Algorithm In-sample Predictability Test Results ...... of 223 4.5.7 Candelon, Piplak & Straetmans’ B-B Algorithm In-sample Predictability Test Results ...... 228 4.5.8 JLS and JLS “Negative Bubbles” Integrated Identifications In-sample Predictability Test Results ...... 233 4.6 Out-of-sample Predictability Test Results ...... 237

4.6.1 Markov-switching Filtered Probabilities Out-of-sample Predictability Test Results ...... 238 4.6.2 Markov-switching Smoothed Probabilities Out-of-sample Predictability Test Results ...... 240 University4.6.3 Markov-switching Dichotomized Smoothed Probabilities Out-of- sample Predictability Test Results ...... 242 4.6.4 Naïve Moving Average Out-sample Predictability Test Results ...... 243 4.6.5 Naïve Moving Average Negative Return Out-of-sample Predictability Test Results ...... 244 4.6.6 Lunde & Timmermann’s B-B Algorithm Out-of-sample Predictability Test Results ...... 246

4.6.7 Candelon, Piplak & Straetmans’ B-B Algorithm Out-of-sample Predictability Test Results ...... 247 4.6.8 JLS and JLS “Negative Bubbles” Integrated Identifications Out-of- sample Predictability Test Results ...... 248 4.7 Concluding Remark ...... 249

CHAPTER 5: DISCUSSION AND CONCLUSION ...... 253 5.0 Introduction ...... 253

5.1 Recapitulation ...... 254

5.2 Discussion ...... 256

5.2.1 Markov-switching Filtered Probabilities ...... 261 5.2.2 Markov-switching Smoothed Probabilities ...... 262 5.2.3 Markov-switching Dichotomized SmoothedMalaya Probabilities ...... 263 5.2.4 Naïve Moving Average ...... 264 5.2.5 Naïve Moving Average Negativeof Return ...... 265 5.2.6 Lunde & Timmermann’s B-B Algorithm ...... 266 5.2.7 Candelon, Piplak & Straetmans’ B-B Algorithm ...... 267 5.2.8 JLS and JLS “Negative Bubbles” Integrated Identifications ...... 268 5.2.9 Predictive Consistency of Test Variables ...... 269 5.3 Accomplishment of Research Objectives ...... 270

5.4 Implications of the Research ...... 271

5.4.1 Market Predictability for Traders’ Consideration ...... 271 5.4.2 Theory and Knowledge Development ...... 274 University5.4.3 Policy Consideration ...... 280 5.5 Limitations of Research ...... 281

5.6 Suggestions for Future Study ...... 283

5.7 Conclusion ...... 285

REFERENCES ...... 289

APPENDIX A ...... 360

APPENDIX B ...... 364

LIST OF PUBLICATIONS ...... 390

Malaya

University

xii

LIST OF FIGURES

Figure 2.1: Theoretical Framework ...... 103 Figure 3.1: Empirical Framework...... 115 Figure 4.1: Markov-switching Results ...... 187 Figure 4.2: Markov-switching Filtered Probabilities Bear Markets ...... 188 Figure 4.3: Markov-switching Smoothed Probabilities Bear Markets ...... 189 Figure 4.4: S&P 500 Index vs Markov-switching Dichotomized Smoothed Probabilities in Bear Markets ...... 190 Figure 4.5: S&P 500 Index vs Naïve Moving Average Bear Markets ...... 191 Figure 4.6: S&P 500 Index vs Naïve Moving Average Negative Return Bear Markets ...... 192 Figure 4.7: S&P 500 Index vs Lunde & Timmermann’s B-B Algorithm Bear Markets ...... Malaya...... 193 Figure 4.8: S&P 500 Index vs Candelon, Piplak & Straetmans’ B-B Algorithm Bear Markets ...... of ...... 194 Figure 4.9: S&P 500 Index vs JLS’ Crashes (Turning Points and Limited Windows) ...... 195 Figure 4.10: S&P 500 Index vs JLS’ Negative Bubbles Rebounds (Turning Points and Limited Windows) ...... 196 Figure 5.1: General Categorisation of Theories on the Espousal of Rationality Assumption ...... 278 Figure 5.2: General Categorisation of Theories on the Espousal of the Bubbles Proposition ...... 279 Figure B-1: S&P 500 Index and Real S&P 500 Index ...... 364 Figure B-2: S&P 500 Index and Return (%) ...... 365 FigureUniversity B-3: Real S&P 500 Index and Return (%) ...... 365 Figure B-4: Dividends ($) and S&P 500 Index ...... 366 Figure B-5: Earning ($) and S&P 500 Index ...... 366 Figure B-6: Real Dividends ($) and S&P 500 Index ...... 367 Figure B-7: Real Earnings ($) and S&P 500 Index ...... 367 Figure B-8: Cyclically Adjusted Price Earnings ($) and S&P 500 Index ...... 368

xiii

Figure B-9: Consumer Price Index and S&P 500 Index ...... 368 Figure B-10: 3-Month Treasury Bill Yield (%) and S&P 500 Index ...... 369 Figure B-11: 5-Year Treasury Note Yield (%) and S&P 500 Index ...... 369 Figure B-12: 10-Year Treasury Note Yield (%) and S&P 500 Index ...... 370 Figure B-13: 10-Year – 3-Month Yield Curve Rate (%) and S&P 500 Index ...... 370 Figure B-14: 5-Year – 3-Month Yield Curve Rate (%) and S&P 500 Index ...... 371 Figure B-15: M1 (SA, $ Bil) and S&P 500 Index ...... 371 Figure B-16: M2 (SA, $ Bil) and S&P 500 Index ...... 372 Figure B-17: RM1 (SA, $ Bil) and S&P 500 Index ...... 372 Figure B-18: RM2 (SA, $ Bil) and S&P 500 Index ...... 373 Figure B-19: ISM Manufacturers Survey: Inventories Index and S&P 500 Index 373 Figure B-20: Contracts and Orders for Plant and Equipment (SA, $ Mil) and S&P 500 Index ...... 374 Figure B-21: New Private Permits (SA, ‘000) and S&P 500 Index ...... 374 Figure B-22: Change in Manufacturers’ Unfilled DurableMalaya Goods Orders (SA, $ Mil) and S&P 500 Index ...... 375 Figure B-23: Purchasing Managers’ Index and S&P 500 Index ...... 375 Figure B-24: University of Michigan Consumerof Sentiment Index and S&P 500 Index ...... 376 Figure B-25: Growth in Real GDP (%) and S&P 500 Index ...... 376 Figure B-26: Commerce Department Composite Index of 11 Leading Indicators (SA) and S&P 500 Index ...... 377 Figure B-27: Chicago Fed Activity Index and S&P 500 Index ...... 377 Figure B-28: Dividends ($) and Return (%) ...... 378 Figure B-29: Earnings ($) and Return (%) ...... 378 Figure B-30: Real Dividends ($) and Return (%) ...... 379 Figure B-31: Real Earnings ($) and Return (%) ...... 379 FigureUniversity B-32: Cyclically Adjusted Price Earnings ($) and Return (%) ...... 380 Figure B-33: Consumer Price Index and Return (%) ...... 380 Figure B-34: 3-Month Treasury Bill Yield (%) and Return (%) ...... 381 Figure B-35: 5-Year Treasury Note Yield (%) and Return (%) ...... 381 Figure B-36: 10-Year Treasury Note Yield (%) and Return (%) ...... 382 Figure B-37: 5-Year – 3 Month Yield Curve Rate (%) and Return (%) ...... 382

xiv

Figure B-38: 10-Year – 3-Month Yield Curve Rate (%) and Return (%) ...... 383 Figure B-39: M1 (SA, $ Bil) and Return (%) ...... 383 Figure B-40: M2 (SA, $ Bil) and Return (%) ...... 384 Figure B-41: RM1 (SA, $ Bil) and Return (%) ...... 384 Figure B-42: RM2 (SA, $ Bil) and Return (%) ...... 385 Figure B-43: ISM Manufacturers Survey: Inventories Index and Return (%) ...... 385 Figure B-44: Contracts and Orders for Plant and Equipment (SA, $ Mil) and Return (%) ...... 386 Figure B-45: New Private Housing Permits (SA, ‘000) and Return (%) ...... 386 Figure B-46: Change in Manufacturers’ Unfilled Durable Goods Orders (SA, $ Mil) and Return (%) ...... 387 Figure B-47: Purchasing Managers’ Index and Return (%) ...... 387 Figure B-48: University of Michigan Consumer Sentiment Index and Return (%) ...... 388 Figure B-49: Growth in Real GDP (%) and Return (%) ...... 388 Figure B-50: Commerce Department Composite IndexMalaya of 11 Leading Indicators (SA) and Return (%) ...... 389 Figure B-51: Chicago Fed National Activityof Index and Return (%) ...... 389

University

LIST OF TABLES

Table 1.1: Dates for Largest Percentage (%) Losses of S&P 500 Index in a Single Day ...... 9 Table 3.1: Business Cycle and S&P 500 Index ...... 117 Table 3.2: Overview of Test Variables ...... 127 Table 4.1: Summary Statistics of Variables ...... 180 Table 4.2: Unit Root Tests: Pre-transformed Variables ...... 181 Table 4.3: Unit Root Tests: Transformed Variables for Stationarity ...... 182 Table 4.4: Summary Statistics of Transformed Variables ...... 184 Table 4.5: Summary of Markov-switching Results...... 185 Table 4.6: In-sample Observations & Out-of-sample Observations for Out-of- sample Tests (Clark & West, 2007) ...... 198 Table 4.7: In-sample Predictability Test Results for Parametric Model: Markov- switching Filtered Probabilities ...... Malaya...... 200 Table 4.8: In-sample Predictability Test Results for Parametric Model: Markov- switching Smoothed Probabilitiesof ...... 204 Table 4.9: In-sample Predictability Test Results for Semi-parametric Model: Markov-switching Dichotomized Smoothed Probabilities ...... 208 Table 4.10: In-sample Predictability Test Results for Semi-parametric Model: Naïve Moving Average ...... 213 Table 4.11: In-sample Predictability Test Results for Semi-parametric Model: Naïve Moving Average Negative Return ...... 218 Table 4.12: In-sample Predictability Test Results for Non-parametric Model: Lunde & Timmermann’s B-B Algorithm ...... 223 Table 4.13: In-sample Predictability Test Results for Non-parametric Model: Candelon, Piplak & Straetman’s B-B Algorithm ...... 228 TableUniversity 4.14: In-sample Predictability Test Results for Scale-invariant Model: JLS and JLS “Negative Bubbles” (6-month Window) ...... 233 Table 4.15: Out-of-sample Predictability Test Results for Parametric Model: Markov-switching Filtered Probabilities...... 238 Table 4.16: Out-of-sample Predictability Test Results for Parametric Model: Markov-switching Smoothed Probabilities ...... 240

xvi

Table 4.17: Out-of-sample Predictability Test Results for Semi-parametric Model: Markov-switching Dichotomized Smoothed Probabilities ...... 242 Table 4.18: Out-of-sample Predictability Test Results for Semi-parametric Model: Naïve Moving Average ...... 243 Table 4.19: Out-of-sample Predictability Test Results for Semi-parametric Model: Naïve Moving Average Negative Return ...... 244 Table 4.20: Out-of-sample Predictability Test Results for Non-parametric Model: Lunde & Timmermann’s B-B Algorithm ...... 246 Table 4.21: Out-of-sample Predictability Test Results for Non-parametric Model: Candelon, Piplak & Straetman’s B-B Algorithm ...... 247 Table 4.22: Out-of-sample Predictability Test Results for Scale-invariant Model: JLS and JLS “Negative Bubbles” (6-month Window) ...... 248 Table 5.1: Summary of In-sample Results for all Models: Highest Goodness of Fit, 푅2 or Pseudo- 푅2 for All Horizons ...... 259 Table 5.2: Summary of Out-of-sample Results of all Models: Highest MSPE-adj or Lowest QPS for All Horizons ...... 260 Table A-1: Data Sources...... Malaya...... 360

University

xvii

LIST OF ABBREVIATIONS

ADF : Augmented Dickey-Fuller AIC : Akaike information criterion AMEX : American Stock Exchange AMH : Adaptive market hypothesis B-B : Brys-Boschan BN : Billion C-CAPM : Consumption capital asset pricing model C-CPI-U : Chained consumer price index for all urban consumers CAPE : Cyclically adjusted price earnings ratio CAPM : Capital asset pricing model CCI : Conference Board’s CompositeMalaya Coincident Index CDLI : Commerce Department composite index of 11 leading indicators [seasonally adjusted] CFNAI : Chicago Fed Nationalof Activity Index CMUO : Change in manufacturers’ unfilled durable goods orders COPE : Contracts and orders for plant and equipment CPI : Consumer price index CPI-U : Consumer price index for all urban consumers CPI-W : Consumer price index for urban wage earners and clerical workers D/P : Dividend yields DF-GLS : Dickey-Fuller generalized least square DJIA : Dow Jones Industrial Average UniversityDVD : Dividends E/P : Earnings-price EAR : Earnings EMH : Efficient market hypothesis FOREX : Foreign exchange G-10 : Group of Ten xviii

GARCH : Generalized autoregressive conditional heteroskedasticity GDP : Gross domestic product GMM : Generalized method of moments INFL : Inflation rate IP : Industrial Production ISM : Institute of Supply Management ISMI : ISM manufacturing survey (inventories index) JLS : Johansen-Lediot-Sornette M : Million M1 : Money Supply M1 M2 : Money Supply M2 M2 : Money Supply M3 MS : Markov-switching MSCI : University of Michigan ConsumerMalaya Sentiment Index MSPE : Mean squared prediction error MTAR : Momentum thresholdof autoregressive NASDAQ: National Association of Securities Dealers Automated Quotations NBER : National Bureau of Economic Research NAPM : National Association of Purchasing Managers NPHP : New private housing permits NYSE : New York Stock Exchange P/B : Price-to-book P/E : Price-per-earnings PMI : Purchasing Managers’ Index UniversityPh.D : Doctor of philosophy degree PP : Phillips-Perron QPS : Quadratic probability score REAR : Real earnings RDVD : Real dividends RGDPG : Growth in real GDP RM1 : Real Money Supply M1 xix

RM2 : Real Money Supply M2 RT100 : Return in percentage S5Y3M : 5-year – 3-month term spreads S10Y3M : 10-year – 3-month term spreads S&P 500 : Standard and Poor’s 500 SA : Seasonally adjusted SIC : Schwarz information criterion T-bills : Treasury bills TB3M : 3-month Treasury bill TN5Y ; 5-year Treasury Note TN10Y : 10-year Treasury Note TIPS : Treasury Inflation Protected Securities UC-MS : Unobserved component model with Markov-switching heteroscedasticity U.S. : United States Malaya USD$ : United States Dollar VWRETD: Value-weighted return exchange-traded derivatives WWI : World war oneof

University

CHAPTER 1

INTRODUCTION

1.0 Introduction

The inaugural chapter of the research begins with a brief review on the background of this research. This section summarily outlines the core theme of the research i.e. “stock market declines” and how the development of literature in this area motivates further investigation of this research. The next section discusses the problem statements of the research followed by sections on research questions and research objectives accordingly. Malaya The introductory sections to the chapterof aims to elucidate the research gaps concisely and to emphatically establish the direction of the research. Then, the research is parameterised in sections of scientific paradigm and scope of research. These sections are aimed to limit the research within a focused area. The research is thus secured as such that the boundary of the research is clearly drawn at the inception. Any related subject, methodology or issue, for example on the selection of variables that are not included in the research frameworks are regarded extraneous, which can be considered forUniversity future studies. The chapter then proceeds to the definitions of key terms and is followed by a section on the significance of the research. Next the chapter lays out the outline of the research that summarily describes the content of the following chapters and ends with the section of concluding remark.

1.1 Background of Research

According to Gonzalez, Powell, Shi & Wilson, (2005), the notion of random walk is deeply ingrained in most contemporary literature on the movement of stock prices and indices. Nonetheless, one area of forecasting in the financial economics that is gaining prominence is the study on the cyclical heterogeneous durations of returns in the stock market (e.g., see Coakley & Fuertes, 2006: and Perez-Quiros & Timmermann, 2000).

Pagan & Sossounov (2003) is one of the earliest studies to specify a set of algorithm on the change of regimes of the bull and bear markets.

The concept found its roots from the cyclical market school of thought and the algorithm introduced by Pagan & Sossounov (2003) was a modification to the Brys-

Boschan algorithm (B-B algorithm hereafter), which is used for dating business cycles.

Other more important literature on the identificationMalaya of bull and bear markets that emerged subsequently include (but are notof limited to) Cunado, Gil-Alana, de Gracia (2010), Gonzalez, Powell, Shi & Wilson (2005) and Maheu, McCurdy & Song (2009).

The development of the study on the bull and bear markets coincides with the upsurge of interest in the log periodic power law (LPPL hereafter) based on the

Johansen-Lediot-Sornette (JLS hereafter) model introduced by the interdisciplinary econophysics that espouse the complex system theory. The JLS model is a ground- breaking methodology for the identification of bubble-induced asset market crashes (exogenousUniversity crashes). Some of the pioneering literature in this area include Johansen & Sornette (1999a; 1999b; 1999c) and Sornette & Johansen, 1997; 1998; etc.

The formal definitions to bear markets and stock market crashes are mathematically represented and functionalised through time series models. Their specifications and distinctions are detailed in Chapter 3. References to both the bear markets and stock market crashes and their theoretical synonyms as postulated by 2 different schools of thought are coalesced and termed as “stock market declines”. Thus in numerous parts of the research, stock market declines will be used in its broadest sense to refer to all downward trends of indices or stock prices notwithstanding the degree of the steepness and the duration of the downturns (except in contexts that call for the specific terms to be distinguished).

The earliest term of “stock market declines” can be found in the literature entitled The Stock Market Barometer by Hamilton (1998), originally published in 1922.

William Peter Hamilton was a renowned economist and journalist who served as the editor to The Wall Street Journal. “Stock market declines” was the favoured term used by Hamilton to describe the downward movements of stock market even in his newspaper reporting prior to the publication of the literature aforementioned.

The word “decline” in reference to stock Malaya market can be interpreted as “to become less in amount”, e.g. prices declinedof or “to slope downward”, e.g. downward trend of market movement (Merriam Webster’s, 2002). Thus, the research reckons that the use of the term “stock market decline” in this research is justified, as it does not explicitly illustrate the acuteness of the drop in the market. Such definitive exegesis is critical as the empirical scope of the research encompasses the “bear markets” (the more subtle and prolonged type of declines” and the “stock market crash”, the more sharp and abrupt type of declines). Further discussion on the definitions is featured in the later sectionUniversity of the chapter. Dissension of theories and scientific paradigms is often healthy as far as research is concerned. However, an un-streamlined development even up to the basic definition of key terms could pose serious issues to this area of study. The challenge is evident as at this juncture, reaching a consensus on a formal specification for the identification of stock market declines is still a work in progress. The contemporaneous development of

3 different schools of thought is in diverse directions. In most instances, literatures on the

JLS model almost never link crashes to the macroeconomic factors or cite studies from the worldview of the bear and bull markets. Vice versa, the boundary of studies on the bear markets is also exclusive from the work of the cross disciplinary econophysics.

1.2 Problem Statements

Ho, Estrada Ruiz & Yap (2017) noted that the trend of stock market declines in recent years is becoming less abrupt. Unlike the historic Black Monday crash which shaved off 20.47% of the S&P 500 Index (SP500 hereafter) in a single day on 19

October 1987, stock market declines in the last three decades have never recorded a single day loss of 10% or more. These include the Malayainstances of major declines such as the burst of the 2000 Dot-com bubble and the 2007 subprime meltdown. The declines were devastating in cumulative term as suchof that the drops tapered off for a prolonged period.

In comparison with the 1987 crash which reached its throng in just one day, the SP500 peaked at 1508.31 on 24 August 2000 during the Dot-com bubble and reached its throng on 11 March 2003 (over 30 months later) at 800.73. Similarly, the SP500 closed at highest of 1562.47 on 10 October 2007 prior to the subprime crisis before tapering off for the next 14 months to 676.53 on 9 March 2009 (S&P 500 Historical Data, Ibid.).

TheUniversity SP500 losses termed in percentage from peak to throng for both the later collapses were 46.8% and 56.7% respectively. Thus both of the “slow burn” meltdowns were by far more devastating compared to the more infamous 1987 crash. Scenarios as such compels researchers to redefine traditional specification for major stock market declines as they are not merely confined to sharp crashes.

Despite the devastating implication, stock market declines, i.e. stock market crashes and bear markets are still two vastly underdeveloped topics in financial economics.

According to numerous literatures, both of the topics have never been satisfactory explained under the dominant efficiency market hypothesis (EMH) (see Cooper, 2008;

Cunningham 2000; Gonzalez, Powell, Shi & Wilson, 2005; Johansen, 1997; Kaizoji &

Sornette, 2008). Research on stock market declines is deemed delicate as no consensus for the definitions of the two types of stock market declines i.e. bear markets and stock market crashes has been reached conclusively. As highlighted previously, the development of literature in this area thus far is scattered in various independent directions with different definitions and specifications for modelling.

Most analyses on stock market declines particularly in the early days were in the form of retrospective case studies on specific episodes.Malaya Theoretical explanation to the complexity of stock market panic (which entails crashes) was first presented by the neoclassical rational expectation school of thoughtof (mostly through the perspective of the general equilibrium theory) and the behavioural finance school of thought (through the behavioural and cognitive psychological theory). Studies on the subject through theoretical discourse by these schools of thought are still extensive at present.

The pursuit of empirical modelling on regime changing in stock markets is later explored independently by a school of thought from the macroeconomics that espouses theUniversity cyclical worldview on the stock market movement (i.e. the bull and bear markets) and the econophysics school of thought which is inspired by the complex system theory, examined the bubble-induced stock market crashes with some very novel techniques that are borrowed from physical statistics. Given the adverse and at times, devastating impact stock market declines could inflict on the economy, the research believes that a

5 further study in this area particularly on the examination on different modelling specifications of stock market declines is timely and warranted.

Fama & French (1988) and Guidolin & Timmermann, (2005) noted that the predictability of financial markets return does not necessarily imply that markets are not efficient. Likewise the proposition should accommodate the predictability of stock market declines (and vice versa). Notwithstanding, it is crucial to note that the modelling of stock market declines should not be equated to predicting stock market return, which is studied extensively for decades in the literature. The modelling of stock market declines are associated with heterogeneous durations of regimes instead of point estimation for the modelling of stock market returns.

Malkiel (2007) argued that extreme market fluctuations such as bubble and panic selling would always succumb to the “financialMalaya law of gravitation”. Thus, unrealistic prices of asset sustained for anof extended period would invariably reverse to their intrinsic values. Distortions and artificial prolongations caused by the monetary authority add further challenge to any forecasting attempt on the market. The change of regimes in stock market and their relationship with the underlying market fundamentals

(e.g. financial variables, industrial indicator, macroeconomic indicators etc.) thus merit a thorough investigation.

The inquiries as laid out above motivate the research to undertake an extensive investigationUniversity on the nexus between various specifications of stock market declines and the underlying conditions of the economy. Linkages between stock market movement and the economic condition have been well established in literature (e.g. see Basistha &

Kurov, 2008; Casarin & Trecroci, 2007; Chauvet, 1999; Smith, Sorensen & Wickens,

2006; and McCown, 2007). The reliability of macroeconomic variables in predicting the direction of the economy is also well documented (e.g. Camacho & Perez-Quiros, 2002;

Estrella & Mishkin, 1998; Layton & Daniel, 2007; and Qi, 2001). The conceptualization of the bull and bear markets in recent years has provided a conducive platform for the investigation on stock market based on the cyclical worldview, resembling approaches from the area of business cycle forecasting.

Studies that specifically investigate stock market declines and the relationship with market fundamentals are still sparse. Apart from macroeconomic leading indicators, which were used selectively in previous literatures, it is highly probable that further important predictors to stock market declines can be discovered from a more extensive study that includes stock market fundamentals, financial market fundamentals, industrial indicators, variables on the market sentiment and indexes of leading indicators. On this aspect, the research seeks to introduce a larger spectrum of market fundamentals that are previously untested in Malayaother related literatures. The development of studies on the ofbear and bull markets are largely inspired by the worldviews and methodologies from the literatures on the business cycles.

Methodologies applied by the school of thought is broadly conventional, i.e. econometric approaches that comprise parametric models, semi-parametric models and non-parametric models (e.g. see Candelon, Piplak & Straetmans (2008), Chauvet &

Potter (2000), Chen (2009), Chang (2009), Cunado, Gil-Alana & de Gracia (2010),

Edwards, Biscarri & Gracia (2003), Gonzalez, Powell, Shi & Wilson (2005), Lunde & TimmermannUniversity (2004), Maheu & McCurdy (2000), Maheu, McCurdy & Song (2009), Pagan & Sossounov (2003) etc).

In comparison, the JLS model from the cross-disciplinary econophysics utilises more avant-garde techniques which include hazard rate and the scale invariant log periodic power law which are unusual in the sphere of financial economics (e.g. see

Feigenbaum & Freund, 1996; 1998; Johansen, 1997; Johansen & Sornette, 1998a;

Johansen, Sornette & Ledoit, 1999). The parallel timeline but independent development in both schools of thought thus presents a research gap for an integrated study on a common platform that the research seeks to investigate further.

The research also aspires to contribute to the extension of some core literature that are meticulously reviewed herein, namely; Chang (2009), Chen, (2009), Lunde &

Timmermann (2004), Perez-Quiros & Timmermann (2000) – for the bull and bear markets and Feigenbaum (2001), Johansen (2004), Johansen & Sornette (2010),

Sornette & Cauwels (2014), Liberatore (2011a, 2011b) and Yan, Woodard & Sornette

(2012) – for JLS’ model of crashes and rebounds.

Crucially, the research reckons an updated study in this area which is apposite as related studies specifically on the identification of bull and bear markets (at the juncture when this research begins) only included observationsMalaya updated to year 2007, i.e. see Chang (2009), Chauvet & Potter (2000) andof Chen (2009). Table 1.1 (updated to 31 Dec 2015) strengthen the justification for the need to have an updated study in this area.

University

Table 1.1: Dates for Largest Percentage (%) Losses of S&P 500 Index in a Single Day

Rank Date Close Net Change % Change 1 19/10/1987 224.84 -57.86 -20.47 2 15/10/2008 907.84 -90.17 -9.03 3 01/12/2008 816.21 -80.03 -8.93 4 29/09/2008 1,106.42 -106.85 -8.81 5 26/10/1987 227.67 -20.55 -8.28 6 09/10/2008 909.92 -75.02 -7.62 7 27/10/1997 876.99 -64.65 -6.87 8 31/08/1998 957.28 -69.86 -6.8 9 08/01/1988 243.4 -17.67 -6.77 10 20/11/2008 752.44 -54.14 -6.71 11 28/05/1962 55.5 -3.97 -6.68 12 08/08/2011 1,119.46 -79.92 -6.66 13 26/09/1955 42.61 -3.02 -6.62 14 13/10/1989 333.65 -21.74 -6.12 15 19/11/2008 806.58 -52.54 -6.12 16 22/10/2008 896.78 -58.27 -6.1 17 14/04/2000 1,356.56 -83.95 -5.83 18 07/10/2008 996.23 -60.66 -5.74 19 26/06/1950 18.11 -1.03 -5.38 20 20/01/2009 805.22 -44.9 -5.28 Malaya Source: Yahoo! Finance. (2015) of On the technical aspect, the research through the literature review (which is discussed extensively in the following chapter) noted that results derived from the parametric model of the Markov-switching are least appropriate to be used for predictability analyses to compare with the results drawn from semi-parametric and non-parametric models (e.g. see Chen, 2009). The predictive power of a variable tested on a parametric model is measured with the adjusted mean square prediction error (MSPEUniversity-adj) while the quadratic probability score (QPS) is used to measure results from semi-parametric and non-parametric models. Thus, the comparison of test statistic is not parallel.

A cursory examination on the data also revealed that the filtered probabilities extracted from the result of the Markov-switching model alternate between regimes more frequently compared to the switching of other semi-parametric and non-parametric

9 models. The chart contrived from the results barely resembles a regime switching time series compared to other models. The frequent alternation between regimes is considered undesirable as it defeats the main purpose of regime switching modelling which is to capture the sustained heterogeneous durations of a time series.

On another note, the semi-parametric model of naïve moving average also produced more frequent switching in regimes relative to most models. The model with relatively simplistic specification is prone to be skewed by outliers that could lead to the misdiagnosis of stock market declines (and vice versa). More in-depth discussion on the shortcoming is featured in Chapter 3. Both the aforementioned incompatibility in comparing test statistics, the noisy filtered probabilities of the Markov-switching model and the shortcomings of the naïve moving average model are methodological issues that the research proposes to address. Malaya of 1.3 Research Questions

The problem statements discussed above provide the general research question as follows: Which predictors from a list of market fundamentals that comprise of stock market fundamentals, financial market fundamentals, industrial indicators, variables on the market sentiment and indexes of leading indicators, most consistently predict the regimeUniversity of stock market declines? The specific research questions are as follows: 1) What are the most consistent market fundamentals to predict the regime of bear

markets that are specified based on various models of parametric, semi-

parametric and non-parametric for a range of predetermined months ahead

within a year?

2) What are the most consistent market fundamentals to predict the stock market

crashes (contrasting with the stock market rebounds) that are specified based on

the econophysics’ scale-invariant approaches of the JLS model & JLS "negative

bubble" model (integrated identifications) for a range of predetermined months

ahead within a year?

3) What are the overall most consistent market fundamentals to predict stock

market declines across all specifications of models proposed for the research for

a range of predetermined months ahead within a year?

1.4 Research Objectives In accordance with the research questions above, theMalaya main objective of this thesis is to identify the best individual predictors of market fundamentals that consistently predict stock market declines across different specificationsof of models. The specific research objectives are:

1) To determine the most consistent market fundamentals to predict the regime of

bear markets that specified based on various models of parametric, semi-

parametric and non-parametric for 1, 3, 6, 9 and 12 months in the future.

2) To determine the most consistent market fundamentals to predict stock market

crashes (contrasting with the stock market rebounds) that are specified based on

Universitythe econophysics’ scale-invariant approaches for 1, 3, 6, 9 and 12 months in the

future.

3) To determine the overall most consistent market fundamentals to predict stock

market declines across all specifications of models proposed for the research for

1, 3, 6, 9 and 12 months in the future.

1.5 Scientific Paradigm

It is imperative for the research to define its scientific paradigm in order to establish the boundary of the study. Apart from the review of literature, the research is wholly built on the “a posteriori” empiricism paradigm which confines it purely within the sphere of modelling, statistical analysis and deduction based on the empirical evidence derived.

This sets the research apart from the scientific paradigms of other schools of thought on the same or related study on stock market declines.

Conjectures on stock market bubbles, i.e. a subject that is closely related to this research are diverse and studies in this area are prevalent. A majority of these literatures are focused on the microeconomic aspects and espouse the rationalism paradigm in examining the behaviour of agents in the market, i.e. how their actions lead to bubbles and the subsequent market corrections. The most commonMalaya schools of thought that are inclined to such research directions are theof neoclassical economics and the behavioural finance. Studies from the neoclassical economics mainly adduce their arguments based on the normativism doctrine with emphasise on rational preferences, utility maximization and action based on full information. On the other hand, studies from the behavioural finance assumes the positivism paradigm, e.g. furnishing evidence on various biasness that are ingrained in the behaviour of agents in the market that contradicts the rationality tenet of the orthodox economics theory.

UniversityAnother more common type of literature in this area is the post-mortem analysis. This approach of retrospective study on specific episode of stock market declines generally espouses the paradigm of causal inquiry. The examination of issues could be purely in a narrative form or supplemented with empirical evidence on the hindsight.

Such causal inquiry approach is also beyond the scope of this research. 1 Extensive review and commentaries on literature is featured in the following chapter. The scientific paradigm and research scope herein institute the basis for the classification of literature which subsequently serve as the structure for the research’s theoretical framework and empirical framework.

1.6 Scope of Research

As defined in the research objectives and the scientific paradigm, the boundary of the research is framed within the scope of simultaneous empirical examination on various models for stock market declines to identify the market fundamentals with the highest and most consistent predictive capability. The researchMalaya scope emulates that of Chen (2009) and Chang (2009). Nonetheless, discussions on the effectuations of market fundamentals across models are furnished inof accordance with Chen (2009).

1.7 Definitions of Key Terms

This section is a prelude to the Literature Review in the following chapter. The purpose of this section is to narrow down the scope of topics covered in the ensuing chapter and to provide an overview on how literatures in the past on stock market declines could be synthesisedUniversity and classified. Before the research deliberates on the etymologies and classifications of theories / hypotheses / frameworks / frontiers that shape the outline of the subsequent chapters, this section will begin with the in-depth definitions of the three important terminologies most frequently referred to throughout the research.

1 The scientific paradigms and economics philosophies in this section are heavily referred to Hauseman (2013). 13

1.7.1 Definitions of Bull and Bear Markets

The advancement of study on the bull and bear markets has expanded the frontier for the research on the dynamics of stock market movement and established the basis for further investigations into other areas such as the one in which the research is undertaking. Albeit the growing interest, it is noteworthy that the formal definitions to the terms of ‘bull market’ and ‘bear market’ have yet to be agreed in consensus.

Bull market is deemed synonymous with the “market rally” term that is commonly found in literature. Stock market rally is defined as the ascendance of the indices or the stock prices (Shim & Siegel, 2001). Likewise, the general understanding of the term “bull market” is the occurrence of a persistent upward trend of stock market indices particularly after a period of downward trend or stagnancy. The run of positive returns during the bull market could stretch for monthsMalaya and commonly typified by high volume of transactions in market. On the contrary,of a bear market is defined as the trend of extended decline of stock prices in market. It is commonly typified by sweeping pessimism on the market outlook. Expectation to economic contraction is the most common catalyst to the occurrence of bear markets in the history (Downes & Goodman,

2010).

Tracing the origins, both the terms of bear markets and bull markets are regular jargons used by fund managers in the financial market to describe the directions of the market.University These terms are later accepted as formal terms for studies on the subject area. In reference to Yanis (2002), one of the definitions from the common traders’ viewpoint

(non-academic) of a bear market is a decline in stock market index of at least 20 percent or beyond. Other addendums to the key specification include the precondition that the stock market has to go through three stages i.e. stage one - a “routine decline” of 5

14 percent or more; stage two – a “moderate decline” of 10 percent or more; and stage three - a “severe correction” of a drop of 15 percent or more.2

The technical specifications for the bear markets and the identifications for heterogeneous durations in the stock markets are featured in Chapter 3. These specifications which are illustrated in the form of mathematical notations and algorithms are operationalised into time series models with a combination of statistical programs (namely Excel, Eviews and Matlab).

1.7.2 Definitions of Stock Market Crashes

In layman’s terms, a stock market crash is defined as a steep and abrupt decline in the prices of stock or stock market indices (Garber,Malaya 1992). It is common for studies in this area to use specific episodes of crashes as reference points without defining objectively the specification of a crash. Stockof market crashes of 1929 and 1987 are some of the most widely referred to examples in the literature (see for examples: Klien,

2001; Nicholas 2008; Rappoport & White, 1993; 1994; Serkin, 1975; Schwert, 1990;

Yang & Bessler; 2008).

The definition of a stock market crash, spelled out by Black, Hashimzade &

Myles (2009) in the Oxford Dictionary of Economics is as follows:

University“A sudden and drastic general fall in the security prices on stock exchange… A stock market crash is always possible, since the present price is heavily

dependent on opinions about future changes. A crash is always likely when a

prolonged bull market has pushed the share to high price-earnings ratio”.

2 Re-emphasise, the definition by Yanis (2002) is only one of the many informal examples found in non- academic literature. 15

According to one of the many views of market fund managers, the occasional 0 to 3 percent fluctuation of stock indices is considered high but not alarming. Any decline between 3 to 4 percent is considered as a sign of alarm in the market and a decline that exceeds 5 percent is considered a severe crash (Vines, 2009).

Johansen (2004) noted that prior to the introduction of the JLS model, there was no single definition that is objective or methodical in the literature to specify the endogenous (bubble-induced) stock market crashes. In this context, it is crucial to emphasise on the “endogenous” characteristic as the counterpart i.e. the exogenous crashes, can be caused by an infinite possibility of external shocks that are assumed to be random and unpredictable (Fry, 2012).

The econophysics’ JLS model is a ground-breaking methodology that introduced an objective specification for crashes in the asset marketsMalaya (e.g. see Johansen & Sornette, 2010; Sornette & Cauwels, 2014; and Yan,of Woodard & Sornette 2012). Johansen & Sornette (2008) argued that most of the stock market crashes in history exhibited an idiosyncratic signature. This signature, which precedes each endogenous crash manifests two unmistakable patterns i.e. a super-exponential surge and an accelerating oscillation in the trajectory of prices (termed as the LPPL signature). The axiomatic specification of the JLS model for the identification of crashes is shown in Chapter 3.

1.8University Significance of Research

The significance of research naturally revolves on the research objectives. The focal point of this research is to determine the most consistent individual market fundamentals for the prediction of stock market declines. Considering that most of the advances in the area of stock market declines have been developed independently,

16 literatures from these differing schools of thought will first have to be thoroughly synthesised to establish a symbiotic framework.

Therefore, the first significance of this research is to provide a unified platform for the study in the stock market declines, both theoretically and empirically. Crucially, it is worth noting that the groundwork of the research ventures beyond the principal requirements of the chapter of literature review which aims to justify the research theoretically and to highlight the empirical research gaps of the research topic. The research also contributes to the building of a coalesced theoretical base and in extending the area of knowledge by examining differing schools of thought, filtering incomparable scientific paradigm and reconciling the direction of various empirical approaches

(conventional econometric and interdisciplinary econophysic) for the investigation of a common issue. Malaya The second significance of the researchof is in the exploration of the Shiller’s Financial Variables on the proposed stock market declines models. These variables that are mostly unexamined in previous similar studies (e.g. Chang, 2009; Chauvet & Potter,

2000; and Chen, 2009) are first introduced by Campbell & Shiller3 (1988; 1989). It comprises a list of important financial market fundamentals namely, the aggregated stock market’s (S&P) dividend, earnings, consumer price index (CPI), long interest rate, real price, real earnings and cyclically adjusted price earnings ratio (CAPE). Such variablesUniversity are mostly used for studies in bubble-induced financial crisis (Shiller, 2005). The research also extends the examination of Estrella & Mishkin’s Financial

Variables on stock market declines that constitute the third significance of this research.

3 Robert Shiller, a Nobel Laureate is widely considered as one of the leading authorities in the area of financial crisis. 17

The variable set, first introduced by Estrella & Mishkin (1998) 4 in predicting recessions

(e.g. Qi, 2001) were only used selectively in previous studies which are related to stock market (see citations in the preceding paragraph). The Estrella & Mishkin’s Financial

Variables set provides a wider spectrum of market fundamentals from a macro aspect for the examination of predictability in stock market declines to complement the micro aspect of the Shiller’s Financial Variables.

As aforesaid, there has been no unified investigation on the efficacy of market fundamentals in predicting various specifications of stock market declines, specifically, one that includes the JLS model. This highlights the fourth significance of the research that is in comparing the predictability results of the proposed sets of market fundamentals, tested extensively with four categories of methodologies in specifying stock market declines. These categories of methodologiesMalaya include parametric, semi- parametric, non-parametric and the econophysics. The summary of the categorisation is illustrated in a figure as featured in Chapterof 3.

1.9 Outline of Research

Chapter 1 herein provides the background of the research, problem statements, research questions and research objectives, scientific paradigm, scope of research, definitions of keyUniversity terms, significance of research, concluding remark, all of which collaboratively

4 Note that the ‘stock prices’ category of variables of the Estrella & Mishkin (1998) is dropped from this research. The aforementioned category includes the New York Stock Exchange composite index, Dow Jones 30 industrials index, monthly average at close and the Standard and Poor’s 500 index (S&P 500), monthly average. The justification of the exclusion being that the dependent variable for the in-sample and out-sample analysis for the research is the S&P 500 index, thus it cancels out itself from being a dependent variable. As for New York Stock Exchange composite index and the Dow Jones 30 industrials index, both moves in tandem with the S&P 500 index most of the time, particularly when the stock market retreats. Therefore their inclusion for the predictability test of S&P 500 index during stock market crashes and bear market is unreliable. Such approach was also adopted by Chen (2009). Further introduction to the ‘Estrella & Mishkin financial variables’ is featured in the section of introduction to data in Chapter 3. 18 advance a meticulous case for the proposed research title, backed with in-depth literature-supported justifications.

Chapter 2 discusses various aspects of the subject of stock market declines that spans across numerous schools of thought in the financial economics discipline.

Reviews of literature on the cyclical bear markets, the phenomena of stock market crashes from the perspective of complex systems as espoused by the econophysicists and the use of market fundamentals for predictability tests is also conducted extensively. Research gaps derived from these literatures that form the basis of the research objectives is underscored and deliberated.

Chapter 3 outlines the dependent variable, the test variables (used interchangeably with the term of market fundamentals, especially in the context of empirical analysis), the duration of observations andMalaya the specifications of models that are used to achieve the proposed researchof objectives. Next, the axioms of the conventional econometric models for regime switching i.e. the Markov-switching, the

B-B algorithm, the naïve moving average model (and their variants) and the JLS models are presented and concisely explained. The subsequent section then describes the use of the probit model to test the predictability of stock market declines, modelled based on different specifications with a list of test variables proposed for the research.

The empirical result is presented and discussed briefly in Chapter 4 and finally, ChapterUniversity 5 will recapitulate the key findings of the research, illustrate the connections of these results with the research objectives defined for the thesis and discuss the inferences derived from the study for probable practical applications and policy implications.

1.10 Concluding Remark

The research aspires to extend the literature in the area of stock market declines that is relatively under-investigated, and contribute to the growing interest in cross-disciplinary approaches for studies in financial economics. The Popperian’s epistemology of falsificationism dictates that the nullification of any theory is always possible.

Inherently, the nature of economy and financial market is very complex due to the continuous interaction of agents. Therefore, the falsification philosophy always runs true to form, particularly in the financial economics discipline. The existence of contradicting views and theories is thus expected.

The general aim of the research is to establish a unified research framework for diverse schools of thought in the examination of stock market declines (bear markets & crashes). The research begins with the synthetisationMalaya of historical reviews and traces the evolution of competing theories on stockof market declines. A common platform is subsequently established for the examination and comparison of frontier empirical approaches used to model market declines. The research is motivated by the diversity and often contradicting worldviews on stock market and aims to bridge some research gaps specifically in the area of stock market declines.

Finding a reconciliatory ground with emphasis on pragmatism is the standpoint of the research. In consonant with other prior studies (e.g. Chang, 2009; Chen, 2009; andUniversity Qi, 2001), the research espouses the conviction that regime-switching forecasting cannot be sufficiently accommodated with the highly generalised multivariate or single- index models. Accordingly, the research seeks to conduct a directional prediction on stock market (i.e. declines or otherwise) with the most consistent market fundamentals instead of focusing on the future quantitative values of variables from the models’ predictions.

The performance of test variables is evaluated through both in-sample and out- of-sample tests across the proposed ex-post models for stock market declines. A test variable is considered reliable if it is consistently ranked highly (based on results derived from either the predictive regression model or the probit model, depending on the types of models) among other variables when tested on different specifications of models. Previous studies (e.g. Chen, 2009; Estrella & Mishkin, 1998; Qi, 2001) contended that ad hoc ex-ante out-of-sample test such as one using the probit model is a very flexible and robust approach to test the efficacy of individual test variables in predicting regime-switching.

In summation, the principal undertakings of the research, premeditated correspondingly to achieve the proposed research objectives are as follows: 1) to establish a unified research framework, both theoreticalMalaya and empirical, for the subject of stock market declines; 2) to model (ex-post) the different specifications for stock market declines; 3) to explore new innovations onof selected models; and 4) to examine a list of test variables which reflect the market fundamentals on different models of stock market declines to investigate their predictive power.

University

CHAPTER 2

LITERATURE REVIEW

2.0 Introduction

The review of literature begins with the assessment on the conventional worldviews on stock market declines and stock market prediction. Review in this section includes literature from the orthodox schools of thought that formed the bedrock of the financial economics discipline and literatures that are inconsistent with the orthodox financial theories. Orthodox financial theories generally dismiss the predictability of stock market and regard the occurrences of stock market declinesMalaya as mere outliers or random fluctuations of the market. On the contrary,of literatures that are on the other side of the divide typically espouse paradoxical notions such as stock market predictability or market irrationality that leads to bubbles and crashes.

Literatures are reviewed methodically to cover the key definitions, the etymology for various schools of thought, evolution of theories and the argumentation among competing viewpoints on common issues. Critical comments and inferences are inserted accordingly in-between passages and the summation section builds a solid justificationUniversity in support of the research objectives. The flow of subjects reviewed in the section of conventional worldviews on stock market declines is in the sequence as follows; origin of stock market prediction; random walk theory; efficient market hypothesis (EMH); divergence of theories and practices; stock market bubbles; business cycle and stock market; and market fundamentals and anomalies.

Next, the chapter moves on to the section on the advancement and alternative worldviews of research which includes topic on the departure (of theories and approaches) from conventional worldviews, adaptive market hypothesis, introduction to the complex system, etymology of the complex systems theory, measurement without theory, complexity of economics, evolution of the complex systems theory in economics and complexity of financial markets and stock market crashes. The overall section highlights the emergence of the complex systems theory which departs from the predominant economics rationales and the critical influence of the econophysics approaches on the area of stock market declines, both theoretically and methodologically.

In-depth discussion on the topics reviewed is featured in the classification of literature section that serves as the foundation ofMalaya the theoretical framework of the research. The theoretical framework is the penultimate part of the chapter that is in the wrapped up section of concluding remarks.of

2.1 History and Conventional Worldviews of Research

The overall section on the history and conventional worldviews of research examines the principal theories of stock market in context of its declines, predictability and the fundamental assumptions. It is noteworthy that most of the orthodox schools of thought areUniversity in discordance with the proposition of stock market predictability. Such disagreement may not be argued explicitly in most literatures. Nevertheless the obvious contradiction on the fundamental assumptions of these orthodox theories infers the irreconcilability in no uncertain terms.

2.1.1 Etymology of Stock Market Declines

This research adopts the generalised term of “stock market declines” as the unifying theme for the investigations of stock market crashes and bear markets. Voluminous literature related to the study on stock market declines had designated diverse definitions to the phenomena. As aforementioned, one of the earliest references of

“stock market declines” could be found in the literature by Hamilton in 1922 to describe the downward movements of stock market (Hamilton, 1998). The word “decline” according to the Merriam-Webster’s Collegiate Dictionary (2002) is defined as “to slope downward or descend”. This definition embodies the cardinal characteristics of stock market crashes and the run of bear markets without specifying the nature of such descent i.e. abrupt and dramatic drop or otherwise.

Gilbraith (2009) who authored one of the earliestMalaya literature that used the phrase “stock market crash” (i.e. The Great Crashof 1929), described the phenomena of crash as a massive and abrupt decline in the prices of stocks due to a confluence of various events in reflection of the underlying economic condition. The market panic that sets in is caused by the psychologically insecure traders, manifested in their herding behaviour in liquidating their hold of stocks. One of the earliest literature that cited the terms of bull markets and bear markets is found in Santoni (1987) in his study on the extended run of bull market between the period 1924 to 1929 and 1982 to 1987 of which both eventuallyUniversity resulted in the two most devastating crashes of the century. Arneodo, Muzy & Sornette (1998) used the term “cascade in stock market” in reference to stock market downturns. The meaning of cascade, according to the dictionary (Merriam Webster’s, 2002), is described as “something falling or rushing forth in quantity”. Therefore, the definition is exclusively more suited for abrupt crashes

24 in a stock market and is deemed unsuitable to depict the prolonged and persistent gradual fall of a stock market epitomised by the term of bear markets.

The term “drawdowns” is almost exclusively used in econophysics in reference to crashes of financial assets. One of the earliest literature that used “drawdowns” interchangeably with crashes was by Johansen & Sornette (1998b). Sornette & Johansen

(2001) provided a superficial definition to drawdown i.e. cumulative losses. Johansen

(2004) assigned a more detailed definition of drawdown specifying it as “loss from the last local maximum to the next local minimum disregarding noise fluctuations”.

The subject of stock market crashes and bull / bear markets in general are two provoking topics in financial economics. It is increasingly acknowledged that academic literature on the stock market are deeply ingrained within the Gaussian framework which inappropriately fails to reflect the non-GaussianMalaya characteristics of the real market (Cross, Grinfeld, Lamba & Seaman, 2005)of resulting in the huge fluctuation in the form of rallies5 and crashes being rendered by outliers (Johansen & Sornette, 1998b; 1998c).

Lo and Mackinlay (2001) echoed that it is fundamentally flawed to assume the disturbance term (휀푡 ) under the random walk model, which is fundamental to the formulation of the efficient market hypothesis (EMH) as independent and identically distributed (i.i.d.) as Gaussian random variable. However, there is growing empirical evidence that shows the financial market regularly exhibits fluctuation that does not conformUniversity to the Gaussian distribution. In the past two decades, the phenomenon of stock market crash has persistently challenged one of the most fundamental assumptions of the EMH, the orthodox theory

5 Rally is considered as a direct inference to the bull and bear markets as discussed in Subsection 1.7.1; Definition of Bull and Bear Markets. 25 on capital assets – rationality6 (French, 1988; Lo, 2004). According to the EMH, all investors in the market are assumed to have perfect rationality, which is reflected in their optimal stock trading decisions. Thus, there would be no mispricing in the market, no speculative bubble could exist and therefore, no stock market crash would occur

(Cooper, 2008; Cunningham 2000: Kaizoji & Sornette, 2008).

Arthur (1995) disputed the EMH and suggested that the stock market was more complex than the prevailing hypothesis, which since its mass espousal by financial economists in the 1960s, was established as the fundamental basis on which the technical mechanisms of the financial market was built on. Lo (2005), on the same wavelength, also noted “it is difficult to overturn an orthodoxy that has yielded such insights as portfolio optimization, the capital asset pricing model (CAPM), the arbitrage pricing theory, the Cox-Ingersoll-Ross theory of theMalaya term structure of interest rates, and the Black Scholes/Merton option pricing model, all of which are predicated on the EMH in one way or another.” Jung & Shiller (2006),of meanwhile, showed evidence to support the ‘Samuelson’s dictum’ that the stock market is “micro efficient but macro inefficient”.

The earliest literature on bubble induced stock market crashes were widely credited to the work of Galbarith (1954) and Kindleberger (1978) through reasoning discourse from a historical perspective. The topic was later examined meticulously throughUniversity theoretical modelling, empirical research by the “behavioural finance” economists (pioneering literature i.e. Bondt & Thaler 1984; Camerer, 1987; Khaneman,

Slovic & Tversky, 1982; Khaneman & Tversky, 1979) and neo-classical economists of rational expectations (pioneering literature i.e. Blanchard 1979, Blanchard & Watson

1982; Tirole 1982, 1985; West, 1987).

6 Rationality of investors, neutralisation of irrational investments by rational arbitrageurs and rational reaction to costless information are the fundamental assumptions of EMH (Shleifer, 2000; Yalcin 2010). 26

According to Gilbraith (2009), who authored one of the earliest literature on stock market decline (i.e. The Great Crash 1929), the phenomena of the massive and abrupt decline in the prices of stocks, defined conveniently as “stock market crash” is a confluence of various events in reflection of the underlying economic condition and the psychological insecurity of traders, manifested in their herding behaviour in liquidating their hold of stocks. One of the earliest study that cited the terms of bull markets and bear markets can be found in Santoni’s (1987) - study on the extended run of bull market between the period 1924 to 1929 and 1982 to 1987, both of which eventually resulted in two of the most devastating crashes of the century.

Since the emergence of ‘econophysics’, a term first used by Eugene Stanley in

1995 (Stanley et al., 1999), many novel methodologies in the form of statistical and mathematical physics were introduced into the economicsMalaya discipline. Examples of new concepts which completely depart from the conventional econometrics include the scale invariance, hierarchical systems, 1/f-noiseof and the use of hazard rate in modelling the build-up of bubbles prior to stock market crashes (Johansen, 1997; Stanley, Amaral,

Canning, Gopikrishnan, Lee & Liu, 1999); These methodologies in general were heavily influenced by the fractal concept introduced by Mandelbrot (see for e.g.

Mandelbrot, 1997a; 1997b; 1999b & 2000 etc.). The conundrums of market bubbles and stock market crashes eventually become the research foci for econophysics. UniversityEconophysics prioritise methodological precision above the theory in seeking answers to a problem. Econophysics approaches therefore are generally more complex, considered more efficient (e.g. nonlinear dynamic methodologies which stem from statistical mechanics) and has higher sensitivity and accuracy (e.g. hierarchal diagnostic processes) as compared to conventional methods in economics. Savoiu & Iorga-Siman,

2008 concluded aptly the efficacy econophysics’ methodologies as follow:

“Econophysics was from the beginning the application of the principles of

physics to the study of financial markets, under the hypothesis that economic

world behaves like a collection of electrons or a group of water molecules that

interact with each other, and the econophysicists are always considered that,

with new tools of statistical physics, and the recent breakthroughs in

understanding chaotic systems, they are making a controversial start at tearing

up some perplexing economics and reducing them to a few elegant general

principles with the help of some serious mathematics borrowed from the study

of disordered materials.”

Similarly to the concept of stock market crashes, the notion of “bull and bear markets” which is a real world phenomenon is also considered very underdeveloped over the years. Gonzalez, Powell, Shi & Wilson (2005)Malaya underlined that although the crude term of ‘bull and bear markets’ is commonly used by traders in the stock market as reference to the extended rally in pricesof and the otherwise protracted run of prices decline, research in this area is somewhat circumscribed due to its contradiction with the orthodox financial theories, e.g. random walk theory as encapsulated in the EMH. The unequivocal occurrence of cyclical trends in the stock market only began to gain serious interest since the 2000s, notably by the works of Candelon, Piplak & Straetmans (2008),

Chauvet & Potter (2000), Chen (2009), Chang (2009), Cunado, Gil-Alana & de Gracia

(2010), Edwards, Biscarri & Gracia (2003), Gonzalez, Powell, Shi & Wilson (2005),

LundeUniversity & Timmermann (2004), Maheu & McCurdy (2000), Maheu, McCurdy & Song

(2009), Pagan & Sossounov (2003) etc.

This spark of interest on the topic of bull and bear market albeit long overdue, could be due to 2 factors; the availability of more advanced methodologies to analyse the changes of regimes and proliferation of literature that dispute the foundation of

28 modern financial theories7 from behavioural finance and the econophysics. The disputes are mainly focused on issues of irrationality and bubbles for behavioural finance and stock market crashes for econophysics.

The decisive rejection of the random walk theory by Lo & MacKinlay (1988) with a simple specification test was an important turning point in the literature. The finding has given rise to the question of ‘what could it be if it is not a random walk?’ among academicians, and thus served as a catalyst to the subsequent proliferation of interdisciplinary researches that did not prescribe the conventional dogma of financial economics.

2.1.2 Origin of Stock Market Prediction Malaya The investigation on the predictability of stock market, specifically in the context of stock market declines, has brought the evolutionof of theories on the stock market to a full circle. The most rudimentary analysis on the stock market began with the technical analysis introduced by Charles Dow in 1884, then followed by the conception of the idea that stock market return was inexorably linked to market fundamentals (i.e. the general economic condition and the performance of firms of stocks), hence the introduction of fundamental analysis in 1938. Through the course of time, such analyses hasUniversity advanced considerably. Despite the EMH emergence in the 1970s which nullified the credence of predicting stock return and the subsequent evolution of theories in financial economics behind the scene of the real trading floor, both the technical analysis and the fundamental analysis are still widely used by the investment community up to the

7 The evolution of theories in the financial economics specifically in the stock market is reviewed and discussed in in-depth in the following section. 29 present. The EMH in general stipulates that any attempt to forecast the future trajectory of security price using the information available in the market is futile.

The complex systems theory and the chaos theory introduced by the econophysics school of thought through the application of multifractality and scaling

(including the LPPL formula, e.g. see Bree & Joseph, 2010; Johansen, 2003; Johansen

& Sornette, 1998a; Johansen, Sornette & Ledoit, 1999; Mantegna & Stanley, 1997;

Mantegna, 1999; Vandewalle, Ausloos, Boveroux, & Minguet, 1999; Matsushita, da

Silva, Figueiredo & Gleria, 2006; Sornette & Johansen, 1998; 2001; Sornette, Johansen

& Bouchaud, 1996 etc.) are not dissimilar with the philosophy embraced by technical analysts of which the means of analyses is focused almost entirely on the perfunctory momentum of stock market fluctuation.

On the other hand, studies on the cyclicalMalaya bull and bear markets which are inextricably linked to the movement of businessof cycle (e.g. see Avramov & Chordia , 2006; Bordo, 2003; Casarin & Trecroci, 2006; 2007; Chordia & Shivakumar, 2001;

2002; Cochrane, 1989; 2005; Hovakimian, 2011; Justiniano, Primiceri & Tambalotti,

2010; McCown, 2007; Smith, Sorensen & Wickens, 2006 etc.) and the use of market fundamentals to examine the predictability of stock market declines (e.g. Ang &

Bekaert, 2007; Binsbergen & Koijen, 2010; Boudoukh, J., Michaely, Richardson &

Roberts, 2007; Boudoukh, Richardson & Whitelaw, 2008; Campbell, 2008; Chang, 2009;University Dopke, Hartmann & Pierdzioch, 2008; Hartmann, Kempa & Pierdzioch, 2008; Nyberg, 2011; Pierdzioch, Dopke & Hartmann, 2008; Rapach & Wohar, 2005) both can be considered as the basics of fundamental analysis.

Technical analysis is the oldest and one of the most widely practiced methods by the contemporary investment community. Traders who espouse the technical analysis approaches investigate the momentum of market prices and the trading volume, both of

30 which reflect the real time supply-and-demand equilibrium of the market. Market fundamentals which include indicators that reflect the economic condition are meaningless to a technical analyst (Colby, 2002; Pring, 2002).

The emergence of the Dow theory in 1884 marked the inception of the technical analysis practice. According to Charles Dow, the father of technical analysis, the movement of stock prices is akin to the rush of water. Fundamentally, there are three typical trends of stock prices fluctuation: 1) massive movement that mimic the rush of tide; 2) middle movement that is similar to the charge of waves; 3) short term movement that resembles the ripples (Edwards, Magee & Bassetti, 2007).

Generally, trading rules of technical analysis can be classified into four main domains: 1) the prices and volume analysis, 2) the tracking of momentum, 3) the rules of contrary-opinion and 4) the strategy of following Malayawhere the smart money goes. These trading rules of technical analysis are developedof from its principal philosophy which espouse that the historical movements of prices and transactions of the aggregate market and individual stocks contain useful information that can be utilised to make profitable investment decisions. Therefore, technical analysis is predominantly dependent on data that can be derived solely from the market i.e. various statistical permutations of stock prices and trading patterns such as charts, moving average and trading volume etc. In principle, technical analysis has minimal contemplation on market fundamentals such as leadingUniversity indicators of the economy, industrial performance outlook and financial information of corporations. (Reilly & Brown, 2011).

The more common literature on technical analysis is in the form of self-help reference text tailored for real time traders. Academic literature in this area on the other hand is generally categorised in a broader theme of “market timing.” However, not all studies on market timing are focused explicitly on the efficacy or advances in

31 techniques used on the trading floor. Market timing could also imply the use of various other methodologies, including fundamental analysis to assist in the decision process of buying, holding or selling of market assets. Some of the more common tools of technical analysis include the analyses of cycle and seasonality, bar charts, support and resistance, trends, ‘consolidation, congestion and correction’, breakout, continuation patterns, reversal patterns, moving averages, momentum, and divergence etc. (Kahn,

2006).

The modern studies on stock market have evolved principally from two competing approaches in stock analysis namely the fundamental analysis and the technical analysis, both origins that date back to the 1930s. These approaches along with the analysis of anomalies and attributes trending can be broadly categorised as active investing or portfolio management strategiesMalaya in the context of portfolio theory. Active portfolio management seeks to maximise return by exploiting the inefficiency of the market. Contrary with passive portfolioof management which builds a portfolio on the foundation of buy and hold strategy and indexing, active investors seek to maximise the expected utility of risk-adjusted return from active trading (Roll, 1992; Sharpe, 1964;

Zhao, 2007).

The Theory of Investment Value by John Burrs William is often credited as one of the earliest literature in the area of fundamental analysis (Williams, 1997). Abarbanel & BusheeUniversity (1998) defined the fundamental analysis as: “A practice that relies heavily on the analysis of current and past financial

statement data to identify when underlying firm value differs from prevailing

market prices”.

Fundamental analysis is based on the assumption that every stock is anchored to an intrinsic value which can be derived primarily from the analysis of financial 32 statements of firms, the firms’ specific prospect in the industry and the general economic condition. The question of whether the stock of a firm is under-priced or overpriced cannot be decided objectively unless its intrinsic value is estimated through the fundamental analysis and compared with the price of stock traded in the stock market. (Graham, Dodd & Klarman, 2008; Graham, 2009).

Fundamental analysts in essence seek to identify undervalued stocks in the market based on economic and financial information to outperform the market return.

Market return is commonly benchmarked with stock-indexed fund (i.e. S&P 500 indexed fund) which replicates the broad movement of the market. The core endeavour of fundamental analysis lies in the thorough examination of a firm’s past performance through its financial statements and deduces the probable trajectory of a company’s prospect in the foreseeable future. Financial statementMalaya analysis is primarily focused on accounting items such as revenue, earnings, operational costs, debts, assets, liabilities etc. The terms of comparison for the analysisof of financial statements are commonly denoted in the form of ratios i.e. price-to-earnings ratio, dividend yield, debt-equity ratio etc. (Damodaran, 2012; Ritchie, 1996).

Apart from financial statements, fundamental analysts also investigate various aspects of a company’s management as well as the company’s industry prospect and the general economic outlook 8 . Apart from the financial strength and the management acumenUniversity of a firm, financial analysts also examine other aspects such as the condition of the specific industry, the state of economy of the country and the global economic outlook. Such information are crucial in forecasting a company’s future commercial

8 “There are two general approaches to the valuation process: 1) the top-down, three-step approach or 2) the bottom up, stock valuation, stock picking approach. Advocates of the top-down approach believe that both the economy market and the industry effect have a significant on the total returns for individual stocks. In contrast, those who employ the bottom-up, stock picking approach contend that it is possible to find stocks that are undervalued relative to their market price and these stocks will provide superior return regardless of the market and industry outlook” (Reilly & Brown, 2012). 33 performance and the potential return on investment that its stock could generate

(Damodaran, 2006; Palepu & Healy, 2007).

Industry analysis is a narrower scope of economic analysis which focuses on the market environment. The objective of this valuation process is to compare the prospect of various industries under different economic conditions. Different industries tend to perform better than others and vice versa at varying stages of a business cycle (Koller,

Goedhart & Wessels, 2009; Palepu & Healy 2007).

Studies have shown that some of the factors that significantly explain an individual stock’s return include the changes in the rate of return of the aggregate stock market and the changes in the rate of return of the aggregate industry that the firm operates in (Bae & Duvall, 1996; Livington, 1977). Meyer (1973) suggested that the combination of both the movement of the market andMalaya the industry are good predictors to the trajectory of individual stocks althoughof the significance of the market movement diminishes in a longer term.

While recessions by and large have an adverse impact on the entire market due to the prevalence of systematic risk; the extent of damage is not the same for different industries in general. One of the main emphasis of industry analysis thus is to evaluate prospective industries for stock investment in anticipation of the economic condition.

The performance of firms and the price of stocks are closely tied with the business cycle (Stovall,University 1996).

It is not uncommon for empiricists to criticise the methodologies of fundamental analysis for not being dependable due to the use of non-quantitative valuation approaches to determine the intrinsic value of stocks. Such criticisms include the disputable assessment on the quality of a company’s management and the subjective judgments on competition within the industry. The major opponent to the fundamental 34 analysis nevertheless comes in the form of theoretical argument from the EMH perspective.

2.1.3 Random Walk Theory

Random walk theory (in the context of financial assets) hypothesises that there should be no detectable trend in the movement of prices across a time-series. The changes of the succeeding prices are independent from the past prices (Beinhocker, 2007; Downes

& Goodman, 2010). The “random departure” of succeeding prices from the past prices is due to the constant flow of information into the market in a random manner and the continuous adjustment of stock prices based on the information. Since the content of information is unpredictable, the subsequent adjustmentMalaya of stock prices based on that information is also unpredictable (Malkiel,of 2003). The much revered random walk theory was conceived in 1900 by Louis

Bachelier in his Ph.D thesis entitled The Theory of Speculation. In the words of

Bachelier (2006) in the translated thesis, he introduced the concept of his work as follow;

“The influences which determine the movements of the stock exchange are

innumerable. Events past, present or even anticipated, often showing no Universityapparent connection with its fluctuations, yet have repercussions on its course. Beside fluctuations from, as it were, natural causes, artificial causes are also

involved. The stock exchange acts upon itself and its current movement is a

function not only of earlier fluctuations, but also of the present market position.

The determination of these fluctuations is subject to an infinite number of

factors: it is therefore impossible to expect a mathematically exact forecast…”

According to Leroy (1989), the Brownian motion, which was first theoretically analysed by Albert Einstein in year 1905 and subsequently widely studied across the field of physics, came much later than Bachelier’s random walk. It is also noteworthy that the random walk theory has benefited enormously from the earlier work on the martingale (see Samuelson 1965 and Mandelbrot 1966).

Academic interest on the random fluctuation of economic time series data was revitalised when Kendall & Hill (1953) made an intensive analysis on a list of economic variables and made a five-point conclusion proposing that - the movement of prices in close interval are random and “behave like wandering series”; it is not easy to determine whether the movement of a series is truly random or otherwise statistical; it is almost impossible to model a random series and perform hypothesis testing on the data; aggregated series exhibit higher predictability but theMalaya emergence of systematic element should be treated with caution as it could be spurious; the movement of stock prices shows limited evidence of serial correlationof within series as well as lag correlation between series. Thus the price of a stock is impossible to predict without “extraneous information”.

Based on the parallelism drawn by Kendall & Hill (1953) that the movement of stock price is akin to the outcome of a roulette wheel which “has no memory” (i.e. each current outcome is independent to the earlier ones), Roberts (1959) simulated a series of dataUniversity on ‘market changes’ from random numbers and found striking resemblance in details between the artificial data and the real data of Dow Jones, thus reasserting that the movement of price is by chance.

Samuelson (1965 & 1973b) provided further proof of price randomness in financial assets and refined the ‘information induced random walk theory’ a few decades after Bachelier’s thesis. Samuelson extended the fair game idea proposed by

Bachelier and in the same vein argued that in a market where information is efficient, price would be properly anticipated and therefore the changes of price is unpredictable.

Any return prospect derived from opportune information is instantaneously offset by the feedback of the very action that investors took advantage of on the price.

According to Samuelson (1973a) and Mikeal (2003; 2007), the application of the random walk theory in the stock market has enormous implications. Effectively it denotes that stock prices in the short-run are not possible to predict. The common practices by investment consultants in providing advisory services and the endeavours of investors in forecasting stocks performance through various analyses (i.e. fundamental, technical, economic, time series etc.) are therefore rendered futile. The profit of average investors in the long-run would not consistently surpass the performance of benchmark indexes such as the DowMalaya Jones Index and the S&P Index. The initiative by Samuelson (Beinhocker,of 2007) along with Cootner (1964)9 who edited a compilation of papers on random walk model which included Bachelier’s study in the highly referenced The Random Character of Stock Market Prices had resurrected the interest of random walk theory with great intensity among economists

(notably Fama, Mandelbrot, French, Sharpe, Scholes, Merton etc.) for the next three decades. The subsequent evolution of theories spawned from the notion of random walk in asset prices would then establish as the bedrock of modern financial theory.

UniversityRoberts (1959) stipulated that on grounds of rational assumption of individual traders in the market, all new information that flows into the market should lead to an instant adjustment of prices as proposed by the random walk model. In the event of slow absorption of new information i.e. changes to the market condition not reflected in

9 Cootner’s The Random Character of Stock Market Prices was cited in most of the founding papers of random walk and EMH, among others: Fama (1970), Leroy (1989), Mandelbrot (1966) and Samuelson (1965, 1973a & 1973b). 37 prices would indicate that there are opportunities for traders to take advantage of to make profit as such opportunities are not fully capitalised yet.

2.1.4 Efficient Market Hypothesis

Subsequent to the random walk theory, efficient market hypothesis (EMH) has emerged to become the central scholastic thesis in stock market. EMH is an operationalised extension to the Samuelson’s work (Lo, 2004) which in the most simplistic and well- known phrase proposes that “a market in which prices always fully reflect available information is called efficient” (Fama, 1970). Thus, based on the assumption of

‘frictionless market’ 10 , all potential gains from market information are already counteracted and fully accounted in the price. Malaya The EMH is commonly tested withof the random walk model, fair game model and martingale model to prove the unpredictability of prices. In layman’s terms, EMH proposes that with all the up-to-date news that is openly disseminated in the market, the price of financial assets is at all times correctly valued. Even if the price of an asset may seem over-valued or under-valued occasionally, by the assertion of EMH, this perceived mispricing is merely a delusion (Shiller, 2005).

The characteristics of the weak form efficiency and strong form efficiency were firstUniversity distinguished by Roberts (1967). Fama (1970) in his meta-analysis on the topic of efficient capital markets which provided an extensive theoretical and empirical review on the work in market efficiency of the era and proposed a concise proposition of EMH to the academic world for further research. The concept is subdivided into three forms

10 “The market is assumed to be competitive and perfect; that is, individuals perceive prices as beyond their influence, there are no transaction costs or taxes, assets are perfectly divisible, and the full proceeds from short sales can be invested”, (Meyer, 1999) 38 namely weak form, semi-strong-form and strong form. These three forms are tested based on the extent of respective information set, Ωt. In the weak form of market efficiency, a test should show that future return cannot be predicted from Ωt, which contains the current and the past prices of an asset or any other market variables such as trading volume or ratios of puts to calls. In a semi-strong form of market efficiency, Ωt is expanded to include all publicly available information (i.e. various market-based data as explained in the weak form efficiency, information on economic condition, financial reports and updates of companies etc.)11, and a test should show that current prices adjust instantaneously to fully reflect this information. Finally, in the strong form of market efficiency, a test should show that no excess return can be garnered from Ωt which includes all public and non-public information (see for detail interpretation, Elton, Gruber, Brown & Goetzmann, 2007; Fama,Malaya 1965; Jones & Netter, 2008; Timmermann & Granger, 2004).

Jansen (1978) concisely elucidatedof the EMH in the context of a martingale model. In the condition of a strong form of market efficiency, trading on information set

Ωt would not yield economic profits (which is defined as returns adjusted for risks and minus all costs). Thus the derivation of asset prices (adjusted for required return) in a speculative market operating under a non-profit condition and constrained by the assumption of zero cost (i.e. transaction and storage costs) will produce the martingale model with regards to the information Ωt. This indicates the existence of perfect competitionUniversity in the microeconomic sense where the availability of new information induces the changes to the prices of stocks. Owing to the swift incorporation of the new

11 “The distinction between the weak and semi-strong forms is that it is virtually costless to observe public market data, whereas a high level of fundamental analysis is required if prices are to fully reflect all publicly available information, such as public accounting data, public information regarding competition, and industry-specific knowledge.” (Jones & Netter 2008) 39 information, the expected changes to the future price fluctuate randomly, approximately to the intrinsic value of stocks (Pass, Lowes & Davies, 2005).

Fama (1991) in his subsequent follow-up review on the EMH supplemented the definition of the type 1 efficiency. The categorization of weak form tests was amended to a broader definition of return predictableness. Corresponding to the categorisation, the semi-strong form efficiencies were changed to studies on specific occurrences and news announcements. According to Jones & Netter (2008) assertion, EMH is an extension to the various well established economic pricing theories that accentuate on

“the zero profit competitive equilibrium condition”. Such postulation can be construed as an amalgamation of the “classical price theory” under the stable market condition with the “dynamic behaviour of prices in speculative markets” under the unstable market condition. Malaya

Lim & Brookes (2011) in one of theof more updated surveys on EMH found that a majority of empirical researches on the EMH over the past decades were solely focused on determining whether or not the stock market of the specific study of complete weak form is efficient. Such studies also tend to assume that the degree of market efficiency stays constant for the entire period of estimation. The survey showed evidence that the degree of market efficiency could be time-varying and the predictability of stock return has been evolving over time. Lo (2004a) also concluded that the EMH of financial marketsUniversity is still inconclusive albeit voluminous research published in this area over the past decades.

2.1.5 Divergence of Theories and Practices

According to Leigh, Paz & Purvis (2002), as the debate on the soundness of the EMH continues, many researches have provided evidence that showed traders employing ranges of technical approaches based on historical data in predicting the prices of stocks have yield positive results to a certain measure (e.g. Chan, Ikenberry & Lee, 2007;

Glassman & Riddick, 2006; Jiang, Yao & Yu, 2007; Park & Irwin, 2007). Examples of trading strategies include the metric-based trading rules such as moving averages and momentum measures (e.g. Gencay, 1999; Hong, Lim, & Stein, 2000; Hong & Stein,

1999) and other assorted technical trading rules (e.g. Chavarnakul & Enke, 2009; Leigh,

Modani & Hightower, 2004; Leigh, Modani, Purvis & Roberts, 2002; Leigh & Purvis,

2008; Pesaran & Timmermann, 2002; Thomakos, Wang & Wu, 2007).

The EMH proponents also argue that instantaneousMalaya stock prices in the market have already incorporated all informationof pertaining to the economic condition, the industry environment and the firms’ financial information portray precisely the value of stocks as they should have been. In short, there would be no previously undiscovered information of a firm by researching its fundamental as all information is already known and reacted upon by all stock traders in the market. Going by the EMH supposition, fundamental analysis is thus deemed to be an extraneous and insubstantial practice

(Fama, 1965; Jones & Netter, 2008).

UniversityIn the context of the EMH, the core assessment to the efficacy of the fundamental analysis is to determine if the employment of such valuation on stocks could result in significantly higher return, benchmarked against the market’s return.

Many studies in accounting and finance have shown that abnormal returns is possible as in most times, prices of stock do not instantaneously reflect the market information accessible to the public, in particular the newly released information on the firms’

41 earnings. Therefore the market mispricing can be taken advantage of through means of fundamental analysis. (Bernard & Thomas, 1989, 1990; Sloan, 1996).

Abarbanel & Bushee (1998) in their examination on a set of indicators that mirror the commonly used contemporaneous financial information for fundamental analysis (i.e. information related to the inventories, working capital, sales, expenditures, margin etc.) concluded that their simulated portfolios which exploited such financial information could beat the market average return by 13.2 percent. As such, their evidence showed that “fundamental signals” are useful in predicting future returns and a large fraction of the abnormal returns of the portfolios was produced only after the firms of stock published their earnings eventually through financial reports.

Richarson, Tuna, & Wysocki (2010) too asserted that the valuation of stock through fundamental analysis is very much relevantMalaya by the fact that it is still widely accepted and practiced by fund managers ofof stock market; and empirical researches of accounting information in predicting the prospect of stocks continues to show positive evidence in supporting the validity of this school of thought.

As highlighted earlier, apart from the microanalysis, fundamental analysis also encompasses the macroanalysis aspect which focuses on eliciting useful information from various macroeconomic variables or indicators. The efficacy of information of the general market in fact could even be superior in forecasting the performance of stock as notedUniversity by Avramov & Chordia (2006):

“Indeed, a formal framework that uses macroeconomic variables generates

trading strategies with long-only positions in individual stocks that outperform

strategies that take long and short positions in the size, book-to-market, and

momentum benchmarks as well as strategies that hold stocks with the same

(potentially time varying) size, book-to-market, and momentum characteristics.” 42

The utilization of macroeconomic variables or indicators as part of the fundamental analysis is a testament to the general acknowledgement of the economic impact towards the price of stock and thus provides a crucial foundation to the subsequent subtopic which focuses on the link between the business cycle and the stock market.

2.1.6 Stock Market Bubbles

The difference in methodological frameworks between behavioural finance and econophysics on the subject of stock market crashes is enormous although both of these disciplines are brought together under the common notion of “bubble”. Behavioural finance primarily concerns the psychological aspectsMalaya of investors such as biasness and heuristic decision-making (Chang, Jiang & Kim, 2009; Hirshleifer, 2001; Kahneman, Knetsch & Thaler, 1990; 1991; Ortoleva of2010; Tversky & Kahneman, 1991; Shefrin, 2005), irrationality (Odean, 1998; Barberis, Shleifer & Vishny, 1998; Barber & Odean,

2001) and herding (Banerjee, 1992; Lux, 1995; Shiller, 1995; Welsh, 2000; Wermers,

1999).

Apart from the behavioural finance and the econophysics, the neo-classical economists offer another viewpoint in the “rational bubbles” which attempts reconciliation with the EMH. These economists who built from the economics’ rational expectationsUniversity foundation were among the pioneers, as aforementioned in acknowledging the existence of bubble in the financial market along with the behavioural financial economists. Blanchard (1979) in one of the earlier literature on rational expectations in the context of asset price movement recognised the importance of not rejecting the existence of speculative bubble in stock market crashes.

Blanchard & Watson (1982) contributed immensely to the foundation of the bubble theory by elucidating the crucial assumptions required to derive a bubble model in a stochastic market condition, i.e. the asset price is inflating exponentially and rational agents are well aware of the imminent crash but completely have no knowledge to the time of market reversal. They stipulated that any deviation of price from the expected fundamental value does not attest to irrational behaviour of investors because

“there can be rational deviations of the price from this value” in the name of rational bubbles.

Camerer (1989) noted that the development of stock market’s rational bubbles was very much influenced by the conceptualisation of “growing bubble” which postulates that the dynamic structures of macroeconomic model (i.e. the monetary model and hyperinflation model) permits price levelMalaya to have indefinite exponential growth within the rational expectations framework. Works by Black (1974), Brock

(1974; 1975), Sargent & Wallace (1973a;of 1973b) Shiller (1979) and Taylor (1977), were some of the immediate literature that served as a catalyst to the development of rational bubbles.

The fundamental assumptions of rational bubbles effectively reject the proposition of irrational behaviours of investors in the behavioural finance and the notion of self-organisation under the complex systems theory put forth by econophysicists.University It can be viewed as an extension theory to explain the unjustified fraction of the EMH – the bubbles formation and the imminent crashes in stock market.

Rational bubbles which began with the measuring of deviation in stock price against its expected value or intrinsic value (future return of the stock i.e. expected return and expected dividend) (Blanchard & Watson, 1982; Diba & Grossman, 1988b) provide a

44 crucial platform for empirical researches with the utilization of tangible variables in investigating the predictability of movement of stock.

Whilst rational bubbles advocates according to Dezhbakhsh & Demirguc-Kunt

(1990) proposed that:

“Fluctuations in the equilibrium price of an asset are believed to reflect changes

in corresponding market fundamentals. Accordingly, stock market booms,

depressions, and crashes can be explained by stock price fundamentals defined,

e.g., as the expected present value of the future stream of dividends. However, it

is also noted that a persistent deviation of stock prices from the path determined

by these fundamentals is possible even when market participants have rational

behaviour and rational expectations. Such deviations, if induced by self- fulfilling expectations, are called speculative Malayabubbles” According to De Grauwe & Grimaldiof (2004), it is of foremost importance to first define clearly the specifications of a bubble before the examination on economic bubble could begin. The accurate meaning to a bubble was conceived as owing to the emergence of the rational expectations theory and its modelling. The efficacy and flexibility of rational expectations model has enabled the solving of unlimited problems related to asset price. Apart from deriving solutions to various fundamental economic issues, rational expectations model has also found a solution to the bubble problem.

UniversityBubble represented with the rational expectations model depicts an exponential path of asset price that diverge incrementally from its intrinsic value but preserve the

“no-arbitrage” condition. Stress-testing the bubble theory would unveil a glaring shortcoming that either assumes that the building up of bubble goes on perpetually or that there exists an expectation of an imminent downturn in the future which would have impeded any intention of investment on the asset in the first place. Owing to such

45 rationale, the EMH school of thought thus argued that asset bubbles should have never existed.

The integration of the bubble concept in rational expectations models is seen as a crucial expansion to the literature in this area. Nevertheless, it is noteworthy that the notion of rational bubble is not flawless. As aforementioned, the most disputable part is in the reconciliation of the notion of an expected future crash into the theory. Thus modelling the bubble-induced crash within the rational expectations framework is contradictory and difficult to be operationalised. The crash component therefore is appended in an ad hoc manner into the model and is not considered as part of the model’s structure. The model assumes that there is exogenous reason that halts bubbles from inflating perpetually (De Grauwe & Grimaldi, 2004).

Heterogeneity of traders was later developedMalaya within the theory in which traders are classified into two categories. The firstof category consists of rational traders whose trades are guided by market fundamentals. The second category consists of noise traders who are characterised as being uninformed and their trading pattern mimics the herding behaviour. Models that accommodate the heterogeneity of traders include specifications that restrict rational traders from capitalizing the build-up of bubble to maximise profit.

The no-arbitrage restriction is based on the assumptions of capital constraints or aversion to risk. The noise traders extension to the rational expectations theory has also subsequentlyUniversity helped conceive the speculative bubble theory (De Long, Shleifer, Summer & Waldmann, 1990; 1991).

Abreu and Brunnermeier (2003) further enhanced the models with constraints that operationalised “arbitrage failure”, specified as the unsuccessful attempt by rational traders to exploit profits from the bubble build-up due to the divergence in views in identifying the time of crash and the failure to coordinate their collective asset disposals.

De Grauwe & Grimaldi (2004) redefined bubble as a “fixed point equilibrium”. The concept is an innovation to the model that applied the bounded rationality assumption instead of the rigid rationality expectations assumption.

“Bounded rationality” finds its root in the behavioural finance discipline (see

Barberis, Shleifer, & Vishny, 1998; Barberis & Thaler, 2007; Simon, 1972; Tversky &

Kahneman, 1981). The theory assumes that rational agents in the market have constrained capacity in making the best decision with the available information due to their cognitive limitation. Thus, these agents would resort to easy, predictable rules but simultaneously comparing and continuously switching for the profitable rule (De

Grauwe & Grimaldi, 2004; Johansen & Sornette, 1999c).

The theory of rational bubble holds to the assumption that each financial asset has its idiosyncratic intrinsic value (fundamental value).Malaya In rational bubble models, the intrinsic value, which is not observable, of is annotated as 푉푡and an observable trading price (transaction price) is represented as 푃푡. The speculative bubble which is built on the foundation of rational bubble as mentioned prior stipulates that the stock market bubble could occur when there is an anomalous deviation of the trading price from the intrinsic value (퐵푡 = 푃푡 − 푉푡 ≫ 0). The subsequent decline or crash in the market resulting from the speculative bubble is construed as an abrupt “evolution” of 퐵푡 =

푃푡−푉푡, reversing from an inflated positive value, to a “0” and possibly all the way to a negativeUniversity value. The most difficult part in speculative bubble is to determine the asset’s intrinsic value (Pele & Mazurencu-Marinescu, 2012).

2.1.7 Business Cycle and Stock Market

Based on Burns and Mitchell (1946) historically significant research, the definition business cycle was coined as follows:

“A cycle consists of expansions occurring at about the same time in many

economic activities, followed by similarly general recessions, contractions, and

revivals which merge into the expansion phase of the next cycle; this sequence

of changes is recurrent but not periodic; in duration business cycles vary from

more than one year to ten or twelve years; they are not divisible into shorter

cycles of similar character with amplitudes approximating their own.”

According to Taylor (1998) the business cycle or sometimes referred to as the growth cycle of the economy is commonly categorised into fourMalaya main stages, namely: “1) economics expansion from below trend growth to a “normal” rate of growth;

2) economic growth above a sustainableof noninflationary level; 3) a cyclical peak

in growth, followed by a decline in the rate of growth toward the trend rate and

4) growth falling below trend toward a cyclical trough.”

Financial markets are highly susceptible to the fluctuation of the business cycle.

Every phase of the cycle has its own repercussion and impact towards the returns of diverse class of asset in the market respectively. Stocks regularly perform better when theUniversity business cycle is at the expansion stage where there is still room for prospective economic growth. Interestingly, the steepest aggregate return of the stock market occurs months prior to the recovery of the economy from a recession (Taylor, 1998).

The inextricable relation between the financial markets and the real economy has been well documented by Cochrane (2005). The robustness of the current economy and its prospective outcome in the foreseeable future are the main factors being

48 considered in most investment decision and analysis. Therefore, the most pivotal analysis to time the market is the evaluation of the economic condition, the state of which the economy is at within the business cycles and the probable direction in which the economy is moving towards in the future. In short, development of the economic condition through time is conditioned by the stage of fluctuation of the business cycle which in turn affects the outcome of the stock market returns.

The dynamic relationship between stock market fluctuations and the business cycle has been explored in many researches and according to Chauvet (1999), stock market index is useful in predicting the state of the business cycle in real time especially the onset of recessions. Casarin and Trecroci (2007) in their findings concluded that macroeconomic and financial indicators follow the common volatility patterns. Apart from viewing the stock market cycles vis-a-vis theMalaya business cycle, some of the other shorter-term cycles tested to have statistical significance in relation with the stock markets are the seasonality effect and variousof calendar effects.

Gorton (1998) argued that financial panics over the course of history were not random or merely “sunspot” occurrences but were caused by contractions in the business cycle. Siegel (2008) showed evidence that since the early ninetieth century, 91 percent of recessions (42 over 46 recessions) had caused at least 8 percent of drop

(either prior or post) in the stock market index. McCown (2007) concurred that in the recentUniversity history, prices of stocks has been consistently retreated prior to or at the stage where the economy was at a decline. The decline of stock prices however was usually gradual and could take up to 18 months to hit the lowest level before any rebound. For various reasons, stock traders did not seem to offload their stock investments abruptly even when an apparent recession was looming.

2.1.8 Market Fundamentals and Anomalies

The notion that stock market movement can be forecasted with information that is accessible by the public is a contravention to the dominant EMH. The endeavour for alternative theories persists nevertheless. In close scrutiny, the undertakings of using publicly available information on the economic condition for stock market forecasting is closely related to the researches on the nexus between the business cycle and the stock market performance.

Evidence from voluminous literature suggests that predicting the future price might not be beyond empirical means with the use of appropriate variables that reflect the fundamental condition of the market such as “dividend yield, price-earnings, price- earnings ratio price-book value, earnings announcements, company size, share repurchases, initial public offerings, etc.” (Thaler,Malaya 1999). Nonetheless, the orthodox theory of financial economics regards theseof evidences of market predictability as merely anomalies within the EMH framework (Yalcin, 2010).

Some classical literatures are generally recognised to yield enormous influence on the perspective of investors specifically within the fundamental-based investment fraternity. These selection of important literatures to name a few include Fama &

French (1988), Goetzmann & Jorion (1993) and Rozeff (1984) who found significant evidence of dividend yields (D/P) in predicting the stock market (i.e. dividend yield is positivelyUniversity correlated with stock price). Basu (1977), one of the more widely cited classic literature in this area has shown evidence that price-per-earnings (P/E) ratio is a good fundamental tool for stock selection. Campbell and Shiller (1988) in the same direction found earnings-price (E/P) to be a very potent ratio in predicting stock returns for an average time frame of one decade.

Pontiff & Schall (1998) similarly found evidence that support the use of fundamental ratios to forecast market return in the book-to-market ratio. Lakonishok,

Shleifer, Vishny (1994) concluded that the stocks with low price-to-book ratios (P/B) perform better in the market compared to otherwise. Keim and Stambaugh (1986) had shown evidence suggesting that a range of market fundamentals can be used to forecast the direction of stock market. Lewellen (2004) substantiated the validity of various financial ratios which include most of the aforementioned fundamentals such as the D/P ratio, E/P ratio and P/B ratio. Hong, Torous & Valkanov (2007) conducted a more in- depth study on the industry and found that the returns of industry portfolio are good indicators to predict stock returns.

Schwert (2003) in his seminal work on market anomalies had among others discussed extensively the efficacy of financial Malaya ratios (as per elaborated prior), macroeconomic variables (such expected inflations and interest rates) and industry level variable (i.e. firm size and liquidity) in forecastingof the movement of stock market. A comprehensive review on market anomalies and market fundamental in stock market forecasting could also be found in Sewell’s (2010) systematic review.

Timmermann & Granger (2004) however provided a contrary view on the use of market fundamentals in predicting the stock market. They argued that as soon as an established market anomaly is made known to the public, such advantage would be exploitedUniversity to the fullest and the said anomaly would not appear anymore in the future observation. There is also the possibility of spurious anomaly resulted from undue “data mining”. This renders the nexus between the spurious anomaly and the stock market return as a pure coincidence. Therefore such anomaly is not expected to be significant in the future too.

Cai, Chou & Li (2009) in the same vein also noted that albeit the existence of voluminous quantitative evidence on the efficacy of economic fundamentals in predicting the movement of financial markets, the significance of such anomalies have yet to receive much acknowledgement in economics. On the other hand, studies on influence of financial ratios to the movement of stock markets are very common in the areas of accounting and finance and evidence on the efficacy of economic fundamentals in predicting market returns are well accepted in these disciplines.

2.2 Advancement and Alternative Worldviews of Research

The section on the advancement and alternative worldviews highlights the unconventional perspectives on stock market. It beginsMalaya with the review on various limitations of the bedrock theories in the financial economics, the introduction and the etymology of the complex systems theory,of followed by the emergence of one of the most important methodological frontiers in the area of stock market crash – the

Johansen-Ledoit-Sornette (JLS) model. The section also discuss the complex nature of the financial economics and how the complex systems theory which focused more on the behaviour of the whole system within itself suits more for researches in this area instead of the more common theoretical framework which is founded on the assumption of rationality of agents in the market.

University

2.2.1 Departure from the Conventional Worldviews

Shiller (2005) noted that over the decades, the orthodox financial theory that is developed based on the conjecture of complete rationality of investors in the market has resulted in the detrimental disposition by academicians and fund managers to structure

52 unknown future outcomes of the market into forms of presumably predictable expectations with the utilization of mathematical modelling and various analytical tools.

Critics of such development highlighted that the generally accepted assumption of the market being a function of an “efficient processor of financial information” has a perilous influence on the structure of wealth management globally.

By adopting the meticulousness of the scientific principle, financial solutions that are derived from the EMH foundation tend to be too generalised in the expression of elegant mathematical equations. There are serious implications resulting from the dependence on oversimplified financial models when resolving economic issues that largely involve social behaviour, moreover for policymaking. There are vast differences in using models with “scientific precision” to solve scientific problems as compared to problems in financial economics because over specifiedMalaya models which are aimed for precise financial forecasting have high tendency of being too narrow and could deviate enormously from reality should rare eventsof occur, according to Shiller (2005).

Likewise, Johansen & Sornette (1998b; 2002) argued that the shortcomings of

Gaussian based financial theories are too dangerous to be ignored especially when dealing with stock market crashes:

“In a million years of GARCH-trading, with a reset every century, never did 3

crashes occur.” University“Under the hypothesis that daily losses are uncorrelated from one day to the next, for the Oct 1987 crash, the sequence of four drops in making the largest

drawdown occurs with a probability 10-23.”

Mandelbrot (2004) on the same wavelength, echoed that based on standard theories, the probability of a stock market crash is treated as outliers with examples such as: 53

“July 2002: The 3 steep falls within a week (probability one in four trillion)”

“19 October 1987: Dow Jones Index fell 29.2% (probability less than < one in

1050).”

Bakshi & Madan (1998) proposed that given the unequivocal stock market crashes and rallies that are not “statistical artefacts”, the topic of stock market crash & bull market ought to be prevalent features in any credible research on the stock market.

Gonzalez, Powell, Shi & Wilson (2005) in similar wavelength accorded that the notion of bull and bear markets is circumscribed due to its contradiction with the bedrock theories in financial economics. They argued:

“Bull and bear markets are distinct financial economic phenomena, instead believing that they are the result of ex-post categorizationMalaya of random data.”

2.2.2 Adaptive Market Hypothesis of

One of the newer frameworks that is fast gaining acceptance as a viable opponent to the

EMH is the adaptive market hypothesis (AMH); a term coined by Lo (2004a). The pioneer framework along with Mackinlay are credited for their earlier work that showed evidence that movements in financial markets are not absolutely random as there are elements that can be forecasted in the returns of stock and bond, a finding that ultimatelyUniversity nullified one of the major cornerstones of theory in financial economics (Lo & Mackinlay 2001).

The AMH proposes the use of evolutionary concept in explaining the phenomena in financial economics. Albeit being qualitative in nature, the AMH has managed to provide plausible explanations to bridge some renowned gaps in EMH through the cross-disciplinary amalgamation of theories such as “bounded rationality,

54 complex systems, evolutionary biology, evolutionary psychology and behavioural ecology”. The conjectures of AMH are based on the theory of evolution, of which the

AMH proposed that the level of efficiency in the market is associated to the

“environmental factors characterizing market ecology such as the number of competitors in the market, the magnitude of profit opportunities available, and the adaptability of the market participants” (Lo, 2004a; 2005).

The implications of the AMH is profound to the research as the theory recognises amongst others, the movement of asset prices can be influenced by heterogeneous trading strategies (i.e. fundamental analysis and technical analysis) where different methods yield different degree of performance at different times. The

AMH also proposed that the behaviour of the market is driven by the opportunities for arbitrage. These assumptions accommodate the standpointMalaya of the research that the movement of stock prices is not merely random, therefore the modelling of stock market declines is possible; and the association ofof the market movement to trading strategies, particularly the fundamental analysis implies that market fundamentals could be effective in predicting the direction of the market.

2.2.3 Introduction to the Complex Systems Theory

Economics discipline in the recent decades has experienced its most significant evolutionUniversity for over a century. The evolution signifies an extensive transformation in the epistemological and empirical direction in the field (Beinhocke, 2006). The paradigm shift is owed mainly to the amalgamation of theories and methodologies that originated from natural sciences. One of the more prominent crossovers is the complex systems theory that is widely applied in financial economics of late.

The meanings of ‘complex’ according to Merriam Webster’s (2002) in verbatim are as follows: “1) a whole made up of interrelated parts; 2) a group of obviously related units of which the degree and nature of the relationship is imperfectly known; and 3) hard to separate, analyse, or solve”. A ‘system’ according to the dictionary is defined as

“1) a regularly interacting or interdependent group of items forming a unified whole; 2) a group of interacting bodies under the influence of related force; and 3) a group of objects or an organisation forming a network.”

Combining the two root words, complex systems in the layman’s understanding can be interpreted as a network of interacting simple units with indistinctive relationships among themselves and is obscure if analysed in separation, but on a whole, the aggregate behaviour of the simple units emerges in an orderly pattern and takes form of a system. Each of the linked-up units in the systemMalaya has its own capacity of self- organising that inaugurates new aggregated behaviour of the whole system. The system thus evolves continuously to adapt to newof environments and has the attribute of involuntary feedback loops.

The theory of complex systems in relative term is defined as an area of study that is set right at the boundary of chaotic order and deterministic order (Kauffman,

1993; Avery, 2003). In contemporary terms, the theory complex systems is a system that comprises many individual components that act according to embedded reaction functions,University which are usually assumed to be the same for every individual. The components are described as cellular automata. The salient characteristic of such systems is that their dynamic properties cannot be derived analytically from the knowledge of reaction functions of the components. The only recourse in studying their outcomes is to resort to dynamic simulations, usually conducted with digital computers.

Problems that are concerned with the self-organizing individual units that give rise to a system are not easily discernible and are hard to solve. Such system is usually in a dynamical order or chaotic in nature and the structure of the system could be hierarchical due to the emerging behaviour from the micro level of individual units. The intricacy of the emerging pattern often prevents prediction or forecasting of the systems directly from the simple units’ specification (Wolfram, 2002).

Over the years, the features inferred to be exhibited in complex systems have been expanding owing to its wide application across various fields of study. Taken in different contexts i.e. biology, physics, computer science, sociology and economics, the combination of features used to describe complex systems are wide-ranging, suited to the fields accordingly. The most common descriptions nevertheless are: 1) the system is dynamic, non-equilibrium and usually non-linear; 2)Malaya the system is self-organising and the relationships among the simple units in the system are woven in a network-like connection; 3) the composite behaviour of ofthe system from an elevated view reflects the merging behaviour of simple units from the micro level. Due to the inherent complexity of the concept, methodologies used to experiment, model or simulation, the system is also diverse across fields (Meyer, 2011).

Since the late 1990s, the interest of physicists and mathematicians in researching economics phenomena has been on the rise. The proliferation of cross-disciplinary researchesUniversity with the application of solutions originating from the physics principle has entailed a gradual paradigm shift in the theoretical articulations and methodological approaches in the economics discipline, primarily in financial economics.

Onnela (2006) noted that the subject of complex systems theory has been acknowledged of late as a novel discipline that not only bridged the gaps between various well established areas i.e. chemistry, physics and biology, the theory has also

57 evolved beyond the boundaries of natural science into areas such as economics, psychology and social science.

As a recap, complex systems are integration of numerous interacting components that form a complete system in which each of the linked-up components in the system has its own respective capacity for self-organisation to inaugurate new aggregated behaviour of the whole system. The system thus is continuously evolving to adapt to new environments and has the attribute of involuntary feedback loops. In other words, complex systems are beyond a mere combination of components that manifests a systematic behaviour or operation but a system of collective dynamic components which could trigger each other to react.

Some of the main attributes of complex systems include the enormous degree of receptiveness to preliminary conditions and the Malaya emergent patterns that are non- deterministic and difficult to predict. Therefore,of reductionism philosophy commonly applied in empirical studies is inadequate to examine the phenomenon. Knowledge on the behaviour of the discrete components is not enough to make an inference or to be used for predicting the collective behaviour of the complete system. Such unique and confounding theory has spawned many novel ideas and advanced mathematical and modelling approaches in the areas of financial economics and social sciences. Among the latest approaches and methodologies developed in this area are mathematical physics,University nonlinear statistical modelling, scaling (note the frontier of log periodic power law combines all the aforementioned techniques), ergodic theory to cellular automata, fractals, game theory and network theory (Meyer, 2011).

2.2.4 Etymology of the Complex Systems Theory

The concept of complex systems can be traced back to the work of Henri Poincare, a mathematical philosopher in the late nineteenth century who introduced nonlinear mathematical solutions after he discovered the fundamental limits of the conventional equations. Through nonlinearity, he illustrated how minor effects could cause far- reaching impact to end result, an idea which years later developed into the chaos theory.

He famously worked on the “three body problem”, a simple system that gives rise to unpredictable trajectories that defy analytic description owing to the phenomena of three mutually attracting bodies being in motion together. He discovered that there can be orbits that are bounded and non-periodic and that do not devolve to a stable cycle.

This upset some of the presumptions of Newtonian mechanics. In that, he showed that it was mathematically impossible to derive equations Malayato predict the trajectories for even a simple system that contains only three planetsof interacting in nonlinearity (Taleb, 2007). The epistemology of deterministic order came under close scrutiny in the 1960s following the translation of Karl Popper’s ogik der Forschung (The Logic of Scientific

Discovery) on the falsification of theories into English in 1959. The original work was published in the German language in 1934 (Popper, 1959). The “Popperian falsificationism” had spurted many ground-breaking explorations by philosophers, scientists and academicians in search for alternative hypotheses and theories to explain theUniversity complex phenomena. Throughout the 1960s, literatures by Edward Lorenz (i.e. Deterministic non- periodic flow, The nature and theory of the general circulation of atmosphere, and

Three approaches to atmospheric predictability), the father of chaos theory who coined the term “the butterfly effect” were widely acknowledged to be the impetus to the conceptualization of the chaos theory. On the other hand, a series of papers by Friedrich

A. von Hayek, a Nobel laureate (i.e. Rules, perception and intelligibility, Kinds of rationalism, and The theory of complex phenomena) laid the foundation for the complex systems theory in the same period (Wible, 2000; Zeidan, 2007). Some literature nevertheless proposed that the seed of chaos theory was sown earlier in 1959 by Andrey

Kolmogorov, who found the solution to the long standing conundrum of entropy of dynamical system in the ergodic theory by applying the thermodynamics approach

(Sinai, 2010).

The two themes of complex systems and chaos systems used in the modelling of social and ecological phenomena are axiomatically distinct, but they tend to coalesce.

Notably, there was no clear distinction in explaining the phenomena of complex systems and the chaos system in the early literature. Work that gave specifications to the two systems might not have existed when the paradigmMalaya shift of non-deterministic order and nonlinear dynamics was at the infancy stage in the early days. As a comparison, works by Lorenz and Kolmogorov were predominantlyof mathematical whereas works by

Hayek were inclined to the philosophy and epistemology on complexity. Their works have gone into the book of history as some of the most important contributions to a new branch of knowledge that was almost unfathomable prior to their era. They had brought about a whole new dimension in researching and resolving problems that are seemingly chaotic and highly complex as the theories’ namesake. UniversityThe subsequent decades of 1970s and 1980s saw a significant rise in the acceptance of these theories, especially among mathematicians to such a degree that academic courses in the areas of interest were made available in universities (Hogkin,

2005). In the early 1980s, Stephen Wolfram began to embark on a research with computer simulation to observe the evolving behaviour of computer systems that is based on simple binary sequences of zeroes and ones, known as cellular automata. He

60 discovered that a simple simulation rule with simple binary values within a system could generate a seemingly chaotic order after many successions of iteration. However, through the extremely complicated outline, regularities of fractal pattern would emerge eventually, forming a similar looking large fractal when combined as a whole. Wolfram named his discovery as the complex systems theory (Wolfram, 1984; 1988). His progressions on the origins of complexity thereafter helped establish a solid foundation for the theory to gain recognition and flourish into different fields (Stephen Wolfram, n.d.).

Over the following two decades, more meticulous descriptions of the characteristics of the chaos theory and the complex systems theory had emerged.

However the boundary between the two theories was still very vague. The distinction between the two systems was rather ambiguous asMalaya their specifications often overlap although the chaos theory at that juncture was more clearly defined compared to the complex systems. As noted prior, Stuartof Kauffman, one of the most prominent biologists in modern era famously termed complex systems as an order “at the edge of chaos”, drawing a clear distinction between the two theories (Kauffman, 1993).

Wolfram (2002) in the synthetisation of his life’s works entitled “A New Kind of Science” treated the chaos theory and the complexity theory as two different entities and used the term complex systems theory and complexity theory interchangeably. OtherUniversity more recent literatures on the other hand considered chaos theory as a subset to complex systems theory (Wible, 2000; Kaneko & Tsuda, 2001; Yu, 2006, Cencini,

Cecconi & Vulpiani, 2010).

According to Kaneko and Tsuda (2001), the cause of ambiguity in the definitions of complex systems is attributed to its wide appeal across many fields of studies because the characterisations for the theory are conceived rather arbitrarily in

61 each respective field based on their needs and interests. It was noted that some researchers even perceived the assigning of a set of unequivocal specifications for the complex systems would inhibit the development of their studies. On the contrary, other researchers advocated the need for clearer specifications to the theory as not all systems with complicated orders are complex systems.

Meyer (2011) concurred and stipulated that over the years, the features used to describe the complex systems have been expanding owing to its wide application across research disciplines. Taken from different contexts such as biology, physics, computer science, sociology and economics, the combination of features for complex systems are becoming more diverse, suited accordingly to the respective subjects.

Onnela (2006) also underscored the growing acknowledgement of the complex systems theory as a discipline that bridges the gapsMalaya between various well established sciences such as chemistry, physics andof biology and its increasing influence that transcends the boundaries of natural science into areas such as economics, psychology and social science.

In the early days, the complex systems theory was generally more prevalent in the areas of computer science and biology, and to a lesser extent in sociology and economics whereas the chaos theory was more widely applied in mathematics, physics and to some extent, economics. Of late, the complex systems theory has established a firmUniversity footing in economics, particularly in financial economics (apart from maintaining its inherent influence in natural sciences and mathematics).

2.2.5 Measurement without Theory

The two main theories that form the pillars of this research are the complex systems theory and the notion of the “measurement without theory” with the application of leading indicators. The research attempts to be faithful to the fundamental tenets of both theories particularly the complex systems theory in its emphasis of allowing patterns to emerge in stages and avoid generalising data too early. Hayek (1964) perhaps laid the first foundation on the philosophy of how complex phenomena should be dealt with when he lamented:

“… Statistics, however, deals with the problem of large numbers essentially by

eliminating complexity and deliberately treating the individual elements which counts as if they were not systematically connected.”Malaya According to Sornette (2003), a complex system must be scrutinised at the appropriate level of specification to captureof the essence of the phenomena. Thus, when such a system is being examined, decomposition of stages or disintegration of some parts may be required to exclude some details until the right level of conception is achieved. Arthur (1995) in the same wavelength suggested that all researches on complex systems should take into consideration the complication catalysed by the multi-facet components that adapt or react to the behaviour that emerges within the system. These components would continuously adjust to form a cumulative pattern and theUniversity pattern on the other hand would react to the components in a continuous loop. With the exception to the emergence of an asymptotic state or equilibrium, the evolution of the dynamical complex systems would perpetually continue.

2.2.6 Complexity of Economics

It is prevalent for economic theories and hypotheses for empirical studies to be derived from the notion that the economy is of a deterministic order, linear and founded upon a static equilibrium system bounded by a restrictive set of assumptions. The contradictions between economic theories and the reality of economy could not be further argued as meticulously as Beinhocker (2006) and Sornette (2003).

The divergence is ironic to the fact that in view of economics complexity, not dissimilar to the complex systems theory, is not something novel and has been deeply entrenched in the very early works of some very prominent philosophers in economics.

For example, Karl Popper who served as a professor at the London School of

Economics in 1940s (who is also widely acknowledged to be one of the greatest philosophers in the 20th century) through his workMalaya in the problem of induction had advocated academia to gain insight into of non deterministic order in economics (SEP, n.d.). Friedrich A. von Hayek, a recipient of the Nobel Memorial Prize in Economics and a former professor in London School of Economics on the same page concurred and advocated rigorously on the need of adopting a holistic approach for economic analysis in a series of papers (Nobleprize.org, n.d.).

Hayek (1964) in his defining forethought on the nature of complexity articulated that “the phenomena of life, of mind, and of society” are much more complex than that of Universitynatural sciences. The emerging patterns from the complex interactions of irregular phenomena in economics and social sciences has a higher degree of complexity compared to patterns that emerge from the complex combination of elements with constant relationships within a deterministic structure in fields like physics. The very fact that economics has to impose the ubiquitous ceteris paribus restriction in the

64 construction of theories and the frequent irregularities of outcomes that defy empirical predictions demonstrate that economics is essentially a field of complex phenomena.

Colander (2000) chronicled that economic thought initially evolved from informal stories of political economy in the era of Adam Smith to the classical economics, which assigned values for theories. It later progressed to Neo-classical economics, which advanced theories in the form of general equilibrium. In the course of time, economics has gradually progressed into the path of natural sciences. Researches are focused in simplifying complicated hypotheses into structural mathematical equations or statistical model and theories are tested on empirical modelling of historical data.

According to Colander (2000), econometric modelling using macroeconomic variables has in fact accentuated the fallibility of modernMalaya economic analyses because most microeconomic and general equilibriumof hypotheses do not conform to empirical findings albeit sharing a common theoretical framework.

The economy is much more complex than as perceived by standard economics because the conventional assumption of “far-sighted rationality” of individuals does not hold. In reality, individuals cannot rationally handle all aspects of the economy on their own. Institutions were formed to draw up policies to deal with economic issues. These institutions in turn would influence the behaviours of individuals in the economy. The interventionUniversity of institutions on the market that rarely conforms to the rational expectation from a micro level thus renders the proposition that the outcome of economy is based on the rational expectation of individuals as invalid.

Over the decades, the proliferation of literature based on various simplified premises in explaining the complex world has drawn many criticisms. Carroll (2001) in his assertion on The Epidemiology of Macroeconomic Expectations reiterated the flaw 65 of rationality assumption and argues that elaborated mathematical models parameterised with the rational expectation assumption are ineffectual and should be replaced with more realistic and explicit models which capture the dynamism of the economy as a whole. No single set of the few variables in a model could consistently produce accurate forecasts in the long run. Beinhocker (2006) concurred and noted that economic theories since the early days are too intertwined with the mathematics of equilibrium which require the contrivance of highly restrictive assumptions. Such developments have increasingly detached theoretical economics from the real world.

The Lucas Critique famously argued that forecasting results derived from econometric models would immediately become obsolete and ineffectual when the optimal decision rules are negated by the policies enforced based on the model itself. Subsequently, the outcome would also react systematicallyMalaya into the model and change its original structure (Lucas, 1978). of Echoing the proposition of Lucas Critique, Arthur (1995) in similar rationale argued that the complexity of the economy and financial markets are due to many reasons and the underlying causes that perpetuated failure of conventional forecasting are due to the explanation as follows:

“Actions taken by economic decision makers are typically predicated upon

hypotheses or predictions about future states of a world that is itself in part the Universityconsequence of these hypotheses or predictions. When we attempt to model how such predictions might be generated we become stymied: the predictions some

economic agents might form depend on the predictions they believe others might

form; and the predictions these might form depend upon the predictions they

believe the original group might form. Predictions or expectations can then

become self-referential and deductively indeterminate. This indeterminacy

pervades economics and game theory.”

One of the classic examples of the Lucas Critique is the flawed interpretation that the Phillip curve i.e. the empirical evidence that showed inverse relationship between inflation and unemployment could be capitalised to calibrate the economy to a desired outcome. The structure of the model would change should there be an attempt to artificially perpetuate inflation through monetary policy as the market would alter employment decisions, based on the expectation of inflation (Ljungqvist, 2008).

Another example that illustrates the complexity of economy is a scenario where the deliberate attempt by the Federal Reserve to prevent the stock market from an overdue correction through monetary easing could instead exacerbate the inflation of the market bubble and increases the level of speculationMalaya (Vines, 2009). Such instances create a chained action-reaction feedbackof loop between the Federal Reserve and investors as the Federal Reserve on one hand would adopt monetary measures to support the market and on the other hand, investors would take advantage of the measures to maximise their return. The emerging behaviour of the stock market continues on until it reaches a tipping point and falls like an avalanche.

2.2.7University Evolution of the Complex Systems Theory in Economics In the mid-1980s, a group of renowned scientists from interdisciplinary sciences banded together and established one of today’s foremost research centres in complexity sciences, the Santa Fe Institute. The establishment created a very important foundation for the diffusion and cross sharing of ideas based on the complex systems theory from across disciplines such as physics, chemistry, biology, computer science and economics.

In due course, the institute had helped produced many literature that offered alternative theories and quantitative methods in resolving problems based upon the general principle of “inclusiveness and broad perspective, one that comprehends the components of a system but views those elements as actors in a large, interconnected, often unpredictable world” (SFI, n.d.). The Santa Fe Institute has compiled some of the most important literature on economics in the context of complex systems, namely as to date (2015), three volumes of “The economy as an evolving complex systems”.

Hayek (1964) in one of the earliest literature in this area that suggested that the simplification of abstract patterns with the general statistics methodologies overlooks the actual complexity of phenomena. By addressing problems from an elevated view, the general modelling approaches have the tendency of disregarding the changes of fundamental dynamism of relationship among elementsMalaya and the organisation of structure within a system that occurs over time. The development of the economics nevertheless took a very different direction and the seedof of complexity epistemology did not gain a footing into mainstream economics until many decades later.

Approaches used in researches that are based on the complex systems theory are distinctive and wide ranging. The diversity in the approaches is attributed to the wide espousal of the theory by different fields of studies. Methodologies conceived by researches that examined their problems from the lens of the complex systems theory departUniversity significantly from the conventional methodologies for all of the respective fields. This puts Hayek’s (1964) criticism into perspective on the general treatment for complex problems and the importance to rectify such deficiencies. Among some of the methods are network models (i.e. small-world networks, random Boolean networks, and neural networks etc.), Markov processes (i.e. Levy’s flight and Brownian motion), bifurcations and diffusion, fractal and cellular automata etc. (Gros, 2008).

In summary, methodologies used for complex systems researches generally allow the information of individual units to be retained as much as possible and permit the dynamism within the structure and the correlation among units to change as when required. Some methodologies (e.g. network graphs, fractal and cellular automata) are not developed for the purpose to prove or reject hypotheses like the conventional statistical approaches. On the contrary, predictive patterns are allowed to emerge freely from their initial abstract forms. The main objective is to observe how the complex systems unfold into patterns that allow for some extent of generalisation and predictability rather than fitting observations into a predetermined framework or simplifying a phenomenon, which is inherently not simple. The flexibility in applying the complex systems theory, albeit not necessarily straightforward has enabled research problems to be scrutinized from a wider perspective.Malaya Methodologies can be developed hierarchically or in stages through simulations or exploratory trials and errors as the observations move along. of

2.2.8 Complexity of Financial Markets and Stock Market Crashes

Financial market crash depicts a meltdown of financial institution and stock market crash is commonly described as a brief but abrupt and sharp drop in the price of stocks or stock market indices. These two events commonly occur synchronously and the causalityUniversity could go either way. Stock market crash is specifically caused by panic that sets into the market resulting in overwhelming sell orders all at once. Market crashes have great adverse impact on the economy and devastating social implications.

A crash which is caused by the inherent market mechanism such as speculative bubble is categorised as endogenous. The most typical definition of speculative bubble is the prices of stocks (or prices of assets in general) being overvalued due to an 69 unreasonable market demand. A crash could entail prolonged decline in the market that lasts for months or years (Jacobsson, 2009). For centuries, the world had witnessed numerous endogenous crashes, some more devastating than others. Among the most severe and well documented crashes include the 17th century “Tulip Mania”, the 18th century “South Sea Bubble”, the “Great Depression” in the 1930s, the “Black Monday in 1987, the “Dot Com Bubble” in the late 1990s and the “Subprime Financial Crisis” in the late 2000s (Malkiel, 2007; Pele & Mazurencu-Marinescu, 2012; Vines, 2009).

Throughout the course of history, random events that created shock in the market such as the “9/11 tragedy” and the breakout of war i.e. the WWI had caused crashes too.

Such isolated non-economic events that triggered crashes are categorised as exogenous

(Jacobsson, 2009).

A healthy economy thus rarely triggers a financialMalaya crisis or a stock market crash. This perfectly logical and almost universally accepted postulation nevertheless runs into contradiction with the efficient market hypothesisof (EMH) of which most financial literature were built upon. According to Cooper (2008), the EMH assumes that asset prices are always in equilibrium and mirror the asset value correctly at any time, adjusted based on all available information in the market. The hypothesis is thus ignorant to the fact that in most times, a market rally occurs even when stock prices have reached an overly inflated and unsustainable level due to speculative herding

(Shiller, 2005). Shiller (2002) and Cross, Grinfeld, Lamba & Seaman (2005) argued thatUniversity the widely accepted notion of the EMH is fundamentally flawed because it fails to capture the critical attributes of the market behaviour in reality and the shortcomings

“manifests itself most clearly in the real-world phenomena of non-Gaussian market statistics such as fat-tails, excess kurtosis and volatility clustering (and the corresponding market bubbles and crashes)”.

Cooper (2008) highlighted that the brink of stock market crashes throughout history was often marked by the occurrence of sharp increases in market volatility.

From the behavioural finance viewpoint, the severe volatility at this juncture is caused by the continuous feedback loop of actions and reactions between cautious traders looking for indicators to pull out from the market and the monetary authority (i.e.

Federal Reserve, central bank etc.) strategising to keep the market afloat based on the traders’ manoeuvres. The market movement, which generally trends with the diffusion of information creates a perception that the EMH is the most viable explanation to a crash when the final piece of information sends a market to a free fall. Notwithstanding, the information could be purely a trigger that bursts a bubble that has built up over time from a prolonged market rally. Thus as noted by Arthur (1995), a financial meltdown or a stock market crash is merely the tipping point of theMalaya underlying complex phenomena that occur throughout an extended course of time.

White (2008) concurred and notedof that in almost all financial crises in history, the overpricing of assets was largely due the credit expansion measure which precedes a crash. The excessive flow of credit into the market enhances the market’s optimism and increases risk-taking endeavour among investors. Progressively, the asset prices would deviate from the intrinsic asset values and the fundamentals of the economy due to the artificial market sentiment. The distortion of economic fundamentals would then manifest in the change of the consumption-investment pattern. At the critical point whenUniversity the market has transcended the psychological threshold where the realization of an unrealistic level of asset prices catches up with the market expectation, “the whole endogenous process would go into reverse”. When the speculative bubble bursts, the economy would be the collateral damage and adversity would be exacerbated by the strain in the financial market due to the prior credit expansion. The feedback loop is once again set in motion but on a reversed direction. White (2008) stressed that most of

71 the forecasting models only describe a section of the whole occurrence. Due to the complexity of the market, a qualitative assessment would seem to be a more appropriate option.

From the angle of econophysics (Sornette, 2003), a stock market movement is similar to the business cycle where the market goes through cyclical transitions over time from a stable state to an unstable state before finally crashing. There are five common stages to the build up of bubble that leads to an imminent stock market crash: displacement, take-off, exuberance, critical stage and crash. A stock market crash is caused by the gradual evolvement of the market towards the state of instability due to the progressive ascendance of market price over an extended period of time. The inherent herding nature of traders in the market especially during the market’s upward trend reinforces the market optimism and creates Malaya a loop that further inflates market bubble. Therefore, the explicit cause that triggers a market collapse is merely superficial. When the market passes the of instability threshold, any minor exogenous disruption would catalyse a meltdown. As such, the market is essentially a complex system that encompasses a network of individual systems that are dynamic and resembles self-similarity behaviour. Interactions among the extensive integrated units within the system usually exhibit emerging and self-organising patterns.

2.3University Classification and Evaluation of Literature

The section on the classification and evaluation of literature synthesises and provide critical comments on the key studies related to the research. The main aim of this section is to illustrate how most schools of thought that studies the subject of stock market declines are separated into two categories. Discussion in this section forms the foundation to the theoretical framework of the research. The first category contains 72 schools of thought and studies in which the methodologies are incompatible with the general objective of the research, which is to empirically determine the best market fundamentals for the prediction of stock market declines. The second category conversely groups together schools of thought and key studies that can be congregated and empirically tested in the same platform to achieve the overall research objective.

2.3.1 Post-mortem Reviews and Theoretical Studies

The earliest literature on stock market declines concentrated almost entirely on specific events of dramatic crashes. Notable crashes that received the most extensive study are the 1929 Wall Street Crash (the Black Tuesday) and the crash of 1987 (the Black Monday). The early literature on the 1929 crashMalaya were predominantly post-mortem reviews with minimal quantitative analysis. Such limitation might be due to methodological constraints in modelling a offew decades ago. Furthermore, it is apparent that most of these older studies commonly restricted their focus on the unique causes for selected cases of market crash instead of comparing and deducing the general trend that causes crashes. The research reckons that this could be due to the lack of crash samples for meaningful comparison in the early days.

Speculation and manias were the main themes used by Galbraith (2009) to describe the 1929 crash in his widely cited classic literature. Edie (1930) examined the roleUniversity of banks on the catastrophic event. Currie (1934) proposed that the botched monetary policy was to be blamed for the crash and the ensuing depression. Willis

(1930) argued that there was a market panic which resulted in the crash. Robinson

(1930) faulted investment companies for their role in the stock market meltdown.

Bolton (1960) and Heller (1987) revisited the 1929 major stock market decline many years later and compared it with a series of subsequent stock market turmoil. Romer 73

(1990) on the other hand examined the reverberation of the crash and proposed that the drop in the consumer sentiment subsequent to the crash was the cause of the Great

Depression. Klien (2001) produced a seminal study on the 1929 crash and found wide ranging and inconclusive answers to the cause of the devastating event. Nicholas (2008) proposed that the intangible knowledge capital (i.e. patented innovations) has caused the market to become over optimistic and thus overvalued prices of stocks prior to the crash.

Although patents are an imperfect measure of innovation, and intangible capital covers a broader set of assets than patented inventions, the measures used here provide important information about how technological change influenced the expected increase to a firm's future free cash flow.

The rare occurrence of dramatic crashes has Malayaled to the general treatment of such events as outliers in the early days. Marketof declines that were less spectacular, with prolonged and persistent downturns (i.e. bear markets) had mostly eluded the attention of economists too, until recently. The lack of statistical means thus had resulted in the absence of viable platforms for empirical analysis on the 1929 stock market crash as evident in the old literature. As the paradigm in financial economics shifted gradually in tandem with the advancement in empirical methodologies, studies on stock market crashes had also progressively improved and allowed researchers to reexamine the crash of University 1929 in new angles decades later along with other more convincing quantitative evidence.

Sirkin (1975) applied the price per earning analysis and rejected the common view in the early days that speculation had led to the 1929 crash. De Long & Shleifer

(1991) found that close-end mutual funds played a role in the market catastrophe.

Wilson, Sylla & Jone (1990) postulated the view of panics and excessive volatility as

74 the main factors to the crash. White (1990a, 1990b) suggested bubbles were the cause of crash. Using the “pricing of loans to stock broker” ratio, Rappoport & White (1993) proposed that there were no bubbles prior to the 1929 crash.

Rappoport & White (1994), through the option-pricing models argued that the major decline was expected at least a year ahead as indicated by the increase in “implied volatility”. Donaldson & Kamstra (1996) also questioned the hypothesis of bubbles induced crash with new evidence based on dividend yield of the market. Baur, Quintero

& Stevens (1996) supported the proposition that the mass sell-off of the market was due to negative market sentiment instead of fundamentals. McGratten & Prescott (2004) using the growth theory, concurred and showed that market fundamentals were robust and stock prices were not overvalued at the juncture of the crash. Thorbecke (1994) however argued that systematic risk and economic fundamentalMalaya (i.e. the unprecedented increase of trade deficit by year 1984) playedof a part in the crash. Naturally, studies on the 1987 stock market crash improved immensely in terms of analyses which were backed with empirical evidence in relative to earlier studies on the 1929 crash. The improvement was owed much to the availability of better quality time series data and methodological advancement in progression with time. The most common view pinned cause of the 1987 market meltdown to the speed of computerised trading (more specifically the speed of selling) which triggered a panic herding. Jahne (1987)University proposed that the market was overvalued and the sharp crash was due to the all- round advancement in the financial market, which allowed traders to sell-off speedily in response to new negative information.

Likewise, Furbush (1989) found evidence that support the argument that program trading exacerbated market sell-off. Roll (1988) examined the spillover effect of the 1987 crash to other markets around the world and reasoned in the same

75 wavelength that “continuous auctions and automated quotation” had resulted in sharp declines in these markets. Gennotte & Leland (1990) employed a rational expectations model to explain the role of market liquidity and the hedge towards the crash.

Another popular theory of the 1987 crash is the misalignment between spot prices and future prices of the market (Kleidon & Whaley, 1992). Antoniou & Garrett

(1993) and Bates & Craine (1999) found evidence that linked the 1987 crash to stock index future. Bates (1991) suggested that the crash was expected based on the movements and indications shown in the option prices prior to the crash. Blume,

Mackinlay & Terker (1989) also found evidence of disintegration in the relationship between future prices and the spot index during the 1987 crash.

Harris (1989) also showed evidence of nonsynchronous trading and breakdown in the linkages between the stock market index andMalaya future contracts in the 10-day duration surrounding the crash. Similarly,of Bernanke (1990) also argued that delayed clearance and settlement of future contracts attributed to the crash. Kheidon (1992) showed evidence that stale prices were the main factors de-linking the spot and future markets.

A few other literatures focused specifically on the volatility of the market before, during or/and after the 1987 event. Kim & Kim (1996) utilised the UC-MS model to examine the heteroscedasticity components of the market during the crash and proposedUniversity fleeting transient shock as the reason for the panic disposal of stocks. Choudhry (1996) examined the volatility of six emerging markets before and after the

1987 crash with a GARCH model and found persistent but non-uniform fluctuation over the same timeframe in these markets. Masulis & Ng (1995) used a GARCH model to study the London Stock Exchange after the “Big Bang” restructuring and the 1987 market disaster and increased volatility in intraday trading for both events.

The remaining literature examined wide ranging viewpoints and themes on the devastating event of 1987. Greenspan (1987) made a generalisation between the great declines of 1929, 1987 and the possible repeat of history on a similar magnitude in the future by pointing towards the market’s inherent problem of excessive speculation in the housing market, huge mortgage debt and massive dollar overhang in foreign market.

Rubinstein (1988) related the catastrophe to failed portfolio insurance strategies.

Jacklin, Kheidon & Pfleiderer (1992) concurred and pointed out that the market had underestimated the dynamic hedging systems of portfolio insurance. Black (1988) examined the devastating event using an equilibrium model. Gammill & Marsh (1988) investigated the trading behaviour and the movement of stock prices leading up to the meltdown. Sinai (1988) studied the aftermath of the crash and how the economy emerged in the subsequent year. Malaya Eagle (1994) proposed that 1987 cascadingof scenario (i.e. the snowball effect on the stock market due to the chain of consequential events relating to portfolio insurance and index arbitrage) could be explained with the Leland's backward-bending demand curve. Seyhun (1990) proposed overreaction as the cause of the crash based on the market insiders’ reaction of extensive buying of stocks right after the sharp decline.

Seguin & Jarell (1993) rejected the role of margin activities, as suggested by some literature, as the cause of the decline.

UniversityRaines & Leathers (1994) investigated the relatively weaker immunising effect of the 1987 meltdown compared to the 1929 predecessor and attributed the phenomenon to the speculative appetite of modern-day traders. Levy (2008) introduced the notion of

‘social phase transition’ to explain major market meltdowns like the 1987 event.

Wigmore (1998) revisited the Black Monday and reviewed the major themes and propositions covered in most literature such as panic selling during the crash, the

77 imposition of policy after the crash and also discussed the possibility of a similar crash in the future.

Most discussions on major crashes in recent years are generally assimilated in the context of selected theories from conventional schools of thought. Literature that is independent of influence from any school of thought, which examined major individual crashes occurring beyond the 1987 meltdown, has emerged sparingly. Wang, Meric, Liu

& Meric (2009) for example attempted to identify the underlying patterns to 8 selected dates of major crashes in history by using a range of financial ratios as proxies to reflect the different characteristics of firms in the market.

Other literature is published in the form of commentaries (e.g. Persaud, 2007) which based the analyses of crashes on deductive reasoning. Newer literature on crashes that are least entrenched to the conventional theoreticalMalaya frameworks tend to have an expanded scope which covers the entire of financial market meltdown, e.g. Fernandes (2006), Friedman & Abraham (2009), Harmon, De Aguiar, Chinellato, Braha, Epstein,

& Bar-Yam (2011), Veldkamp (2005) and White (2008).

Beyond the 1980s, quantitative methodologies in economics had advanced leaps and bounds and a number of financial economic theories had become well established, namely the EMH and rational expectations and its variants. The establishment of such well-argued schools of thought had allowed better generalisation on themes. The availabilityUniversity of more advance modelling and time series data had enabled the issues and observations to be analysed more rigorously.

Most theoretical / non empirical literature from the EMH school of thought do not explicitly reject the existence of stock market crashes nor recognise the trend of bull markets. However, the EMH based on the interpretation of Fama (1998) believes that overreaction or under-reaction of the market is bound to happen. Most of the other 78 rejections on such occurrences are implicit and self-evident in the espousal of random walk and the full reflection of information in the price of assets (e.g. Dupernex, 2007;

Fama, 1970; 1991; LeRoy, 1989; Malkiel, 2003; Risso, 2008).

Johansen (1997) and Sornette (2003) in their dissertations related to complex systems also argued explicitly that stock market crashes do not conform to the EMH.

French (1988) on the contrary justified the 1987 major market crash as an evidence of market efficiency with a retrospective qualitative analysis on the various information disseminated prior to the eventful crash. Szafarz (2012) concurred that financial crises could be explained within the EMH framework.

The study of stock market crashes from the rational bubbles school of thought

(which is considered to be an extension to the rational expectations theory in financial economics) evolves fundamentally on the economic-behaviourMalaya theoretical frameworks operationalised with mathematical economicsof e.g. choices of agents, supply and demand, equilibrium, expected payoff etc. The majority of the literature from the rational bubbles theory (and its wide-ranging permutations) does not examine crashes in asset markets categorically, but bubbles being the antecedents to eventual market corrections are self-explanatory in the context of crashes.

Examples of literature that cover specifically the context of rational expectations, bubbles and crashes include Lux (1995) in his “Herd behaviour, bubbles andUniversity crashes” and Sornette & Malevergne (2001) who gave an in-depth explanation on the topic in their literature, titled “From rational bubbles to crashes”. Caplin & Leahy

(1994), Bulow & Klemperer (1994) and Madrigal & Scheinkman (1997) respectively proposed a model of market dynamics with multiple stages that generalised the behaviour of traders in reacting to new information leading up to a market bubble and subsequent crash. Barlevy & Veronesi (2003) elucidated how crashes in stock market

79 are due to “rational panic”. Sandroni (1998) devised a dynamic market model free of intrinsic uncertainty with the exception of sunspot to explain the heterogeneous expectation in the market that results in crashes.

Among other classic literature on rational bubbles theory, Blanchard & Watson

(1982) first proposed that bubbles in financial market are consistent with rationality;

Camerer (1989) reasoned that speculative bubbles are due to the deviation of asset prices from intrinsic values; Diba & Grossman (1987, 1980a, 1980b) focused on reconciling rational bubbles theory into the erratic and sometimes explosive movement of stock price; Grossman & Stiglitz (1980) counter-argued the validity of the EMH theory based on the “noisy rational expectations” model; Smith, Suchannek & William

(1988) simulated a trading environment to show that bubbles and crashes are caused by deviation of stock prices from intrinsic value (i.e.Malaya dividend) fuelled by capital gains expectation; and Tirole (1982, 1985) reconciled rational expectations theory with asset speculation and examined the “rational foundation”of for asset prices respectively. Stiglitz

(1990) wrote an extensive review on bubbles.

More diverse concepts extending from the rational bubbles theory have emerged in recent literature. Chen & Duan (2011) explained the herding behaviour that leads to financial bubbles with a combination of the well-established bounded rational expectation with “adaptive expectation belief”. Conlon (2004) introduced a "finite horizonUniversity nth order" rational bubble for prices of assets. Froot & Obstfeld (1991) conceived the idea of “intrinsic bubbles” for stocks.

Other contemporary literature tested the rational bubbles based on various assumptions in simulated environments. Loewenstein & Willard (2000) introduced a probabilistic methodology to examine bubbles in a condition where agents trade non- stop. Hugonnier (2012) on the other hand showed how portfolio constraints under

80 equilibrium conditions could induce rational bubbles. Loisel (2009) utilised a “closed” rational expectations model to examine the “bubble-free” feedback rules.

A number of literatures re-examined the rational bubbles more comprehensively;

Santos & Woodford (1997) revisited the theory through a methodical study within the framework of intertemporal competitive equilibrium; Lux & Sornette (2002) showed that “fat” power tail could be predicted with rational bubbles; Rubinstein (2001) furnished extensive evidence of events which seemed to reflect market irrationality but proposed that markets are at times “minimally rational” and at other times “hyper- rational” due to overconfidence; Smith, Van Boining & Wellford (2000) studied the effect of dividend timing on asset bubbles under a simulated environment; Zeira (1999) proposed that bubbles and crashes are caused by “informational overshooting”; and Barbarino & Jovanovis (2007) examined the causesMalaya of growth in optimism of the market prior to crashes with a modified Zeira-Robof model. Separately, Loewenstein & Willard (2006) questioned the core assumption of rational bubbles with their criticism on the famous DeLong, Schleifer, Summers, and

Waldmann model (DSSW) which highlighted the role of noise traders on the over- inflation of asset prices. This influential literature argued that non-clearing markets and limited liability of investors are the main causes to the divergence in asset prices and risks from their expected level instead of the irrational behaviour of noise traders.

UniversityIn summary, most of the literatures on rational bubbles do not deal with the topic of stock market declines directly. Literature in this area is focused in examining the decisions and behaviour by individuals in the financial market and attempt to explain how the investors’ collective behaviour would lead to the overpricing of financial assets and the generation of bubbles within the scope of rational expectations assumptions.

Most specification tests for rational bubbles revolve on examining the divergence of

81 actual asset prices with their fundamental valuation (i.e. dividend yield) from their variance bounds (e.g. Shiller, 1981; West, 1987; and LeRoy & Porter, 1981). McQueen

& Thorley (1994) noted that tests based on variance bounds have a shortcoming of assuming linearity for the entire series of data to the preceding observations with a single set of parameters.

Behavioural finance in the same context however, perversely argues that irrationality is prominent in the decision making process of investors (e.g. Barberis &

Thaler, 2007; Daniel, Hirshleifer & Teoh 2002; Shiller, 2005; Vissing-Jorgensen, 2004).

The evolution of behavioural finance began with the clinical cognitive studies by psychologists on the behaviour of individuals in their decision-making process pertaining to their economic interests, specifically the investment process (Sewell, 2010). Malaya This school of thought graduallyof expanded into the studies on how the behaviours of individuals influenced by various factors such as cognitive biases, sentiment, framing effects, heuristics, herding, bounded rationality etc. (in which are contrary to the fundamental economic assumption of rationality) affect the economy in diverse aspects such as the fluctuation of prices in the asset markets (Odean, 1999). It is noteworthy that another focal facet of behavioural finance includes the “identification of anomalies in the efficient market hypothesis that behavioural models may explain” (Dargham,University n.d.; De Bondt & Thaler, 1985). Among some of the other more recent and impactful literature related to stock market decline (bubbles generation in particular) from the behavioural finance school of thought include; Abreu & Brunnermeier (2003), who examined the nexus of bubbles and market crashes from the behavioural perspective; Brunnermeier (2008), a seminal review on bubbles; Choi & Jayaraman (2009) on stock market declines due to

82 overreactions; Duffy & Unver (2006) with a self-explanatory title of “Bubbles and crashes with near-zero-intelligence traders”; Focardi, Cincotti & Marchesi (2002) on the self-organising behaviour of traders and how it leads to crashes; Guerrero, Stone &

Sundali (2012) on the anxiety that sets in through and after a sharp stock market decline; Scharfstein & Stein (1990), Trueman (1994) and Welch (2000) on the herding behaviour in asset markets; Kindleberger & Aliber (2005) reviewed the history of financial crises and illustrated how extreme optimism and the subsequent panic and overreaction lead to major crashes; and Ofek & Richardson (2003) who focused on the irrational optimism during the internet stock mania and the ensuing crash in the late

1990s.

Some sections of literature are not focused directly on bubbles or stock market declines but examined the various states of psychologyMalaya and behaviours of investors (e.g. optimism, overconfidence etc.) in contributing to the erratic movements of the markets.

Barberis, Shleifer & Vishny (1998) and Deof Bondt (1995) examined the sentiment and psychology of traders in asset markets. Barone-Adesi, Mancini & Shefrin (2012) estimated the “risk aversion, optimism and overconfidence”.

Camerer (1987) and Chang, Jiang & Kim (2009) investigated the bias judgments in trading. Cross, Grinfeld, Lamba & Seaman (2005) modelled the psychological threshold in investing. Daniel, Hirshleifer & Subrahmanyam (1998) and Hong & Stein (1999)University examined the overreactions and under-reactions of the markets. Greenwood & Nagel (2009) studied how the trend-chasing behaviour of the least experienced traders in the market contributes to asset bubbles.

Hong, Scheinkman & Xiong (2008) proposed that bubbles are caused by the existence of intelligent and naïve investors in the market taking advice from heterogeneous advisors which leads to biases and overreactions. Howard (2012) did a

83 seminal study on how cognitive psychology and neuroscience impacts the judgment of investors in the market. Lo, Repin & Steenbarger (2005) studied the “fear and greed” of traders. Shiller (2000, 2002) investigated the causes of bubbles by human factors i.e. confidence, judgments and opinions. Brav & Heaton (2002) examined the conflicting theories between irrational investors (behavioural finance) with rational investors with incomplete information (rational expectations) and found that the “mathematical and predictive similarities” of both school of thoughts are very close. Shiller (2003) reasoned how the eminence of EMH is replaced by behavioural finance.

The emergence of behavioural finance has also challenged the core assumptions in the rational bubbles school of thought. Literature on speculative bubbles evolved at the same time with both the behavioural finance and the rational bubbles schools of thought. The theoretical and methodological approachesMalaya of speculative bubbles are not dissimilar with the rational bubbles with core emphasis on bubbles generation in the asset markets and rational expectations. Nevertheless,of the notion of speculative bubbles proposes that even in the presence of rational behaviour of investors, it is common for stock prices to persistently depart from the projected fundamental values under rational expectations (Dezhbakhsh & Demirguc-Kunt, 1990 and Flood & Garber, 1982).

According to McQueen & Thorley (1994), speculative bubbles tolerates the divergence of asset prices from their fundamental values but on the conjecture that investorsUniversity are not irrational because investors rationalised that they could continue to capitalise on the growth of bubble for capital gain even if the prices of stocks have reached a precarious level.

The earlier literature mostly focused on examining the nature of speculative bubbles and conceptualising specification tests for the theory. Diba & Grossman

(1988b) first termed the speculative phenomena of stock prices in the context of rational

84 expectations as “explosive”. Flood, Hodrick, & Kaplan (1986) proposed that dividend yield is the component to measure rationality for stock valuation and therefore, it has to be utilised as the control variable to test stock market bubbles.

Leroy & Porter (1981) tested for bubbles through the divergence of stocks’ present value from their variance bound rational prices whist Grossman & Shiller

(1981) and Shiller (1981) concurred that the price volatility in reality surpasses the variance bounds of ex-post rational price to a measured extent. Hamilton & Whiteman

(1985) proposed that actions by agents who are seemingly insignificant in the market could produce huge impact to prices of assets and result in the formation of speculative bubbles.

Hamilton (1986) and West (1987) explored specification tests for speculative bubbles. Harrison & Kreps (1978) were one of the Malayaearliest to suggest heterogeneity of expectations among investors leads to speculationof in the market. However, West (1988) in his seminal work dismissed the impact of noise traders on speculative bubbles.

Shiller (1990) highlighted the popular themes and the shortcoming of various models used to study speculative bubble and crash in 1987.

Literature on speculative bubbles gradually advanced to the themes of simulation and modelling under various assumptions and market conditions. Flood &

Hodrick (1990) reviewed and classified literature on speculative bubbles whereas Evans (1991)University highlighted the common issues in testing speculative bubbles and argued that there were many flaws in approaches of classic literature. Ikeda & Shibata (1992) studied bubbles in a simulated condition where dividend growth is random and continuous. Scheinkman & Xiong (2003) investigated a “continuous-time equilibrium model” under the condition where traders in the market could not reach a consensus on the right values for market fundamentals due to overconfidence.

Kirmana & Teyssiere (2005) examined bubbles generation in a simulated environment where heterogeneous agents with differing views on the market interact randomly. Leoni (2009) extended the literature to the classic general equilibrium on financial market through the investigation on how market crashes due to speculation could occur when the agents’ belief and equilibrium prices of assets do not converge under abnormal economic conditions. Wang & Wen (2009) examined an “infinite- horizon heterogeneous agent general equilibrium model” and proposed that storable goods in the economy motivate agents in the market to continue trading despite their awareness of speculative bubbles and the impending crash.

Another popular area of research for speculative bubbles is the empirical investigation of various related themes with wide ranging permutations of assumptions and statistical approaches. Adam & Szafarz (1992)Malaya expanded the coverage of speculative bubbles from the more common stock markets to financial markets.

Dezhbakhsh & Demirgue-Kunt (1990) revisitedof the test by West (1987, 1988) on speculative bubbles in stock markets with an improved and more direct methodology and found opposite results, i.e. there were no bubbles in the observed period of 1871 to

1981 and 1871 to 1988.

McQueen & Thorley (1994) showed and developed an empirical model to test the “duration dependence” hypothesis of speculative bubbles for the period of bubble generationUniversity to market decline. Bohl (2003) used the Enders–Siklos momentum threshold autoregressive (MTAR) model to investigate the “periodically collapsing bubbles”.

Phillips, Wu & Yu (2011) using the similar periodically collapsing bubbles approach concluded that the dot com NASDAQ crash in the late 1990s was caused by

“exuberance” bubbles whereas Pastor & Veronesi (2006) using the Pastor and Veronesi

(PV) valuation model, rejected the notion of speculative bubbles in the NASDAQ crash.

Some of these contemporary literature have departed from the conventional fundamental dispersion model for speculative bubbles identification and proposed various alternatives. McQueen & Thorley (1994) noted that speculative bubbles in general exhibit the characteristics of prolonged runs of bull market and a sharp decline at the end phase also commonly exhibit nonlinear form of abnormal returns. As such, the more recent empirical tests for bubbles gravitate towards the examination of statistical properties such as kurtosis, skewness and autocorrelation.

Van Norden (1996) applied a regime switching model to detect speculative bubbles for FOREX. Al-Anaswah & Wilfling (2011) tested the specification for speculative bubbles with Markov-switching. Anderson, Brooks & Katsaris (2010) using the regime-switching approach concluded that the 1990s dot com speculative bubbles were not merely confined to tech-stock. Asako & LiuMalaya (2013) developed a speculative bubbles model with time varying parameters. Homm & Breigung (2012) examined the numerous specification tests for speculativeof bubbles and compared the power properties of these tests with a “Chow-type break test”.

The adaptive market hypothesis (AMH) conceived by Lo (2004a, 2004b, 2005) does not specifically address the issue of stock market declines. The school of thought integrates multiple disciplines which among others include evolutionary biology and ecology that argue as such (Lo, 2004b);

University“…behaviourists cite as counter -examples to economic rationality - loss aversion, overconfidence, overreaction, mental accounting, and other

behavioural biases - are, in fact, consistent with an evolutionary model of

rational agents learning to adapt to their environment via satisfying heuristics,”

Therefore, the classification of the AMH is not dissimilar to those literature under the behavioural finance which did not investigate the phenomena of stock market 87 declines explicitly. The AMH school of thought in short proposes that the constant learning and adapting nature of agents in the market contributes to the constantly changing behaviours in the market. Kim, Shamsuddin & Lim (2011) is one of the most relevant empirical literature in the area of stock market declines based on the AMH elucidation.

Classic interdisciplinary literature that applied econophysics approaches in the financial market, specifically on the stock market were introduced earlier. The evolution in this area began with the chaos theory which is based on the multifractality concept conceived by Mandelbrot. The various recent forays of econophysicists in the studies on stock market (apart from the LPPL and its variants which is discussed in the following subsection) that are related to stock market declines are more diverse in terms of themes and methodologies. Malaya Most of these literature revolved aroundof the notions of Brownion motion (e.g. Arneodo, Muzy & Sornette, 1998; Andersen, Gluzman & Sornette, 2000; Lillo &

Mantegna, 2000; 2001), Lévy process (e.g. Bates, 2012; Bakshi & Madan, 1998), network and genealogical trees (e.g. Bonanno, Caldarelli, Lillo, Micciche, Vandewalle

& Mantegna, 2004; Mantegna, 1999; Onnela, Chakraborti, Kaski & Kertesz, 2002;

2003; Onnela, 2006) scaling (e.g. Mantegna & Stanley, 1997), and various other mathematical physic approaches which are related to LPPL such as exponential behaviourUniversity (e.g. Watanabe, Takayasu & Takayasu, 2007a; 2007b), log-periodic oscillations (e.g. Bothmer & Meister, 2003).

The use of log periodic oscillations was nevertheless critically questioned by

Laloux, Potters, Cont, Aguilar & Bouchaud (1999). Barunik & Vosvrda (2009) revisited the 1987 crash and showed evidence that the market movement for the endogenous crash was well explained with the cusp catastrophe model (founded on the bifurcation

88 theory) compared to the September 11, 2001 exogenous crash. Kapopoulos & Siokis

(2005) and Siokis (2012) proposed that the aftershock of stock market crashes resembles the seismology’s Gutenberg–Richter law.

In summary, literature that focused explicitly on stock market crashes, particularly those in the earlier years were generally event specific i.e. “storied” chronicles, narrative reviews on the plausible causes of a certain crash or discussions on the impact of such dramatic catastrophe (and a combination of the above). Literature from the EMH, rational bubbles, behavioural finance, AMH and to a certain extent, the econophysics approach as classified above ranges between theoretical, clinical or indirect in regard to the investigation on stock market declines. All of these literature are classified as “Post-mortem Review and Theoretical” (i.e. non-empirical) in the context of this research as they do not provide a viableMalaya platform for the modelling of stock market declines and the subsequent forecasting model based on market fundamentals. of

2.3.2 Empirical Studies I: Defining Stock Market Regimes

The second section of the literature classification categorises studies on the predictability of stock market declines into more specific themes. These themes are self- explanatory, i.e. “Cyclical” comprises empirical studies on the predictability of the bear marketsUniversity and bull markets and the predictability of stock market cycles through its close relationship with the business cycle; “Complex Systems” covers the econophysics’ methodological frontiers through the complex systems theory such as the use of multifractality and scaling (e.g. the application of the LPPL formula); and “Market

Fundamentals” consists of studies that focuses on the use of market fundamentals in predicting or deducing the causes of stock market declines i.e. literature on speculative 89 bubbles and literature that employed various financial variables and macroeconomic indicators.

The phenomenon of cyclical movement in stock markets is well observed and documented. The cyclical pattern was a primary catalyst to the development of the bull and bear market theory. Santoni (1987) contributed one of the earlier insights on the bull markets for the period of 1924-1929 and 1982-1987 (in which the run ended in a devastating fashion) without providing a clear specification to what tantamounts as a bull market and vice versa.

Santoni & Gerald (1990) published one of the earliest thematic literature based on bull markets, examined in the context of bubbles and market fundamentals. Treynor

(1998) explored the rational expectations theory in the same breath with bull and bear markets, a rare theme of research in the early days,Malaya although the discussion was more inclined to rational bubbles. As underscoredof earlier, many literatures had noted that a formal definition to bull and bear market does not exist. Lunde & Timmermann (2004) highlighted the attempt by Sperandeo (1990) in giving a general definition to the phenomenon as follows:

“Bull market: A long-term… upward price movement characterized by a series

of higher intermediate… highs interrupted by a series of higher intermediate

lows. Bear market: A long-term down trend characterized by lower intermediate Universitylows interrupted by lower intermediate highs.”

In consonance with the bull and bear market definitions parallel to the one given as above, various models in this school of thought were conceived to detect the change of regime within the data series of stock market indices. Brooks & Katsaris (2005) extended the model of van Norden & Schaller (1999) for a three-regime model to identify the “dormant, explosive and collapsing speculative behaviour” in the stock 90 market. Candelon, Piplak & Straetmans (2008) adapted the GMM model of Harding and

Pagan (2006) used to determine regime changes in business cycle for stock market.

The parametric Markov-switching model and its variants which are widely used for regime changing analysis on business cycle (see classic literature e.g. Filardo

(1994), Filardo & Gordon (1998), Hamilton (1989)) are extensively employed for studies on the bull and bear markets. Maheu & McCurdy (2000) extended Hamilton’s

(1989) Markov-switching model to identify the bull markets and bear markets synchronization of the U.S. market. Hess (2003) carried out similar study for the Swiss stock market. Gordon & St-Amour (2000) incorporated the C-CAPM model into a two- regime Markov process to capture the essential characteristics of the data used for the bull and bear markets identifications.

Guidolin & Timmermann (2005) utilised theMalaya Markov-switching model for asset allocation decisions between stocks andof bonds. Maheu, McCurdy & Song (2009) extracted the bull and bear markets using the VWRETD Index

(NYSE+AMEX+NASDAQ) with few variants for both the Markov-switching and the

B-B approach. Sarno & Valente (2005) explored the regime change of stock return with a hybrid Markov-switching vector equilibrium correction model. Cunado, Gil-Alana, de

Gracia (2010) tested the mean reversion properties of regime change in the stock market modelled with the B-B approach.

UniversityOthers calibrated the B-B algorithm based on various rationales that resulted in slight deviations in the dating of bull market and bear market over the course of history, e.g. Edwards, Biscarri & de Gracia (2003), Gonzalez, Powell, Shi, Wilson (2005),

Lunde & Timmermann (2004), Pagan & Sossounov (2003). Chang (2009), Chen

(2009), Perez-Quiros & Timmermann (2000), Ritter & Warr (2002) tested the efficacy

91 of various sets of macroeconomic indicators on their regime switching models for bull and bear markets.

Away from the domain of theory-laden financial economics, an offshoot group of econophysicists has developed an exponential oscillation regime model which is based on log periodic power law (LPPL) for the detection of endogenous crashes in the financial market in the late 1990s. Literature by Feigenbaum & Freund (1996; 1998),

Johansen, Sornette, & Bouchaud (1996) and Johansen, Sornette, & Ledoit (1999) were attributed to be the foundations to the breakthrough in modelling bubble-induced crashes. Nevertheless, the Johansen, Sornette, & Ledoit model (JLS model) were more referred to in academia due to the continuous expansion of literature based on their own model particularly by Johansen and Sornette (e.g. Johansen & Sornette, 1999a; 1999b; 1999c; Sornette & Johansen, 1997; 1998; etc.). Malaya Johansen, Sornette, & Ledoit (1999),of the pioneers of the JLS model interpret the stock market as a trend-chasing system which leaves a trail of positive feedbacks.

Traders in the market in some measure based their trading decision on others’ decisions and the loop of such interactions determines the prices of stocks. Such interactions also lead to the formation of self-similar clusters of traders which could result in the creation of market bubble. Therefore the stock market is considered as an epitome of self- organising complex systems, which resembles nature’s other “dynamically driven out of equilibriumUniversity systems such as earthquakes, avalanches and crack propagation.” The JLS model, which adopts the assumptions of the rational expectation theory

(one of the mainstream economic foundation) for its theoretical framework categorises the traders in the market into two groups i.e. rational traders and noise traders. Rational traders are defined as traders who make sound and informed decision based on the market fundamentals whereas noise traders are defined as irrational traders who

92 collectively imitate the trading decisions of others (i.e. herding behaviour). (Fantazzini

& Geraskin, 2013; Johansen, Sornette & Ledoit, 1999).

The intricate technicalities (assumptions, equations and algorithm) of the LPPL frontier are well elucidated in literature by Fantazzini & Geraskin (2013); Johansen,

Sornette & Ledoit (1999) Pele (2012) and Pele & Mazurencu-Marinescu (2012).

Criticism of the method can be found in Chang & Feigenbaum (2006, 2008), Fantazzini

& Geraskin (2013), Feigenbaum (2001), and Van Bothmer & Meister (2003). Response to the criticisms by some of the pioneers was published in Sornette, Woodard, Yan &

Zhou (2013).

The LPPL frontier has brought much attention to the econophysics’ complex system approach in the academia of financial economics. Models based on the LPPL have spawned many literatures in a relatively shortMalaya span of time. Most of these literatures utilised the formulae to examineof issues of bubble generation in various asset markets across countries. Examples of issues covered include bubbles in various financial markets (e.g. Bartolozzi, Drozdz, Leinweber, Speth & Thomas, 2005;

Johansen & Sornette 2001; Sornette & Zhou, 2006; Zhou & Sornette, 2003a; 2003b;

2006) review and analysis on the application of the LPPL (e.g. Bree & Joseph, 2010;

Cajueiro, Tabak & Werneck, 2009; Drozdz, S., Kwapien, Oswiecimka & Speth, 2008;

Fantazzini & Geraskin, 2013; Feigenbaum, 2001; Johansen 2003; Matsushita, da Silva, FigueiredoUniversity & Gleria, 2006; Vandewalle, Ausloos, Boveroux & Minguet 1999; Sornette, Woodard, Yan & Zhou, 2013), technical calibration or enhancement to LPPL models

(e.g. Chang & Feigenbaum, 2006; 2008; Yan, Woodard & Sornette, 2012). The list above is not exhaustive as literatures in this area are voluminous, albeit repetitive in themes (i.e. application of the LPPL formulae for various different assets and countries).

The LPPL is scale-invariant, which is a defined self-similar pattern that replicates itself across discrete time and space 12 (Sornette, 1998). Thus it is also commonly referred to as “discrete scale invariance”. The association of LPPL with scale-invariant is aptly summarised by Yan (2011) as follows:

“Log-periodic oscillations are associated with the symmetry of discrete scale

invariance, a partial breaking of the symmetry of continuous scale invariance,

and occurs in complex systems characterized by a hierarchy of scales.”

2.3.3 Empirical Studies II: Predicting Stock Market Returns / Regimes

Studies on the bull and bears market are closely linked with the worldview on the nexus of stock market movement and business cycle. In general,Malaya the majority of literature in this area examine how the business cycle fluctuation and the changes in the macroeconomic environment affect stockof market returns. Investigations on such relationships are justified on the grounds that the performance of the stock market is linked to the robustness of the economy. Therefore it is widely acknowledged that the stock market moves in tandem with the fluctuation of the business cycle. Selected macroeconomic indicators are statistically recognised as proxies to gauge the wellbeing of the economy and as means to determine the phase of the business cycle. UniversityGarcia (2005) presented a theoretical model to illustrate that the synchronisation of stock market bubbles and crashes with the business cycle is consistent with the rational expectations theory and the EMH. Cochrane (2005) compiled a theoretical survey that explained the link between macroeconomic and financial market. Barro &

12 Technical detail of the concept can be found in Sornette (1998). In the study, discrete scale invariance is concisely defined as “an observable O which depends on a “control” parameter x is scale invariant under the arbitrary change x →λx (1) if there is a number μ(λ) such that O(x)=μO(λx)”. 94

Ursua (2009) found a significant relationship between economic downturn and stock market crashes in 30 countries over an extended period of observation.

Separately, Casarin & Trecroci (2007), Hamilton & Gang (1996) and Smith,

Sorensen & Wickens (2006) showed evidence on the linkage between business cycle and stock market volatility whereas Chauvet (1999), McCown (2007), Munira &

Muradoglu (2008) and Perez-Quiros & Timmermann (1988) found significant dynamic co-movement between business cycle and stock market. Edwards, Biscarri & de Garcia

(2003) examined the characteristics of bull and bear market based on the countries’ extent of financial liberalisation. Basistha & Kurov (2008), Bernanke & Kuttner (2005) and Ehrmann & Fratzscher (2004) explicated the significant impact of monetary policy towards the movement of stock market. Chung, Hung & Yeah (2012) showed evidence that the NBER recession index which was used asMalaya the proxy to gauge the investor sentiment, is a good predictor of stock marketof returns. The close relationship between the business cycle and the stock markets also highlighted the efficacy of the utilisation of economic indicators in examining the predictability of stock market. Bordo (2003) documented how stock market crashes in the history were linked to productivity level within a business cycle. Chen, Roll & Ross

(1986) examined how innovations in macroeconomic indicators influenced the stock market returns.

UniversityAvramov & Chordia (2006), Chordia & Shivakumar (2001, 2002) and Cochrane (1991) combined the usage of various business-cycle related indicators to examine the predictability of stock returns. In the same vein, Chauvet & Potter (2000) tested a set of coincidental macroeconomics and leading indicators on the stock market and concluded that the stock market reacts to the foreseeable short-term contractions and economic recessions.

Hartmann, Kempa & Pierdzioch (2008) concurred and supplemented that economic indicators on looming financial crises have similar influence on stock market returns. Likewise, Nyberg (2011) used a set of macroeconomic indicators of recession for a probit model to predict the excess returns of the stock market. Lettau & Ludvigson

(2001) found significant correlation between stock market returns with the aggregate consumption–wealth ratio, on top of dividend yield, the dividend pay-out ratio and other macroeconomic indicators.

Albeit the obvious link between the stock market and the macroeconomic fundamentals, the task of specifying the most consistent indicators to even predict the business cycle has proven to be difficult. Attempts in employing macroeconomic indicators to predict the stock market returns over the years expectedly had produced mixed results. Among the earlier literature, BranchMalaya (1976) concluded that many indicators to a certain extent have predictive content for stock market performance, particularly the rate of treasury bill and theof mutual funds’ cash position. However, their predictive ability wanes over time. Breen, Glosten & Jagannathan (1989) concurred that the rate of the Treasury bill has significant correlation with stock market returns. Flood

& Garber (1980) proposed that money supply and money demand influence the performance of stock market and the generation of speculative bubbles.

Keim & Stambaugh (1986) found risk premium as a good predictor to the performanceUniversity of stocks and bonds. Fama & French (1989) in the same vein show evidence that risk premium along with dividend yield and term structure have significant predictive value for both the returns of stocks and bonds as these three variables are good macroeconomic indicators to the business cycles. Nelson (1976) and

Fama & Schwert (1977) found significance in the linkage between inflation and stock market performance. Balvers, Cosimano & McDonald (1990) showed evidence that

96 stock returns are forecastable with aggregate output. Summers (1986) nevertheless argued that stock market does not necessarily mirror its fundamental value based on

“economic realities” (i.e. macroeconomic indicators and various financial evaluations) due to the speculative nature of the market.

Newer literature on the stock market and economic fundamental linkage also revolve around similar macroeconomic indicators albeit using a larger pool of variables for each study and commonly with the combination of newer approaches. Chen (1991) used the dividend-price ratio, term structure, default premium, rate of growth in production and interest rate as indicators to the well-being of the economy to exploit investment opportunities in the financial market. Pesaran & Timmermann (1995) proposed a recursive model to test the robustness of a set of macroeconomic variables (i.e. market dividends & earnings, industrial productionMalaya rate of change, inflation rate, 12-month discount bond rate, average of bid and ask yield for Treasury bill and narrow monetary aggregates) in predicting the stockof market during different historical time frames.

Similarly, Rapach & Wohar (2005) examined the predictability of stock market returns across 12 markets with the rate of inflation, various interest rates, term structure, money supply, unemployment rate and industrial production and concluded that interest rates have the best predictive power among the rest. On the other hand, Flannery & ProtopapadakisUniversity (2002) showed evidence of a significant relationship between stock market return with money growth and inflation while Domian & Louton (1995) proposed an asymmetric model for the linkage between stock index returns and the rate of unemployment.

Campbell & Vuolteenaho (2004) also found inflation to explain the largest percentage of variation in stock market mispricing. Avramov (2002) found term

97 structure to be a good predictor for future stock market returns compared to ratios like dividend yield and book-to-market. Hong, Torous & Valkanov (2007) found that a significant number of industrial indicators could predict the direction of stock market for up to two months ahead.

Whilst most empirical researches investigate the role of macroeconomic variables on stock market returns and - to a lesser extent - regime change in the stock market, other literature depart from examining these direct linkages but focus on other related issues which nevertheless corroborate the significance of the relationship between macroeconomic indicators and the stock market. For example, Dopke,

Hartmann & Pierdzioch (2008) compared the real-time economic indicators for ex-ante stock market forecasting with revised data and found similar yield.

Pierdzioch, Dopke & Hartmann (2008)Malaya showed the linkage between macroeconomic indicators with stock marketof volatility. Racine (2001) compared the statistical properties of a linear and non-linear model for stock market prediction using financial and economic variables. Birz & Lott (2011) showed that the stock market is sensitive to news of the latest update on the macro economy. Zivot & Andrews (2002) conversely examined the unit root of various macroeconomic indicators by using the

1929 crash and the 1973 oil shock as the breakpoints for these data.

It is noteworthy that very few literatures narrow the scope explicitly into the relationshipUniversity between business cycle and bear market run or stock market crashes but this research theme has been gaining momentum of late. Among others, Chang (2009)

Chan (2009) Guidolin & Timmermann (2005) Perez-Quiros & Timmermann (2000) and

Ritter & Warr (2002) examined the efficacy of various macroeconomic indicators and financial variables in forecasting the bull and bear markets. Likewise, Kaliva &

Koskinen (2008) conceived a regime switching model and showed that stock market

98 bubbles have an inverse relationship with inflation whereas the “error-correction regime” is correlated with price-dividend ratios.

Apart from the market’s macroeconomic indicators, the relationship between the financial fundamentals and the stock market movements have long been a subject of interest among researchers albeit the contradiction with the bedrock theory of financial economics, i.e. the EMH. One of the crucial focuses of the behavioural finance also includes the reasoning of market anomalies phenomenon which are inconsistent with the EMH assumptions13 (Dargham, n.d.; De Bondt & Thaler, 1985).

Theoretical literature are relatively rare compared to a long list of empirical literature on market anomalies. Some of the examples are Marsh & Merton (1986) who proposed a theorem to examine the prices of the stock market with variance bounds test and variability of dividends. Menzly, Santos & VeronesiMalaya (2004) alternatively used a general equilibrium model to explain the of predictability of stock market returns using dividend growth and then proceed to prove the hypothesis empirically.

Schwert (2003) summarised the core essences of market anomalies and demonstrated the extensiveness of this area with themes ranging from calendar effect, data snooping, momentum effect, size effects, return from different types of investors etc. The subsets of market anomalies discussed in the context of behavioural finance which are specifically related to this research include value effect and various “predictableUniversity return through times” (e.g. earning per shares, stock returns, dividend yields, expected inflations, interest rates etc.). Singal (2004) authored an extensive guide on anomalies and mispricing in the financial markets, which is almost covered on the same ground as Schwert (2003).

13 As highlighted in the earlier section. 99

Most of the existing literatures on such anomalies accentuated on the predictability of market returns and not precisely on bear markets or stock market crashes14. Nevertheless, the prolonged negative stock market returns is synonymous to a run of bear market and a sharp and dramatic decline in stock prices is equivalent to a stock market crash. Therefore, such relationships are self-evident. It is noteworthy that literatures on the stock market relationship with macroeconomic indicators often overlap with literature on stock market anomalies. In general, literatures presented in the context of macroeconomic indicators incline to illustrate the effect of business cycles on the stock market whereas stock market anomalies may not necessarily be linked to real economic conditions. However, the employment of common variables such as inflation, term structure, interest rate etc. blurred out the line.

Examples of the stock market anomalies foundMalaya in some earlier literature include the predictability of stock returns with one or a combination of a few accounting based ratios i.e. price-earnings (Basu, 1977; Campbellof & Shiller, 1998), earnings (Campbell &

Shiller, 1988), earning yields (Basu, 1983; Reinganum, 1981), dividend-price

(Campbell & Shiller, 1998), dividend yields (Fama & French, 1988; Hodrick 1992;

Kothari & Shanken, 1997; Rozeff 1984), expected dividends (Campbell & Shiller,

1998), changes in dividends (Shiller, 1981), market value (Basu, 1977; Reinganum,

1981), firm size (Schwert, 1983; Fama & French, 1992; 1995; Daniel & Titman 1997), book-to-market (Fama & French, 1992; 1995; Daniel & Titman, 1997; Kothari &

Shanken,University 1997; Pontiff & Schall, 1998). Apart from that, Ball (1978) argued that information dissemination i.e. public announcement of corporates’ financial performance is a good predictor to stock returns. Campbell (1987) found that term structure / yield curve rate has a significant relationship with stock returns.

14 Some literature specifically that examine market fundamental / anomalies in relation with bull market e.g. Guidolin & Timmermann (2005), Perez-Quiros & Timmermann (2000) and Ritter & Warr (2002) are classified under the sub-theme of “Bull and Bear Market”. 100

Anomalies which were found to have significant predictive power on stock market returns in the more recent literature include earnings growth (Ferreira & Santa-

Clara, 2011), book-to-market ratio (Ali, Hwang & Trombley, 2003; Jiang & Lee, 2007;

Lewellen, 2004), dividend yields (Ang & Bekaert, 2007; Boudoukh, Richardson &

Whitelaw, 2008; Jiang & Lee, 2007), dividend-price ratio (Binsbergen & Koijen, 2010;

Campbell, 2008; Campbell & Yogo, 2006; Ferreira & Santa-Clara, 2011; Lewellen,

2004; Park, 2010) dividend growth rate (Binsbergen & Koijen, 2010), payout ratio - defined as the summation of dividends and repurchases divided with price (Boudoukh,

Michaely, Richardson & Roberts, 2007), earnings-price ratio (Campbell & Yogo ,

2006; Campbell & Shiller, 1988; Lewellen, 2004), yield spread (Campbell & Yogo,

2006) Ibbotson & Chen (2003) deconstructed the historical stock market returns and concluded that the long-term risk premium is closelyMalaya related to factors such as “inflation, earnings, dividends, the P/E, the dividend-payout ratio, book value, return on equity, and the GDP per capita”. of

Campbell & Thompson (2008) did a compendious historical review on the predispositions of stock market predicting in the academic in different periods. They concluded that by the early 2000s, literature in financial economics have become more accommodating to the notion of stock market predictability, particularly on the aggregated stock returns. In the 1980s, the focus was mostly on financial ratios, especially those that are related to dividends, earnings and Treasury yields. Such researchesUniversity were spearheaded by prominent economists such as Campbell, Fama,

French, Shiller, Rozeff (as noted above).

The range of predictors and indicators of the stock market has increased in the

1990s and early 2000s to include various macroeconomic indicators and variables that are related to the exploitation of information on firm level activities. Concurrently, there

101 were also researches that continued to question the validity of stock predicting by raising the issue of bias coefficients (various). The debate rages on and of late, literatures are trending on developing econometric methodologies to rectify the biased problems and improving inference on such issues15.

2.4 Theoretical Framework

The theoretical framework as illustrated in Figure 2.1 shows that the literature in the specific area of stock market declines in general can be classified into two streams, i.e.

1) Post-mortem Review / Theoretical; 2) Empirical. The diagram intends to demonstrate tentatively the flow of theoretical review for the research and how it leads to the derivation of the empirical framework in the followingMalaya section. As an overview, literature that cover the topic of stock market declines (some more comprehensively while some lesser) can be segmented into twoof main categories.

The first category in the framework comprises studies that examine the topic of stock market declines (mostly indirectly) from various perspectives, all except on the modelling of the phenomena i.e. research that are based on post-mortem analyses, reviews and theoretical discourses on the behaviour of agents in the market (e.g. assumptions that are originated from the rational expectation school of thoughts and its antithesesUniversity and the behavioural finance discipline).

15 See Campbell & Thompson (2008) for literature references on points discussed in the passage. 102

Post-mortem Study / Theoretical:

Post-mortem Study EMH Rational Bubbles Behavioural Finance Speculative Bubbles AMH Econophysics (Various) Literature* on Stock Market Declines Empirical Modeling:

Cyclical  Bull and Bear Markets  Business Cycle and Stock Market Linkages Market Fundamentals  Anomalies (Financial)  Macroeconomic Indicators Complex SystemsMalaya  Logof Periodic Power Law *Classification is non-exhaustive. Literature indicated are of the more established theories / hypotheses / frameworks / frontiers

Figure 2.1: Theoretical Framework

One of the more extensive types of literature on stock market declines that can be classified is the post-mortem study. Most post-mortem studies are presented in the formUniversity of narratives on specific crashes (for the most part) and run of bear markets. Although such literature existed mainly in the earlier eras when time series data and econometric models were limited for meaningful empirical analysis, contemporary studies that focus specifically on selected episode(s) of stock market crashes, bear markets and financial crises (a topic that relates closely to stock market declines) are still prevalent.

103

Literature in this category selectively chronicle specific incidence of market decline(s) retrospectively. Discussions include but are not limited to the suggestion on causes, consequences and future preventive measures to the specific episodes. Other types of literature in the first category include those that examine the possibility of stock market crashes and bear markets based on the assumptions of some well-established hypotheses i.e. the efficient market hypothesis (EMH) and the adaptive market hypothesis (AMH). The hypotheses and assumptions proposed by these schools of thought are largely founded on deductions drawn from seminal review of literature.

Further on, the other two types of literature classified in the first section are rational bubbles and speculative bubbles, both of which are founded in the methodological individualism approach. Generally, these schools of thought do not directly inquire the topic of stock market declines. MalayaThe main focus is the formation of bubbles in the market, which understandably lead to the eventual crashes or prolonged periods of prices correction (bear markets).of

The extensive deliberations revolving around the economic behaviour of the utility maximizing agents in the formation of bubbles from both side of views are presented either in the form of Austrian economics-style narrations or expressed in axioms. By and large, the rational bubbles which found it’s roots in the rational expectation school of thoughts postulates that agents in the market are rational in their tradingUniversity decisions leading up to a bubble while the speculative bubbles is the antithesis to that inference.

The behavioural finance is largely in agreement with the speculative bubbles school of thoughts on the non-rationality of agents in market. Instead of axioms and rationales on economic choices, the behavioural finance derived inferences on the behaviour of agents in the market from psychological observations and behavioural

104 surveys. Most studies from the school of thoughts on subjects related to stock market declines are anchored on the subject of cognitive biasness (various) of traders that lead to issues such as misjudgement, herding and panic etc. in the market.

Another aspect of the behavioural finance is on the study of market anomalies.

Primarily, studies on market anomalies investigate the market’s inefficiency empirically. Most studies in this area are concerned on the abnormal returns of the market which are least pertinent to the examination on stock market declines. Both the speculative bubbles and behavioural finance are considered heterodox in the financial economics for not conforming to the fundamental assumption of rationality of agent in the market.

Lastly, the more prominent econophysics branches that touch on the subject of stock market declines e.g. the network theory, ErgodicMalaya theory and multifractal theory are also classified under the first sectionof category. Relevant studies in these areas mainly examine the theoretical characterisation of stock market movements; prior, during or after stock market declines using approaches such as topology of interactions,

Lévy flight, long memory etc. In summation, literatures that are covered in the first category of framework are those that do not share a compatible platform or framework that allows for a viable empirical investigation under a unified theme i.e. the modelling of the changing regimes in stock market, specifically the periods of stock market declinesUniversity and otherwise. Literature classified in the second category of the framework are fundamentally based on empiricism and shared the common motivation of detecting forecastable trends of stock market declines from time series data. The schools of thought under this category generally employ the stock market indexes or their returns as the dependent variables, combining with various methodologies which may or may not include

105 exogenous variables to predict the trend of stock market declines (albeit what tantamount to a stock market ‘decline’ differs and is more specifically defined and modelled by the respective schools of thought).

The salient criteria for the classification of literature in the second category of the framework are empirical studies that examine the modelling / specifications of stock market declines and tests that examine the predictability of the phenomena. This second category of the framework thus includes schools of thought that based their empirical investigations on the cyclical theory (namely the bull and bear markets and the nexus between business cycle and stock market fluctuation), market fundamentals (i.e. stock market anomalies which are derived from the financial analyses and macroeconomic indicators) and the approaches developed by the interdisciplinary econophysicists based on the underlying theory of complex systems (i.e. variousMalaya models permutated from the base formulae of log periodic power law). of Over the last decade, one of the most dominant complex systems methodologies in financial economics is the application of log periodic power law (LPPL) formulae as used in the Johansen-Ledoit-Sornette (JLS) model to predict the building up of financial bubbles. The JLS model (and its variants) is able to capture the signature of an impending bubble-induced crash and estimate the risk of crash along the timeline with hazard rate (among the earliest literature, see Johansen, Sornette, & Ledoit, 1999; Sornette,University Johansen & Bouchaud, 1996; Sornette & Johansen, 1997; 1998). This section in essence filters the least related studies in the area of stock market crashes and bear market for further discussion. It also illustrates how the thesis builds up the foundation for the empirical framework which is introduced in the subsequent section.

106

2.5 Concluding Remarks

Over the years, studies on the stock market declines are mostly focused indirectly on the theoretical perspective of market agents’ behaviour in the creation of bubbles. These theories on bubbles are wide-ranging and often contrasting, which depend according to the promulgating schools of thought. Apart from the vast dissension in theories and paradigms, the empirical direction in this area is equally diverse too.

Although the research as reflected in the research objectives is principally empirical in nature, it has nonetheless taken a keen interest on the theoretical development of stock market declines. Since the summation to the literature review, particularly on the empirical facet is presented concisely in the section above, the section on the concluding remarks here designedly avoids repetition to the discussions above and emphasises more on some thought provokingMalaya discussions and theories that justifies the need of different perspectives ofand approaches in investigating the theme of stock market declines.

As a reiteration, the stock market essentially operates in a very complex system.

The complexity of the market is attributable to the volatility of stock transactions across the market which is subjected to a variant of trading strategies and styles of investment, different attitude towards risk, different time horizons in buying, holding and selling, contrasting interpretations on information and the erratic behaviours of traders in a high stakeUniversity game environment. The intricacy in which the whole system revolves has generated unabated interest for decades and lead to a constant evolution of theories.

Returning to the rudimentary un-institutionalised worldview, a stock market is perpetually filled with information of which investors based upon to make their investment decisions. Each new piece of information emanated from within the stock market such as the financial fundamental stocks or the volume of trading brings about 107 different reactions among different clusters of traders such as the “fundamentalist” and the “technical analyst” or “chartist”, and each turn of event external to the propagation mechanism of the market such as a change in the macroeconomics fundamentals i.e. rate of interest or a decline in the GDP growth would also trigger reactions on the stock market.

The stock market as illustrated by the simplified real world scenario as aforementioned manifests four distinguish standpoints for traders in the market, namely the trading of stock based on the intrinsic value of stocks as espoused by the fundamentalist, the charting of momentum on the movement of stock for the chartist, risk evaluation based on the volatility of stock movement which is assumed to be random by the “quant” and the more general macroeconomic assessment which is commonly adopted across to gauge the overall well-beingMalaya of the economy and which has enormous effect on the stock market (for further reading, see Hearth & Zaima,

2003). On the point of macroeconomics assessmentof for stock market, it is important to identify the “leaders of leading indicator” as stock market itself is a leading indicator to the economy. A stock market crash is also a common precursor to an impending recession or a financial crisis (McCown, 2007).

It is noteworthy that the proposition that the behaviour of stock traders are non- conforming and every new piece of information, which would elicit different responses fromUniversity investors is not without valid justification despite contradicting the conventionally accepted notion of efficient market hypothesis (EMH) which is discussed in detail at the latter part of this paper. According to Johansen (1997), should the EMH holds, trading of stock between two parties i.e. A and B would not occur since both of them would share the same information and expectations.

108

The primary concern of academic literatures on stock market over the course of history has been mostly on the short term predictability of stock return. The evolution of theories in the stock market began with the two competing stock paradigms in the

1930s, i.e. the intrinsic value of stock, which is the basis of the fundamental analysis and the Dow theory that is the foundation to technical analysis or charting, both analytic models which are still prevalent practices of modern day professional traders16 (for further reading, see Malkiel, 2007; Rhea, 1993). The focus gradually evolved to the notion that the movement of stock is a motion of chance which was the basis of the random walk theory (i.e. Kendall & Hill, 1953; Roberts, 1959; Samuelson, 1965,

1973b; Mandelbrot, 1966 etc.) and eventually leads to the inception of the EMH (Fama,

1970, 1991).

As a recap, Fama (1970) categorized the predictabilityMalaya of stock market return with the use of historical data as the “weak-form EMH”. The term “weak-form EMH” is renamed as “tests for return predictability” of

The establishment of the EMH nullified the analytic models of the fundamental analysis and the technical analysis from the academic viewpoint as the strong version of the efficient market hypothesis postulates that information in the market is complete and all investors are fully informed on every single piece of information available. Under such hypothetical assumptions, gain from stock speculation through means such as technicalUniversity analysis which chart the historical movement of stock prices for future prices prediction is not possible (Fama, 1965; Jones & Netter, 2008)

Apart from some well documented market anomalies which raise some reservation on the EMH, it is noteworthy that there were a few alternative views on the

16 The assertion that the earliest development of theories on the stock market is notwithstanding the earliest related work by Louis Bachelier ‘s 1900 dissertation entitled The Theory of Speculation (Bachelier 2006) which was not furthered developed until 1950s. 109 stock market that are theoretically sound but underdeveloped. According to Cooper

(2008), three of the most important theoretical frameworks opposing the EMH are the theories that drew parallelism of stock speculation to the “The Beauty Contest” which is contained in The General Theory of Employment Interest and Money, Keynes (1936), the theory that the stock market is a non-equilibrium system that is inherently unstable by Minky17 (1986, 1992), and the theory of fractal and scaling in finance by Mandelbrot

(pioneering literatures i.e. 1997a, 1997b, 1999a, 1999b, 2000, 2001a, 2001b, 2001c,

2001e, 2004).

Mandelbrot 18 , a distinguished mathematical physicist who ventured into the discipline of quantitative financial economics was one of the independent pioneers who developed the random walk theory which stipulates that the movement of stock has no memory (Mandelbrot, 1966, 1968). He then famouslyMalaya contradicted his own ground- breaking theory years later and through the econophysics 19 approaches provided evidence that the stock market indeed haveof long memory in the form of self-similarity pattern which he coined as “fractal” (Mandelbrot, 1999a). Mandelbrot’s works in the area of fractal and scaling in finance had laid a very important methodological foundation which has become an integral part of most contemporary econophysics researches, particularly in the application of the complex systems theory in financial economics. University

17 Minsky extended his work from his Stabilizing the Unstable Economy (1986) in The Financial Instability Hypothesis (1992) and reconciled some of his points with The General Theory of Employment Interest and Money by Keynes (1936). 18 Mandelbrot was recognised as one of the most influential and versatile intellectual in the past century. He founded the fractal geometry, the Mandelbrot set and was widely acknowledged to be one of the founding fathers of the chaos theory and the random walk theory. He contributed immensely in the area of mathematical physics and quantitative finance o which he subsequently merge both disciplines in the 1990s and counter-argued the foundation theories of financial economics from the econophysics perspectives (Clapham & Nicholson, 2009; Mandelbrot & Hudson, 2008). 19 Econophysics is broadly defined as a cross discipline that applies statistical physics methodologies which mostly based on the complex systems theory and the chaos theory for economics analysis (Mantegna & Stanley, 2000). 110

From a broader view, these underdeveloped theories originated from Keynes,

Minsky and Mandelbrot had to some extent served as the catalyst to the respective development of behavioural finance (pioneering literatures i.e.: Camerer, 1987; De

Bondt 1991; De Bondt & Thaler 1985; Kahneman & Tversky, 1979; Thaler, 1992), the cyclical or “bull and bear market” view on stock market (pioneering literatures i.e.:

Maheu & McCurdy, 2000; Pagan & Sossounov 2003; Edwards, Biscarri & Gracia,

2003) and the growing influences of the econophysics approaches in the research on the pre-cursory signatures of stock market crashes (pioneering literatures i.e.: Johansen &

Sornette, 1998a, 1998b, 1999a, 1999b, 1999c; Johansen, Sornette & Ledoit, 1999;

Sornette & Johansen, 1997, 1998, 2001; Sornette, Johansen & Bouchaud, 1996).

The AMH emergence marks the beginning of the reconciliatory pursuit in theoretical deliberation on the behaviour of asset markets.Malaya It is an important bridging theory that allows the conventional theories, particularly the EMH to coexist with other competing theories such as those from theof behavioural finance school of thought. The research which is also reconciliatory in essence is founded on the various theoretical viewpoints and empirical approaches as aforementioned (e.g. complex systems, cyclical, AMH, etc.).

More importantly, these alternative theories recognise the existence of bubbles in asset prices, the tendency of abrupt crashes in stock market and the cyclical run of bullUniversity and bear markets due to psychological factor more commonly known as herding which would not occur under the EMH. The crux of the EMH stipulates that the current price of a stock is an “equilibrium price” which reflects the balance of rationality, efficiency of information dissemination and the optimum decision of profit-maximising investors (Basu, 1977; Jansen, 1978; Laffont & Maskin, 1990).

111

These imply that stock market crashes in the history were assumed to be due to a certain piece of critical information that instantaneously caused a sharp adjustment of market prices and the persistent run of bull and bear markets are random coincidence

(see e.g. Abreu & Brunnermeier, 2003; Bakshi & Madan, 1998; Gonzalez, Powell, Shi

& Wilson, 2005; Johansen & Sornette, 1998b; 2002; Mandelbrot, 2004).

In reality, speculative bubbles in asset markets are well documented in history.

Literatures of bubbles from the tulip mania in the 16th century, south sea bubble in the

18th century all up to the recent dot-com bubble in 2001 and the subprime financial crisis in 2008 (Malkiel, 2007; Pele & Mazurencu-Marinescu, 2012; Vines, 2009) demonstrated the pervasiveness of market psychology in inducing speculative herding which results in the run of bull and bear markets and panic selling in stock market crashes (White, 1996). Malaya of

University

112

CHAPTER 3

METHODOLOGY

3.0 Introduction

The classification of literature featured in the penultimate section of Chapter 2 serves as the prelude to the following empirical framework section. The research then proceeds with the discussion on the selection of data and variables and next continues to highlight the innovation and the justifications on the use of the selected data and variables. Each of the variables that is used for predictive analysis is discussed amply in the section of variables descriptions, followed by the technicalMalaya sections on the mathematical / statistical notations of models for stock of market declines i.e. specifications for bull markets (and bull market) and specification of crashes and rebounds. Next, the research illustrates the methodologies for predictive analysis in two sections, i.e. the section of predictive regression model for the in-sample test and out-of-sample test used for parametric approaches and the section of probit model for both the sample tests for the semi-parametric, nonparametric and econophysics scale-invariant approaches.

3.1University Empirical Framework Figure 3.1 in the following shows the empirical framework of the study which narrows the research scope into a viable and compatible common ground that is shared by a selection of methodologies to achieve the objectives of the research. The framework is a continuation to the filtering of study scope from the classification of literature on stock market declines in Chapter 2 (see Figure 2.1).

113

The cyclical worldview on stock market crudely defines a run of stock market declines as bear markets. The switching of regimes (upturn run and downturn run) can be modelled with the parametric approach (i.e. the Markov-switching model and its variants) and the non-parametric approach (i.e. various variants of the B-B algorithm which are first used to date the business cycles). On top of the established methodologies, the research also devised a semi-parametric approach named as the naïve moving average negative return model. The comparison of models is discussed in the following chapters. Either the S&P 500 Index or the stock market returns derived from the index is used as the basic unit of measurement for the dependent variables for sample tests, depending on model types. The specific details are outlined in Section 3.4 and Section 3.5 in the following.

The research also compares the in-sample andMalaya out-of-sample output between the various econometric approaches with one of the Johansen-Lediot-Sornette (JLS) model variants which uses the base formulae of logof periodic power law (LPPL) to model stock market declines. In specific term, the stock market decline in the context of the JLS model is defined as stock market crash. As a reiteration, the definitions to stock market crashes are inconclusive. Therefore, it is important to build a common ground for results comparison between the conventional econometric approaches with the econophysic’s scale invariant approach. University

114

Bear Markets Definitions

Parametric Models:  Markov-switching (Filtered Probabilities)  Markov-switching (Smoothed Probabilities)

Semi-Parametric Models:  Markov-switching (Dichotomised Smoothed Probabilities)* Basic  Naïve Moving Average Dependent  Naïve Moving Average Negative Return* Variables:

 Standard & Non-Parametric Models: Poor's 500  Lunde & Timmermann’s B-B Algorithm (2004 Index / Variant) Return  Candelon, Piplak & Straetmans’ B-B Algorithm (2008 Variant)

Market CrashesMalaya Definition Econophysics Scale-invariant Models:  JLS Model of& JLS "Negative Bubble" Model (Integrated Identifications)

Independent Variables Sets: Predictability Analyses  Shiller’s Financial In-sample & Out-of-Sample Test: Variables  Predictive Regression (Test for Outputs from  Estrella & Parametric Models) Mishkin’s  Probit Model (Test for Outputs from Semi- Financial Parametric, Non-Parametric & Econophysics UniversityVariables Scale-invariant Models)

Figure 3.1: Empirical Framework

115

According to Cochrane (2008), the obvious linkages between the financial markets and the business cycles most of the times are exhibited through the causal relationship between the returns of financial assets such as stock with the various macroeconomic variables that represent the health of the real economy. Schwert (1989) who carried out one of the pioneering researches on the implications of the business cycle fluctuation to the stock market concluded that there is evidence of causality between the fluctuations of economic activities with the volatility of stock returns. Both the volatility of stock market and the changes in the industry output intensify when the economy is on a downturn.

Chauvet & Potter (2000) and Hartmann, Kempa & Pierdzioch (2008) also concluded that the underlying economic changes influence the movement of financial markets. Therefore, financial variables and macroeconomicMalaya indicators that are useful to predict the financial markets returns are also useful to forecast the business cycle and to serve as the general measurement to the robustnessof of activities in the economy. Chang

(2009); Chen (2009); Perez-Quiros & Timmermann (2000); Rapach, Wohar & Rangvid

(2005) in the same vein found evidence that macroeconomic fundamentals are useful to predict the future direction of the stock market.

McCown (2007) showed evidence that historically, prices of stocks have consistently dropped at the juncture when the business cycle is at the brink of entering intoUniversity the stage of downturn. The declines nonetheless were gradual and could take up to 18 months before the downward trend hit the bottom. This indicates that stock traders are inclined to relinquish their holding of shares slower than what is conventionally expected. One plausible explanation is the lack of knowledge of investors that the economy has reached a turning point until some months later due to the lag in the standard reporting of GDP. Table 1.1 shows the close nexus of stock prices and the

116 business cycle. Note that prices of stock decline generally after the economy has hit a peak. Such gradual but impactful descent in the stock market calls for more studies on the area of bear markets which seemingly received lesser attention compared to stock market crashes.

Table 3.1: Business Cycle and S&P 500 Index20

NBER S&P 500 Date of Number Local Percentage Business Cycle Index at Minimum of of Minimum Drop (%) Peak Business S&P 500 Months of S&P 500 Cycle Index from Index after Peak* Peak to Business S&P 500 Cycle Peak Minimum December, 92.06 June, 1970 6 72.72 -21.01 1969 November, 95.96 September, 10 63.54 -33.79 1973 1974 January, 1980 114.16 March, 1980 Malaya2 102.09 -10.57 July, 1981 130.92 July, 1982 12 107.09 -18.20 July 1980 356.15 October, 1990 3 304 -14.64 March, 2001 1160.33 September, of 18 815.28 -29.74 2002 December, 1468.36 March, 2009 15 683.38 -56.52 2007 *End of month

Table 3.1 underscores the validity of using market fundamentals to forecast the peaks and throngs of business cycles and to test the predictability of stock market. The following contents of the chapter explain in stages how the research uses various market fundamentalsUniversity to test the predictability of the proposed ex-post models for stock market declines with predictive regression and the probit model, which is introduced by Estrella

& Mishkin (1996; 1998) and Estrella (1998). Predictive analysis is carried out on the

20 Table 3.1 is Table above is updated from the paper by McCown (2007). Other sources include: Information on Recessions and Recoveries, the NBER Business Cycle Dating Committee, and related topics (n.d.), NBER; New York Stock and Commodity Exchange (n.d.). As of December 2014, no major stock market decline has occurred. The crash of 2007 as shown in the table is the the occurence last updated. 117 outputs of conventional economic models, namely the Markov-switching model and the

B-B algorithm. The predictability analysis is next repeated on the periods of stock market crashes as identified with the Johansen-Lediot-Sornette (JLS) model and the JLS

"negative bubble" model. The outcomes from these three models are compared alternately.

From a theoretical point of view, the linkage between the stock market and the fluctuation business cycle, measured with various proxies, is inherently an essential facet of the fundamental analysis for stocks selection. The analysis of stock is generally categorised into two types, the “macroanalysis” and the “microvaluation”.

Macroanalysis examine the causality between the movement of stock market, the aggregate economic activities as well as the general health of the economy. On the other hand, the more detailed valuation on the micro aspectMalaya of firms, specifically on the financial performance is termed as microvaluationof (Reilly & Brown, 2011). Macroanalysis approach is justified on the ground that stock market mirrors the outlook of the economy because the fundamental valuation of investment is based on the projected cash flows and the required rate of return, both of which are determined by the overall economic condition. Macroanalysis thus aims to identify which macroeconomic variables or leading indicators that are reliable to forecast the market movements. The analyses on the aggregate economy and the market present a critical understandingUniversity to the ensuing analysis on the specific industrial sectors and individual companies. The overall process of stock picking represents the top down fundamental analysis approach (Reilly & Brown, 2011). Essential market fundamentals in the form of macroeconomic variables, industrial indicators and financial ratios at every level of evaluation are mostly selected as test variables for the research.

118

3.2 Data and Variables

The empirical section of the research begins by identifying the periods of stock market declines within the observation period from the beginning of April 1967 to the end of

June 2014 (beginning of second quarter of 1967 to end of second quarter of 2014). The

Standard’s & Poor’s 500 Index (S&P 500 Index) is the composite stock price index selected for the research. The use of the S&P 500 Index follows the precedence of the past related literatures e.g. Boyd, Hu, & Jagannathan (2005), Campbell and Yogo

(2006), Chang (2009), Chen (2009), McMillan (2001), McQueen & Roley (1993),

Rapach, Wohar & Rangvid (2005), Shiller (2005) and Qi (2001). As opposed to the use of monthly average of stock return by e.g. Shiller (2005) and Qi (2001), the research uses stock return, computed as the monthly change in the closing index as used by e.g. Chang (2009) and Chen (2009). Malaya The S&P 500 Index along with the ofDow Jones Industrial Average index or DJIA (which evolved from the oldest index of Dow Jones Averages) are the most important benchmark indices of stock market in the U.S. The other prevalent index commonly used for studies is the NASDAQ Composite Index. The research chooses to use the

S&P 500 Index because it is composed of the most important stocks traded in the market, all of which totalled over seventy five percent of the market’s equity value. The derivation of the index is based on the market value of selected stocks compared to stockUniversity prices. In comparison, the DJIA only includes the stock prices of thirty listed companies and the NASDAQ Composite Index includes many peripheral small-capitals and high tech stocks (Standard & Poor’s, 2009). Therefore, the S&P 500 Index is deemed more suitable for the research.

Most of the selected macroeconomic variables selected for the research are considered as leading indicators to the economy. These test variables therefore possess

119 the potential to gauge the sentiment of the stock market at various stages. The National

Bureau of Economic Research (NBER) was the first to develop and utilise the index of leading indicators as an informal predicting means on the economy in the 1940s. The application of leading indicators to gauge the condition of the economy invited heavy criticism. This approach was once dubbed as “measurement without theory”.

Nonetheless, the leading indicators over the years are proven to be the most accurate tools for economic prediction (Auerbach, 1981).

The feasibility of using leading indicators for the examination of stock market predictability is strongly supported in the literature. Evidence of significant relationship between economic performance and leading indicators is well established in studies by

Bernanke and Kuttner, (2005), Schwert (1989) and Siegel (2008) among others. Baumohl (2008) concurred that leading indicators areMalaya important predictors for economic studies as they have shown to be reliable precursors to signal probable direction of the future economy. Different leading indicatorsof could produce different signals at various turning points of the economy. Resnick & Shoesmith (2002) for example showed that the shape of yield curve transforms correspondingly to the transition of economic condition.

Similarly, some of the other recent studies that showed evidence of relationship between the return of stocks and various macroeconomic and financial variables include Hong,University Torous & Valkanov (2007), Lettau & Ludvigson (2001); Menzly, Santos & Veronesi (2004), Rapach & Wohar (2005) etc. The examples of variables used for these researches are wide-ranging and include various macroeconomic variables such as term structure, interest rates, money supply, consumption–wealth ratio, inflation rate, unemployment rate as well as various industrial production ratios and numerous dividend-related ratios.

120

The significance of test variables for the research is tested using the methods borrowed from studies of the business cycle. One such test method for example is the probit model used earliest by Estrella and Mishkin (1998) in testing macroeconomic leading indicators for recession. Instead of investigating recession, the research emulates Chen (2009) in which the test model is applied for stock market declines. The use of market fundamentals in empirical prediction is not straightforward. Many issues and potential pitfalls have to be taken into consideration.

Heij, Dijk & Groenen (2008) precisely summed up the challenge in forecasting as follows:

“One of the basic questions in empirical forecasting is what information should

be included in the forecast model. For instance, in many macroeconomic and

financial applications, a large number of predictorMalaya variables are available. The forecaster then faces the challenge of employing the available information in the best possible way.”

As such, the research proposes the use of two sets of well-established financial variables commonly used for macroeconomic forecasting, particularly in the area of recession and financial crises. These variables sets are the Shiller’s Financial Variables and the Estrella & Mishkin’s Financial Variables (see for example Campbell & Shiller

(1988; 1989), Estrella & Mishkin (1998), Qi (2001) and Shiller (2005).

University

3.2.1 Shiller’s Financial Variables

The Shiller’s Financial Variables used for this research are made available at “Online

Data Robert Shiller”: http://www.econ.yale.edu/~shiller/data.htm. This data set was used by Shiller (2005) to investigate the causes that drive the mood of the market to

121 excessive levels - extreme optimism or otherwise. The data set is one of the two important sets of market fundamentals deemed most suitable for the analysis of financial market which are selected by the research as test variables. Most variables from the Shiller’s Financial Variables set are grouped under the Stock Market

Fundamentals category in the research.

The data set which represents the aggregated stock market fundamentals comprises of dividends (market aggregate), earnings (market aggregate), real dividends, real earnings, cyclically adjusted price earnings ratio (CAPE) and the consumer price index (CPI), (although the CPI is categorised under Financial Market Fundamentals in the research). Observation of the data set based on the description in the aforementioned website begins from January 1871. The selected fraction of the time series which is used for this research begins from the second quarter ofMalaya year 1967 to commensurate the availability of data for other test variables. of Further on the description of the Shiller’s Financial Variables set, the dividend and earnings are converted from the S&P quarterly data beginning from year 1926 through the linear interpolation method whereas the earlier dates are taken from the

“Common Stock Indexes, 2nd ed. (Bloomington, Ind.: Principia Press, 1939)” by

Cowles and associates. The original data in annual interval was interpolated into monthly interval. The Online Data Robert Shiller also provides data on monthly stock pricesUniversity derived from the mean of daily closing. Data used for this research however follows the calculation of market return as preceded in other studies in related area e.g.

Candelon, Pilak & Straetmans, Chen (2009), Coakley & Fuertes (2006), Maheu &

McCurdy (2000), Perez-Quiros & Timmermann (2000), Qi (2001) etc. The nominal terms of the Shiller’s data set are inflation-corrected using the CPI-U (Consumer Price

122

Index-All Urban Consumers) published by the U.S. Bureau of Labor Statistics since year 1913.

3.2.2 Estrella & Mishkin’s Financial Variables

Another set of variables used by the research follows the widely cited study by Estrella

& Mishkin (1998) who explored the efficacy of an extensive group of financial variables functioning as the leading indicators in predicting the U.S. recession. Qi

(2001) categorised the original financial variables of Estrella & Mishkin (1998) into five sub-categories, namely 1) Interest Rates and Spreads which consisted of 10-year

Treasury note minus 3-month Treasury bill yield curve rate, 6-month commercial paper rate minus 6-month Treasury note rate, 3-month Malaya Treasury bill, market yield, bond equivalent and 10-year Treasury note; 2) Monetary Aggregates i.e. monetary base (monthly averages of daily figures and seasonallyof adjusted), M1 (seasonally adjusted), M2 (seasonally adjusted), M3 (seasonally adjusted), monetary base adjusted for inflation (seasonally adjusted), M1 adjusted for inflation (seasonally adjusted), M2 adjusted for inflation (seasonally adjusted) and M3 adjusted for inflation (seasonally adjusted); 3) Stock Prices namely New York Stock Exchange Composite Index,

Standard and Poor’s 500 Index, monthly average and DJIA, monthly average at close;

4) Individual Macro Indicators include vendor performance, slower deliveries diffusion indexUniversity (seasonally adjusted), contracts and orders for plant and equipment (seasonally adjusted), new private housing permits (seasonally adjusted), MSCI, trade-weighted dollar vs. G-10 countries, change in manufacturers’ unfilled durable goods orders

(seasonally adjusted), Consumer price index, all urban consumers, all items (seasonally adjusted) and Growth in real GDP, lagged 1 quarter (seasonally adjusted); 5) Indexes of

Leading Indicators which comprised of Commerce Department composite index of 11

123 leading indicators, seasonally adjusted (CDLI), Stock and Watson (1989) Leading Index and Stock and Watson (1993) Leading Index.

3.2.3 Innovation to Data and Variables and Justifications

Variables introduced by Shiller (2005) and Estrella & Mishkin (1998) are re-categorised for the research to reflect more accurately on their representation within the context of financial economics in relation with the movement of the market. Some of the data recommended in the aforementioned set are excluded from the research due the discontinuation of publication. Other modifications and exclusions are explicitly justified in the following. The Shiller’s Financial Variables represent anMalaya important selection of aggregated financial fundamentals of the stock market. Variables such as dividends and earnings are most commonly used for fundamentalof analysis of individual stocks. The Shiller’s Financial Variables which reflect the aggregated fundamental values of stock market therefore are categorised by the research as the Stock Market Fundamentals.

The Estrella & Mishkin’s Financial Variables set on the other hand embodies a wider range of proxies that represent the macro-market fundamentals. For the research, variables of interest rates and spreads, money aggregates, CPI and growth in real GDP areUniversity grouped in the Financial Market Fundamentals category. Qi (2001) showed that this set of variables is highly effective in forecasting the turning points of the economy while a number of selected variables from the set were also examined by Chang (2009) and Chen (2009) in similar studies on the stock market.

Other macro-indicators from the set are split into two categories, namely 1)

Industrial Indicators which include the ISM manufacturing survey (inventories index),

124 contracts and orders for plant and equipment (seasonally adjusted), new private housing permits (seasonally adjusted) and the Change in manufacturers’ unfilled durable goods orders (seasonally adjusted); and 2) Market Sentiments, i.e. Purchasing Managers’ Index and University of Michigan Consumer Sentiment Index.

The original category of Indexes of Leading Indicators includes only the CDLI.

The Stock and Watson (1989) Leading Index and Stock and Watson (1993) Leading

Index are excluded from the research as these series retired after December 2013. The

Chicago Fed National Activity Index (CFNAI) is considered to be the most direct and appropriate replacement as the economic data used for the construction of the CFNAI is an extension to the original Stock and Watson Leading Index (Stock & Watson, 1999,

Federal Reserve Bank of Chicago, 2013).

Other data series, namely the monetary base,Malaya M3, trade-weighted dollar vs. G-10 countries, and the 6-month commercial paperof rate minus 6-month Treasury bill rate are removed as they were no longer updated at their respective sites. Stock indexes such as the New York Stock Exchange Composite Index, the Standard and Poor’s 500 Index and the DJIA are not used due to the reason that the dependent variable of the test model is the Standard and Poor’s 500 Index.

In compensation for the removal of the 6-month commercial paper rate minus 6- month Treasury bill rate, the research has replaced it with other types of interest rates andUniversity spreads i.e. the 5-year Treasury bond and the term spread of 5-year Treasury bond rate minus 3-month Treasury bill rate. The inclusion of these two variables follows the study by Chen (2009) in which the selection of test variables from the the Estrella &

Mishkin’s Financial Variables set was also arbitrary and with slight modifications to suit the needs of the study.

125

As a reiteration, one of the principal aims of the research is to determine the market fundamentals that are most significant and consistent in forecasting stock market declines, where the periods of declines are identified with ex-post modelling. Table 3.1 lists the groups, codes, informational lags, and descriptions for all the research variables. In a summation, a total of 25 financial fundamentals selected from the sets of the Shiller’s Financial Variables and the Estrella & Mishkin’s Financial Variables

(with slight adjustments) are used by the research as test variables. These test variables are regrouped into 5 categories. Data series that were discontinued are excluded. A number of interest rates and spreads are included to complement the original variable sets.

While some data of variables are obtainable instantaneously (i.e. stock prices and interest rates), numerous variables on the otherMalaya hand are only accessible later in a month or two. The GDP is published only after one quarter of lag. Estrella & Mishkin

(1998) and Qi (2001) tested the economy ofby applying the ex-ante forecasting approach where only the data that are readily available for the forecasting moment were selected for testing. It is common in most forecasting studies to employ data that contain informational lag due to publication delay (e.g. Estrella & Mishkin, 1998; Qi, 2001;

Chen, 2009; Chan, 2009). In similar fashion, the research also adopts the ex-ante forecasting approach where only data that are readily available for the forecasting moment were selected for testing. The informational lag of the research test variables areUniversity illustrated in Table 3.2. Detail of data sources is enclosed as Appendix A.

126

Table 3.2: Overview of Test Variables

Variable Lag Description Code (Month)

Stock Market Fundamentals DVD 1Q Dividends [Market Aggregate] EAR 1Q Earnings [Market Aggregate] RDVD 1Q Real Dividends [Market Aggregate] REAR 1Q Real Earnings [Market Aggregate] CAPE 1Q Cyclically Adjusted Price Earnings Ratio Financial Market Fundamentals TB3M 0 3-month Treasury Bill TN5Y 0 5-year Treasury Note TN10Y 0 10-year Treasury Note S5Y3M 0 5-year – 3-month Term SpreadsMalaya S10Y3M 0 10-year – 3-month Term Spreads M1 1 Money Supply M1of [Seasonally Adjusted] M2 1 Money Supply M2 [Seasonally Adjusted] RM1 1 Real Money Supply M1 [Seasonally Adjusted] RM2 1 Real Money Supply M2 [Seasonally Adjusted] CPI 0 Consumer Price Index RGDPG 1Q Growth in Real GDP [Lagged 1 Quarter]

University

127

Table 3.2, continued

Variable Lag Description Code (Month)

Industrial Indicators ISMI 1 ISM Manufacturing Survey (Inventories Index) COPE 1 Contracts and Orders for Plant and Equipment [Seasonally Adjusted] NPHP 1 New Private Housing Permits [Seasonally Adjusted] CMUO 1 Change in Manufacturers’ Unfilled Durable Goods Orders [Seasonally Adjusted] Market Sentiments PMI 1 Purchasing Managers’ Index MSCI 1 University of Michigan Consumer Sentiment Index Indexes of Leading Indicators CDLI 2 Commerce Department CompositeMalaya Index of 11 Leading Indicators [Seasonally Adjusted] CFNAI 1 Chicago Fed Nationalof Activity Index

3.3 Description of Test Variables

This section provides a concise glossary for the test variables and the corresponding sources for the retrieval of data (e.g. website links, statistic annuals etc.). Most of the descriptions and technical details in the following are drawn from the sources where the datasUniversity are obtained. Other definitions of layman’s terms are cited from specialised dictionaries of finance and economics. The in-depth review and justification of variable selections are already comprehensively discussed in Chapter 2 and the earlier sections.

128

3.3.1 Dividends [Market Aggregate]

The dividends (market aggregate) in this context is the dividend yield of the S&P index market. Dividend in layman’s term is described as distributions of profit to stock investors. Dividend yield is defined as the percentage derived by dividing the yearly dividend pay-out by the stocks’ closing price (Apostolou & Apostolou, 2004). The data used for this research is in monthly intervals, adjusted from the quarterly data of the

S&P. The dividends (market aggregate) data set together with other data in the following i.e. earnings (market aggregate), real dividends (market aggregate), real earnings (market aggregate), cyclically adjusted price earnings ratio, consumer price index as noted earlier are obtained from the “Online Data Robert Shiller” website, http://www.econ.yale.edu/ ~shiller/data.htm.

The relationship between dividends with the Malayamovement of stock market has been one of the most studied topics over the decades.of Studies in this area are still prevalent and up-to-date. Some of the latest studies on the significance of dividends in effecting the movements of stock market include Avramov (2002), Avramov & Chordia (2006),

Bernanke & Kuttner (2005), Bergeron (2013), Bozos, Nikolopoulos & Ramgandhi

(2011), Chen (2012), Chortareas & Noikokyris (2014), Dasilas & Leventis (2011),

Ferreira & Santa-Clara (2011), Hong, Torous & Valkanov (2002), Milonas, Travlos,

Xiao & Tan (2006), Kanas (2005), Karpavičius (2014), Kuo & Lee (2013), Nasseh & StraussUniversity (2004) and Park (2010). Hartmann, Kempa & Pierdzioch (2008) used the dividend yield as one of the macroeconomic predictor variables of stock returns in their model on economic and financial crises. Chang (2009) found that dividend yield is one of the most efficient test variables in forecasting conditional variance in the study of structural changes in stock price dynamics. Chauvet & Potter (2000) concurred and showed evidence that the

129 growth rate of the stock market’s dividend yield correlates significantly with the upswing and downswing of the stock market.

On the other hand, there are also disputes on the extent of dividends in dictating the direction of the stock market. Boudoukh, Michaely, Richardson & Roberts (2007),

Campbell and Yogo (2006), Park (2010), Robertson and Wright (2006), Valkanov

(2003), Wolf (2000) proposed that the selection of sample period played a very crucial role in forecasting the returns of stock with the use of the dividend-price ratio as the predictive variable, as its predictive power of dividends is not consistent over time.

3.3.2 Earnings [Market Aggregate]

The earnings (market aggregate) variable of the researchMalaya refers to the earning-price ratio of all companies that are listed in the S&P (Shiller, 2005). The study by Shiller (1984) is one of the most prominent studies that usedof both the market earnings and dividends as predictors to stock returns. Earning-price ratio is defined by Shim & Siegel (2001) as the market value per stock in relative to the gain accumulated for every unit of stock held.

In the context of capital structure, the earning-price ratio equals to the price of current share over earning per share (net earnings, minus the preferred dividends, dividedUniversity with the weighted average of ordinary shares in outstanding shares). In the practical terms, the price-earnings ratio implies the amount of dollar a market participant can pay for a company’s stock to gain a return of a dollar of the company’s earnings. The data of price-earnings ratio, similar to the data of dividend yield as discussed in the subsection prior is computed and maintained by Robert Shiller in the same website.

130

Similar to the nexus between dividends and the movement of stocks, another more important variable that is widely established to have significant influence to the sentiment of stock traders is the earnings of stocks (for more recent literatures, see e.g.

Alexander, Peterson & Wang (2014), Barnhart & Giannetti (2009), Bartholdy & Feng

(2013), Clement, Hales & Xue (2011), Cormier & Martinez (2006), Diavatopoulos,

Doran, Ferreira & Santa-Clara (2011), Fodor & Peterson (2012), Eilifsen, Knivsflå &

Sættem (2001), Jorgensen, Li & Sadka (2012), Kothari, Lewellen & Warner (2006) and

Shivakumar (2007)). Lewellen (2004) concluded that the earning-price ratio is one of the better financial ratios in forecasting the aggregated stock market return along with book-to-market ratio for recent sample (beginning from year 1963 to recent years) whereas the predictive power of dividend yield is consistently reliable from the early years (since year 1946 to recent years) and works wellMalaya for different subsamples. of 3.3.3 Real Dividends [Market Aggregate]

Real dividends (market aggregate) is simply the variable of dividends (market aggregate) transformed into real term i.e. the aggregated dividend yield of the S&P 500

Index adjusted for inflation.

3.3.4University Real Earnings [Market Aggregate] As per above, the real earnings (market aggregate) is simply the variable of earnings

(market aggregate) transformed into real term i.e. the aggregated earnings per share of corporate listed in the S&P 500 adjusted for inflation.

131

3.3.5 Cyclically Adjusted Price Earnings Ratio

The cyclically adjusted price-earnings ratio (CAPE), which is also known as the Shiller

PE Ratio is calculated by adjusting the annually aggregated earnings of the S&P market for the past 10 years with the rate of inflation. The result is then averaged out and divided with the aggregated price of the market (Shiller, 2005). According to Faber

(2012), the idea of filtering out cyclical components in the analysis of stocks by smoothing the earnings across numerous years was originally proposed by Graham &

Dodd (2008). The early seminal work on value investing, entitled the Security Analysis, which was first published in year 1934. Tobago (2011) nonetheless credited the origin of the cyclically adjusted PE ratio to the seminal work by Shiller (2005) and noted that

PE ratio alone is much less reliable if computed with volatile earnings without filtering out the fluctuation of the business cycle. Malaya of 3.3.6 3-month Treasury Bill

Treasuries are debt instruments “backed by full faith and credit by the U.S. government”. Treasuries can be classified into 4 types, namely Treasury bills, Treasury notes, Treasury bonds, and Treasury Inflation Protected Securities (TIPS), subjected to their maturities and denominations (Shim & Siegel, 2001). Treasury bills (T-bills) are short-term debts of the U.S. government that mature in not more than one year. They are easyUniversity to liquidate in the secondary market and are considered as minimal-risk investment instruments since they are free of default risk. Interest earned on T-bills is exempted from taxes, dissimilar to short-term corporate bonds (Federal Reserve Bank of New

York, n.d.).

132

The 3-month Treasury bill rate is most commonly used to predict the performance of stock market or stock returns, e.g. Avramov (2002), Avramov &

Chordia, Dopke, Hartmann & Pierdzioch (2008), Flannery & Protopapadakis (2002),

(2006), and Rapach, Wohar & Rangvid (2005). The market fundamental is also widely used for studies that examine and rank the predictive power of test variables in regime change through the in-sample and out-of sample analyses, e.g. Chang (2009), Chen

(2009), Gupta & Modise (2012) and Qi (2001). The predictive efficacy of the the 3- month Treasury bill rate is not merely confined to financial market. Keilis-Borok,

Soloviev, Allègre, Sobolevskii & Intriligator (2005) for example employed the 3-month

Treasury bill rate to predict the rate of unemployment, which represents as an indicator of the general health of the economy.

Malaya 3.3.7 5-year Treasury Note of As a reiteration, Treasury bills are short term government debt instruments with maturity of less than a year. According to the definitions by Kenny (n.d.), Treasury notes are sovereign debt issued with maturities of less than 10 years (i.e. 1-year, 3-year,

5-year, 7-year and 10-year) whereas Treasury bonds which is also synonymous to “long bond” are issued with maturities of up to 30 years (i.e. 20-year and 30-year). In layman’s terms, the main distinction between Treasury notes and Treasury bonds is merelyUniversity the maturity duration.

The 5-year Treasury note therefore is a sovereign debt issued by the U.S. government that has a five year maturity period. Since the US government would not default on its debt, Treasury notes are deemed to be the safest investment and therefore provide the least yield. The 5-year Treasury note was used by Chen (2009) to derive the

5-year Treasury note ‘minus’ the 3-month Treasury bill term spread. Data of the 5-year 133

Treasury note is used as a standalone test variable and also to derive the 5-year Treasury note ‘minus’ 3-month Treasury term spread. Such dual usage of data is similar to the study by Qi (2001) which used the 10-year Treasury note for individual test variable as well as to compute the 10-year Treasury note ‘minus’ 3-month Treasury bill term spread to forecast recessions.

3.3.8 10-year Treasury Note

The 10-year Treasury note is similar to the 5-year Treasury as elaborated above. It is a long-term debt of the US government issued through the US Treasury with a ten year maturity and sold at face value at the inception with fixed rate of annuity paid throughout the maturity period (Shim & Siegel, 2001).Malaya The rate the 10-year Treasury note generally represents is the benchmark rate for most of other interest rates. It is also observed by the Federal Reserve as a guideof in setting the Fed funds rate in the auction market. The rate of the 10-year Treasury note thus reflects the sentiment of investors towards the outlook of the economy and would usually fall during a recession.

(Amadeo, n.d. (a)). The 10-year Treasury note is commonly used for economic forecasting e.g. see Keilis-Borok, Soloviev, Allègre, Sobolevskii & Intriligator (2005),

Nyberg (2011) and Qi (2001) etc. University 3.3.9 5-year – 3-month Term Spreads

The spreads between various interest rates of different terms of maturity are well- established precursors to the turning points of the economy and financial market

(Bhaduri & Saraogi, 2010; Resnick & Shoesmith, 2002). Malik, Ewing, Kruse & Lynch

(2009) and McCown (2007) underscored that a conclusive explanation to the

134 conundrum of relationship between the changes in spreads and the real economic activities is yet to be found.

Of all the possibilities proposed in relation between term spreads and the macroeconomic movement, Bhaduri & Saraogi (2010) proposed a plausible explanation that due to the fact that the expectation of the market for future short term rates is encapsulated in the information of the interest rate of longer-term bond, the general movement of short term interest rates therefore mimics the long-term interest rate on the assumption that the expectation on interest rate is systematically undistorted. In most times, economic conditions would dictate the rate of shorter term interest rates i.e. during economic downturn, a short term interest rate would be lower because the rate of inflation is also lower when the economy is in recession. Moreover, it is common for the Federal Reserve to suppress the interest rates duringMalaya an economic downturn as a measure of expansionary monetary policy to spur the economy. Therefore, a levelled yield curve or a smaller yield spread is cursoryof to an impending decline of the economy.

Vice versa, a larger yield spread could be an indication of a robust economy in the future.

The term spread of 10-year Treasury note rate ‘minus’ the 3-month Treasury bill rate is one of the most common predictors for the forecasting of business cycle and financial market. Qi (2001) used the spread for recession forecasting while Chen (2009) employedUniversity it as the test variables for stock market prediction. Since both the 5-year Treasury note and the 3-month Treasury bill are selected as the test variables for the research, it is convenient to derive from their spread. The narrower spread is thus included as one of the research’s variables, along with the broader term spread for the testing of predictive capability.

135

3.3.10 10-year – 3-month Term Spreads

The rates difference between the 10-year Treasury note and 3-month Treasury bill is constantly monitored closely in the financial market as a yield curve inversion in most instances in history were shown to be the harbinger to an economic downturn. A yield curve inversion is described as an interest rate scenario where a longer term

Treasury note has a lower yield compared to a shorter term Treasury note / bill (see

McCown 2007). Shoesmith (2002) showed evidence that the broad term spread is one of the most significant market fundamentals to help time the stock market. By employing the spread of 10-year Treasury bond rate ‘minus’ the 3-month Treasury bill rate, it was found that impending bear markets could be predicted with remarkable success rate as such that the market timing strategy was able to increase annual compounded return of over 2% compared to “a stock-only-buy-and-hold strategy”.Malaya The term spread was also provenof to be a significant predictor by Hartmann, Kempa & Pierdzioch (2008) and Estrella & Trubin (2006) in their studies on the efficacy of macroeconomic variables in forecasting economic and financial crisis. Issler

& Vahid (2006) on the same wavelength recommended the 10-year minus 3-month term spread as one of the most reliable leading indicators for recession forecasting while

Avramov (2002) successfully used the term spread to forecast financial crises. Avramov

& Chordia (2006), Bernanke & Kuttner (2005) and Flannery & Protopapadakis (2002) alsoUniversity found that the term spread has significant relationship with return of stocks.

3.3.11 Money Supply M1 [Seasonally Adjusted]

The M1 money supply (seasonally adjusted) is the basic and most liquid categorisation of monetary measurement. M1 comprised of currency that could be effortlessly traded

136 for services or goods, i.e. publicly held banknotes and coins, traveller’s cheque, various chequeable deposits such as Negotiable Orders of Withdrawal accounts (NOW),

Automatic Transfer Service accounts (ATS) and demand deposits (checking accounts)

(Shim & Siegel 2001; Wessels, 2012). The data series retrieved for the research is termed in billions of USD$.

Rapach, Wohar & Rangvid (2005) used the narrow money (M1) and broad money (M2) amongst other market fundamentals to test the predictability of international stock market. Some of the recent researches that investigate the relationship between M1 and the stock market include Chen (2009) and Hartmann,

Kempa & Pierdzioch (2008). Other researches could mostly be found in the earlier years. Nonetheless, Laopodis (2013) gave a word of caution that empirical results in this area were by large mixed and contradicting, indicatingMalaya that there is no single and consistent framework in explaining the relationship between the stock market movement and monetary policy. of

Laopodis (2013) stipulated that on one perspective, an increase in money supply would result in an increase in the prices of stock which consequently would spur the performance of the stock market and the general economic activities. Due to the fact that expected dividends and rate of interest predominantly influence the prices of stock, any unexpected adjustment on the monetary policy would have an impact on the stock marketUniversity either explicitly though the interest rate mechanism or indirectly via the factors of dividend or equity premium. An alternative view proposed that a counter cyclical monetary policy would lead to the lowering of the assets’ expected return and consequently impacts the stock market. This is because the ascending of stock prices is deemed to be an indication of potential future inflation which could prompt a reaction from the Federal Reserve to offset it.

137

3.3.12 Money Supply M2 [Seasonally Adjusted]

M2 (seasonally adjusted) is equals to M1 plus assets that cannot be used for direct payment transactions but can be conveniently exchanged for the purchase of consumptions i.e. household savings deposits, time deposits (in small-denomination), repurchase agreements (overnight), money market mutual fund shares and Eurodollars

(overnight) (Shim & Siegel 2001). The data series retrieved for the research is denoted in billions of USD$.

Similar to the discussion in section 3.3.11, a concrete conclusion on the relationship between money supply and the fluctuation of stock market is still yet to be established from empirical studies. As such, a number of studies on recession and stock market in recent years only include this variable as one of the test variables to be parallel with the employment of variables of EstrellaMalaya and Mishkin (1998) without establishing strong evidence on its predictiveof efficacy (see for example Chen 2009 and Qi 2001). Other examples of literature that specifically employed the M2 in the prediction of economy or financial market are Bouwman & Jacobs (2011), Flannery &

Protopapadakis (2002), Heij, van Dijk & Groenen (2011a), Kaminsky, Lizaondo &

Reinhart (1998), Rapach, Wohar & Rangvid (2005), Seip & McNown (2007) etc.

3.3.13University Real Money Supply M1 [Seasonally Adjusted] The real M1 (seasonally adjusted) simply equals to the M1 (seasonally adjusted) adjusted for inflation in 1982-1983 dollars. The series is termed in billions of USD$.

138

3.3.14 Real Money Supply M2 [Seasonally Adjusted]

As per above, the Real M2 (seasonally adjusted) simply equals to the M2 (seasonally adjusted) adjusted for inflation in 1982-1983 dollars. The series is termed in billions of

USD$.

3.3.15 Consumer Price Index

In reference to the Bureau of Labor Statistics (2009), the consumer price index (CPI) is defined in verbatim as “a measure of the average change over time in the prices of consumer items - goods and services that people buy for day-to-day living”. The CPI which is kept by the Bureau of Labor Statistics can be accessed through its official website at http://www.bls.gov/. The CPI of the USMalaya is a complex measurement that integrates advanced survey and sampling techniques with various statistical methods all of which are based on the foundation of soundof macroeconomic theory. The output thus is an apt and accurate calculation of the changes on the average price for the consumption sector of the U.S. economy.

Based on the Bureau of Labor Statistics (2009), there are three types of CPI that are computed and maintained by the department, namely the CPI-W and CPI-U and C-

CPI-U. The “W” in the CPI-W represents the urban wage earner, but more specifically, thisUniversity category comprises of “clerical workers, sales workers, craft workers, operative, service workers, or laborers”. The CPI-U has a wider measure (i.e. the “U” stands for all urban employees) which covers both the categories of workers under the CPI-W as well as the full-time white collars and professionals, the short-term workers, the section of self-run business owners and the population that is out of employment (i.e. the labor force which is temporary unemployed, the unemployable and the retirees).

139

The C-CPI-U which stands for Chained Consumer Price Index for All Urban

Consumers is a relatively new index which was first published in year 2002. The difference between C-CPI-U and both the CPI-U and the CPI-W is the method and weightage in formulating the different indices. The C-CPI-U takes into consideration the households’ ability to attain the equivalent standard of living with the consumption of substitute goods and services compared to otherwise.

Due to the computational complexity of the C-CPI-U which includes data on consumptions that are not accessible instantaneously, the data series requires two subsequent adjustments on schedule after its initial publication in the preceding form.

The CPI-U is nonetheless chosen as one of the research’s test variable because it has a broader measurement, which better reflects the economic environment. Moreover, the CPI-U is one of the variables contained in the Shiller’sMalaya Financial Variables set which the research proposes to test in entirety. of Studies have shown that the fluctuation of inflation has huge implications on the market. Campbell & Vuolteenaho (2004) for example, found that the level of inflation explains about 80% of the residual mispricing of the stock market due to ‘inflation illusion’. Similar evidence is also found in studies by Brandt & Wang (2003), Cohen,

Polk, & Vuolteenaho (2005) and Lee (2010). Oxman (2012) contrarily argued that the level of inflation does not significantly affect the prices of stock because the used of CPIUniversity misrepresents the relationship between price inflation and dividend yield. Baumohl (2008) agrees with the general findings and suggested that the market fundamental is constantly placed on the radar of analysts of financial markets due to the significance of its predictive power. He further added that from the financial market perspective, inflation commonly leads to a higher cost of borrowing for businesses due to the increase in bond rates. The inflation risk could prompt the raise of interest rates

140 by the Federal Reserve and prompts investors to re-evaluate their investment strategies.

At the global level, Cai, Chou & Li (2009) found evidence that the cyclical oscillation of inflation rate, combined with the volatility of stocks have a potent adverse impact on the international stock indices. On the economic aspect, the CPI which is much considered as a macroeconomic variable is also widely used to forecast the economy.

Among more recent studies in this area with the CPI include Hartmann, Flannery &

Protopapadakis (2002), Hong, Torous & Valkanov (2002), Kempa & Pierdzioch (2008),

Rapach, Wohar & Rangvid (2005), Ritter & Warr (2002), Seip & McNown (2007) etc.

3.3.16 Growth in Real GDP [Lagged 1 Quarter] Gross domestic product (GDP) is defined as the aggregatedMalaya output of final goods and services measured in market value that are produced by labour and property within the boundary of a country. GDP is comprisedof of purchases by government and consumer, the differences (net) between exports and imports in goods and services, and private domestic investment. Prior to the year 1991, the Gross National Product (GNP) (which includes the value of goods and services produced by U.S. citizens outside of the U.S. and excludes producers of non-U.S. citizen within the U.S.) was the main economic measure for the U.S. before it was replaced by the GDP to conform to international standards (Downes & Goodman, 2010).

UniversityThe real GDP for the U.S. is the constant dollar GDP adjusted based on year

2005 dollar. It is used by the Bureau of Economic Analysis (BEA) to compute the growth rate of the U.S. The real GDP computed in the form of growth rate is a more accurate measure to account for the changes in the national output because it eliminates the distorting impact of inflation. The GDP growth rate is a crucial macroeconomic variable as it illustrates how quickly the economy is expanding (or otherwise). It is 141 derived by simply comparing a selected GDP figure to the previous quarter or year. The growth rate of GDP is determined by four fundamental components which are part of the GDP computation. Personal consumption driven mostly by retail sales is the most significant determinant to the U.S. GDP growth. This follows by business investment, spending of the government and exports (Amadeo, n.d. (b); Wessels, 2012).

It is common that the GDP growth rate is annualised for the convenience of the year-to-year comparison although the figure is made available by BEA on quarterly intervals. Thus, the GDP report published by the BEA at any particular quarter is in actuality the figure of GDP for the year with the seasonal effects removed. Such measure is necessary as otherwise, the GDP figure and its growth rate for each fourth quarter would see a steep upsurge due to the surge in the household consumption which swells the retail sales during the festive seasons at theMalaya end of the year. The complexity of accounting and computing the GDP causes the frequent revision to the GDP growth rate within a month of release (Bureau of Economicof Analysis, 2006).

Baumohl (2008) underscored that the GDP report and its revision could have a huge effect on the stock market and the bond market (to a lesser extent) as GDP is perceived to be the most important economic indicator to most of the market participants and researchers. Although a high GDP reading is favourable as it indicates a robust economy (which could point to higher corporate earnings), it could also present an Universityuncomfortable dilemma if the growth rate is climbing too steeply. That could be an indication of an economic overheating. Nevertheless, a sluggish economic growth is obviously unfavourable.

Results of the predictive significance of the variable are mixed when tested with the economy and the stock market. Qi (2001) found that the GDP growth rate is among the top most significant macroeconomic indicators in predicting an economic recession.

142

However, Flannery & Protopapadakis (2002) in their study on the nexus between macroeconomic indicators and aggregated stock returns through the GARCH model found that the GNP/GDP are among the variables that have the least effect on stock returns.

3.3.17 ISM Manufacturing Survey (Inventories Index)

The inventories index of the Institute for Supply Management (ISM) is one of the weighted components used to construct the Purchasing Managers’ Index or the PMI.

Other components include new orders, production, employment and supplier deliveries.

The ISM is formerly known as the National Association of Purchasing Managers (NAPM). NAPM started conducting surveys for theMalaya manufacturing industry in the US since 1931 and continued under ISM since 2002 (Baumohl, 2008).

Results from survey are computedof into diffusion indices to gauge the level of activities in the economy. The survey is designed as such that respondents could compare the perceived performance of various activities between the present months with the prior months. Answers are restricted to “better, worse, or unchanged”. The diffusion index derived from the survey that ranged from 0% to 100% is computed by totalling up the “better” percentage and one-half of the “unchanged” percentage. Further on, every component index is seasonally adjusted separately and weighted accordingly to University produce the Purchasing Managers’ Index (PMI) (Lahiri & Monokroussos, 2013;

Peláez, 2003b). Descriptions on the PMI are featured in the following.

The research uses the ISM’s inventories index in replacement of the vendor performance, slower deliveries diffusion index because the publication schedule of later series was found to be not reliable. As of January 2015, the vendor performance slower

143 deliveries diffusion index was only updated to January 2014, which is a full one-year lag. Recognising the empirical significance of the index in previous studies, the research decided to replace it with the next most similar index.

For the record, the Vendor Performance Index is one of the components of the

NAPM index. The NAPM is derived from a survey, which covers more than 250 corporations from 21 industries in the U.S. The index allows market participants to gain insight on the manufacturing and production activities over the phase of the past 30 days

(NAPM, n.d.) that could be pivotal in their trading decisions. The series is somewhat discontinued by the ISM but the data is reconstructed and published irregularly with much delay. The data series is made available in Datastream.

The vendor performance, slower deliveries diffusion index measures the relative swiftness of order distribution in the industrial sector.Malaya An increase in delivery speed is an indication to the growth in demand of for production supplies. Thus, the vendor performance, slower deliveries diffusion index is widely acknowledged as one of the leading indicators to the business cycle (Pring, 2006). Some examples of recent studies that utilised the vendor performance, slower deliveries diffusion index as predictors for economic or financial forecasting include those by Bouwman & Jacobs (2011), Drobetz,

Kaiser & Zimbehl (2012), Heij, C., van Dijk & Groenen (2011a), Peláez (2003a), Stock

& Watson (2002).

UniversityThe relationship between the inventories index and the real economy output is briefly explained by Koenig (2002). The use of inventories indices for the forecasting of

GDP growth is also examined by Peláez (2003) but the results were found to be negative. Lahiri & Monokroussos (2013) however showed evidence that the ISM inventories index is a reliable variable for economic forecasting. Cho & Ogwang (2006)

144 on the other hand focused at the industrial level and showed evidence that inventories indices is significant in forecasting the outlook of the supply chain industry.

3.3.18 Contracts and Orders for Plant and Equipment [Seasonally Adjusted]

The contracts and orders for plant and equipment is an industrial level economic leading indicator. The data series (denoted in USD$ millions) is published by the Conference

Board each month (Downes & Goodman, 2010). The glossary by the United States

Bureau of the Census (1968) described the series as follow:

“This series measure the dollar value of new contract awards to building and

public works and utilities contractors and of new orders received by manufacturers in heavy machinery and equipmentMalaya industries. It is the sum of: 1) Value of commercial and industrial contracts; 2) Value of privately owned

public works and utilities contracts;of and 3) Value of manufacturers’, new orders,

machinery and equipment industries.

Data on value of commercial and industrial construction contracts measure the

value of contracts for work about to get underway on commercial building

(banks, offices and lofts, stores, warehouses, garages, service stations), and

manufacturing buildings (e.g. processing, mechanical). Since January 1956, Universitytheaters have been excluded and some non-industrial warehouses have been excluded.

The value of privately owned public works and utilities contracts component

measures the value of public works, and utilities contracts awarded by private

individuals and agencies. It includes contracts for the following types of

construction: 1) Public works - streets and highways, bridges, dams and

145

reservoirs, waterfront developments, sewerage systems, parks, playgrounds etc.:

and 2) Public utilities - electric light and power, gas plants and mains, pipe lines

(oil and gas wells), water supply systems, railroad construction, airports

(excluding buildings), etc.

Value of manufacturers’ new orders, machinery and equipment industries,

measures the values of new order received by a subgroup of durable goods

manufacturers, specifically manufacturers in the following categories: 1) Non-

electrical machinery – including steam engines and turbines; internal

combustion engines; construction, mining and material handling equipment;

metalworking machinery; special industry equipment; general industry

equipment; office and store machines; service industry machinery; and miscellaneous non-electrical equipment (farmMalaya machinery and equipment and machine shops are excluded); 2) Electrical machinery – including electrical

transmission and distribution equipment,of electrical industrial apparatus, and

other electrical machinery (household appliances, communication equipment,

and electronic components are excluded); and 3) Shipbuilding and railroad

equipment.”

Klein & Iammartino (2009) pointed out that an increase in new contracts and orders for plant and equipment implies a favourable economic outlook and potential robustUniversity growth in the future. Most of the more recent studies however did not investigate the specific relationship between the variable and the economy or the financial market.

The contracts and orders for plant and equipment was more commonly used in univariate model as test variable to establish its individual forecasting accuracy; see for example, Heij, van Dijk & Groenen (2011a; 2011b) & Qi (2001).

146

3.3.19 New Private Housing Permits [Seasonally Adjusted]

The variable of new private housing permits (seasonally adjusted) for this research is retrieved from the U.S. Bureau of the Census. The site which publishes statistics of housing units authorised by building permits in the U.S. is updated on monthly intervals. The official definition in verbatim is as follows (United State Census Bureau, n.d. [a]):

“Building permits represent the number of new privately-owned housing units

authorized by building permits in the United States. A housing unit, as defined

for purposes of this report, is a house, an apartment, a group of rooms or a single

room intended for occupancy as separate living quarters. Separate living quarters

are those in which the occupants live separately from any other individuals in

the building and which have a direct access Malayafrom the outside of the building or through a common hall. In accordanceof with this definition, each apartment unit in an apartment building is counted as one housing unit. Housing units, as

distinguished from “HUD-code” manufactured (mobile) homes, include

conventional “site-built” units, prefabricated, panelized, componentized,

sectional, and modular units. Housing unit statistics in these tables exclude

group quarters (such as dormitories and rooming houses), transient

accommodations (such as transient hotels, motels, and tourist courts), "HUD- Universitycode" manufactured (mobile) homes, moved or relocated units, and housing units created in an existing residential or non-residential structure.

These numbers provide a general indication of the amount of new housing stock

that may have been added to the housing inventory. Since not all permits

become actual housing starts and starts lag the permit stage of construction,

these numbers do not represent total new construction, but should provide a

147

general indicator on construction activity and the local real estate market. The

value of new private housing units is the sum of the estimated valuation of

construction on each building permit authorized in that year by local permit-

issuing jurisdictions.”

The new private housing permits data is commonly used as test variables for macroeconomic studies and its reliability as a predictor for the economy was underscored by Estrella & Mishkin (1998). Some of the more recent literatures that investigated the relationship between the economic well-being and the new private housing permits include Bouwman & Jacobs (2011), Issler & Vahid (2006), Stock &

Watson (2002) and Qi (2001). Case & Shiller (2003) and Shiller (2008) specifically employed the variable to investigate bubbles in the market.

Malaya

3.3.20 Change in Manufacturers’ Unfilledof Durable Goods Orders [Seasonally Adjusted]

The manufacturers’ unfilled durable goods orders is defined as purchase transactions that are yet to be accounted in the sales account. More specifically, the figure of the unfilled orders at the period of closing account is the sum of the figure brought forward from the opening period, add up with the current orders newly acquired and less the net sales amount. Changes in the figure of the manufacturers’ unfilled durable goods orders reflectUniversity the monthly fluctuation in the quantity of “orders variants of durable goods manufacturers” termed in dollar. In other words, it shows the total variation for orders backlogs in the closing period of the present month and the end period of the prior month (United States Bureau of the Census, 1968).

Unfilled orders generally rise when the economy is expanding and decline during an economic downturn. Baumohl (2008) explained that a huge build-up of

148 durable goods orders is an indication that the industry is probably robust in the future because orders are piling up too quickly to be fulfilled on time. Backlogs in production nonetheless could be a cause for concern to the economy as the supply of resources could be stressed. This could consequently trigger inflationary pressures due to the deferments in fulfilling the demand of the market. Conversely, a drop in unfilled orders could jeopardise the economy as a reduction in production would adversely impact the job market in the future.

The manufacturers’ unfilled durable goods orders is most often used as test variable for economic forecasting (Qi, 2001; Zarnowitz, 1992). It is considered as a leading indicator to the economy and some studies combined it with other macroeconomic indicators for the construction of recession indices. Issler & Vahid (2006) for example used the series as one of the componentsMalaya of the NBER recession signal to construct an index to gauge the general direction of the U.S. business cycle and the index compared favourably against otherof existing indices.

3.3.21 Purchasing Managers’ Index

The data of the Purchasing Managers’ Index is maintained since year 1948 through the initiative of the Institute of Supply Management (ISM). As a reiteration, the ISM is previously known as the National Association of Purchasing Managers or NAPM before yearUniversity 2002 (Baumohl, 2008).

Kettel (2001) drew a parallelism that the PMI is akin to a market’s barometer that gauge the sentiment of producers. The index is constructed based on a survey called the “Report on Business” which is sampled on the purchasing executives who are employed in over three hundred firms from the services and manufacturing industries.

149

The survey is designed to collect insights and opinions on the industries to ascertain whether the business environment (represented in various indicators such as

“employment, new orders, business expectations, output” etc.) has changed for the better or for worse in comparison to the previous month. Therefore, questions listed in the survey are very objective in the form of adjectives i.e. “better, same, worse, higher, lower, faster, slower” etc. More important indicators in terms of their impact on the economic activities are given higher weight in their computation.

Results from the survey which are computed separately in two segments are then used to derive a diffusion index which has a range of 0 to 100 percent. A reading of over 50 indicates that the business environment is healthy whereas a figure under 50 shows that the economy is contracting. A higher reading in the index above the 50 percentage level represents a larger expansion, vice Malayaversa, a bigger difference below the middle level denotes a larger contraction (Yamarone,of 2007). Boumohl (2008) highlighted that the PMI is one of the more important economic indicators to the stock market as it serves as the precursor to the performance of corporates in terms of earnings. The stock market is generally more bullish in the anticipation of higher corporate earnings. Conversely, the level of confidence would deflate if corporate earnings are predicted to drop. Qi (2001) replicated the study by

Estrella & Mishkin (1998) in the forecasting economic recession which used the PMI as oneUniversity of the test variables. In the study, it was found that the PMI is relatively significant (ranked at the top half of the table) compared to other test variables, particularly for the longer forecasting horizons.

150

3.3.22 University of Michigan Consumer Sentiment Index

In reference to the official website of the University of Michigan (n.d.), the Surveys of

Consumers was conceived by George Katano in year 1946. The conception of the survey was founded on the premise of sensing the pulse of the national economy based on the fundamental behaviour of consumers, which are their choices on saving and spending. The University of Michigan Consumer Sentiment Index is one of the indices derived from the Surveys of Consumers which is carried out by the Survey Research

Center and overseen by Richard T. Curtin at the University of Michigan.

Results of the Surveys of Consumers are highly regarded and acknowledged to be reliable precursors to the future economic direction of the U.S. The execution of the surveys also meets the highest of standards required for accuracy, reliability and validity. As such, The University of Michigan ConsumerMalaya Sentiment Index is used as one of the indicators in the Leading Indicatorof Composite Index published by the U.S. Department of Commerce, Bureau of Economic Analysis (Baumohl, 2008). A snippet from the website of the University of Michigan (n.d.) describes the following:

“The inclusion of data from the Surveys of Consumers by the Commerce

Department is a significant confirmation of its capabilities for understanding and

forecasting changes in the national economy. Each series included in the

composite index of leading indicators is selected because of its performance on Universitysix important characteristics: economic significance, statistical adequacy, consistency of timing at business cycle peaks and troughs, conformity to

business expansions and contractions, smoothness, and prompt availability.”

The core inquiries of the University of Michigan Consumer Sentiment Index encompass three segments, namely 1) what is the perception of the consumers for their personal financial wellbeing?; 2) what is the perception of the consumers for the overall 151 economic condition in the foreseeable future?; and 3) how do the consumers see the outlook of the economy for the long term? It is noteworthy that the Consumer Sentiment

Index only embodies a fraction of the complete survey data gathered regularly. The complete survey set contains about fifty questions that covers specific traits of the expectation and consumption pattern of over 5000 respondents every month. The sampling methodology for the survey is statistically framed to represent the household population of the U.S. except Hawaii and Alaska.

The main questions are focused on three segments of the consumer sentiment namely the “personal finances, business conditions, and buying conditions”. The estimation of realised and anticipated changes of individual finances are rounded off by approximation of the expected nominal income changes for family and the expected change in real income. The perceptions on the overallMalaya business environment of the economy over the short-term and long-term period are assessed thoroughly. Explicit questions pertaining to various economicof indicators such as the rates of interest, expected inflation rate, expectation on the level of unemployment, the credence in economic policies etc. are parts of the overall questions that complement the more standard ones. Other miscellaneous questions include inquiries on consumers’ perception on the current market situations for “large household durables, vehicles, and houses.” UniversityThe University of Michigan Consumer Sentiment Index is commonly used as the test variable for macroeconomic forecasting. Bouwman & Jacobs (2011) for example used the index to test an innovated state-space model for resolving the problem of missing data due to amendments and lags in publication. Heij, Dijk & Groenen

(2011a) employed the index as test variable to determine the best models for the

152 forecasting of the Industrial Production (IP) growth rates and the growth rates of the

Conference Board’s Composite Coincident Index (CCI).

3.3.23 Commerce Department Composite Index of 11 Leading Indicators [Seasonally Adjusted]

A leading indicator is an economic variable that can be used as a pointer to the probable outcome of the economy for the foreseeable future. Leading indicators commonly move ahead of the cyclical trajectory of the business cycle which make them a prominent feature in economic studies. In contrast, lagging indicators are series of economic data that tail behind the aggregated economic activity whereas coincident indicators are economic series that approximately move in tandem with the peaks and troughs of the business cycle (Reilly & Brown, 2011; Shim & Siegel,Malaya 2001). Likewise, a composite index of of economic leading indicators is a weighted measurement of a combined selection of well-established predictors devised to signal the turning points of the economy. Apart from composite leading indicator index, there are two other types of indicator indices, namely the composite lagging indicator index21 and composite coincident indicator index22 (Reilly & Brown, 2011). According Malonis

(2000), the most widely used composite indicator index for economic forecasting is the

Composite Index of Leading Economic Indicators (CLI) which is made available monthlyUniversity by the U.S. Department of Commerce. The history of the CLI is traced back to the end of the 1950s where it is conceived by a fraternity of economists led by Geoffrey

21 The Composite Lagging Index of the U.S. Department of Commerce is composed of: 1) Average duration of unemployment; 2) Inventories to sales ratio, manufacturing and trade (chain 2000 dollars); 3) Change in labor cost per unit of output, manufacturing; 4) Average prime rate; 5) Commercial and industrial loans outstanding (chain 2000 dollars); 6) Consumer installment credit to personal income ratio; 7) Change in consumer price index for services. 22 The Composite Coincident Index of the U.S. Department of Commerce is composed of: 1) Employees on nonagricultural payrolls; 2) Personal income less transfer payments (chain 2000 dollars); 3) Industrial production; 4) Manufacturing and trade sales (chain 2000 dollars) Source: U.S. Census Bureau (n.d.). 153

Moore. The up-to-date CLI is published in various media and newspapers in the U.S. and is followed attentively by economists, corporate figures, market analyst, policymakers and the financial analysts for the forecasting of the economic outlook and strategising plans for the future.

The CLI is composed with 11 series of economic data which are reflective to the various facets of the economy. They are selected exclusively (with statistical adequacy) for their predictive efficiency on the future direction of the business cycles. The

Department of Commerce occasionally improves the series to ensure that the CLI remains as an accurate tool for economic prediction. The composition and methods used for the computation of the index is revised as demand arise. For example in year 1996, an adjustment was made to remove the percentage of change in the revised index of sensitive materials prices. For further technical readingMalaya on the construction of the CLI, see McGuckin, Ozyildirim & Zarnowitz, 2007of and Zarnowitz (1992). The 10 remaining components in the CLI after the elimination of the percentage of change in the revised index of sensitive materials prices are as follow: 1) Average weekly hours, manufacturing; 2) Average weekly initial claims for unemployment insurance; 3) Manufacturers' new orders, consumer goods and materials (1982 dollars);

4) Vendor performance, slower deliveries diffusion index; 5) Manufacturers' new orders, nondefense capital goods (1982 dollars); 6) Building permits, new private housingUniversity units; 7) Stock prices, 500 common stocks; 8) Money supply, M2 (chain 2000 dollars); 9) Interest rate spread, 10-year Treasury bonds less federal funds; 10) Index of consumer expectations (United State Census Bureau, n.d.[b]).

Camacho & Perez-Quiros (2002) acknowledged the efficacy of leading indicators in predicting the turning points of the U.S. economy and proposed an optimal filtering method to improve the transformation of the CLI into the form of probabilities

154 for economic downturn. Heij, Dijk & Groenen (2011a) concurred with the significance of leading indicators and found that the CLI is a powerful predictor for the growth rate of the industrial production which has a direct bearing to the economic direction. Seip &

McNown (2007) on the other hand employed the CLI to typify the macroeconomic indicators based on timing and accuracy to predict the turning points of the economy.

Flannery & Protopapadakis (2002) utilised the CLI as one of the test variables in their investigation on the impact of macroeconomic variables on the aggregate stock returns but discovered mixed results.

3.3.24 Chicago Fed National Activity Index Stock & Watson (n.d.) suggested that the ChicagoMalaya Fed National Activity Index (CFNAI) is the closest successor to the retired Stock and Watson Leading Index which was widely regarded as one of the most reliableof economic indicators. Qi (2001) found that the Stock and Watson (1989) Leading Index generated the smallest mean square forecast error (MSFE) within the one quarter horizon in predicting recession turning points. It also produced smaller margin of MSFE over the four-quarter horizon compared to the overall benchmark which indicated that the variable is a good recession predictor. Heij, Dijk & Groenen (2008; 2011a) in the same vein recommended the use of the Stock and Watson (1989) Leading Index and its derivation as predictors as a soundUniversity approach to forecast the economy.

According to Marcellino (2006), the “Stock and Watson (1989) Leading Index” was designed with the aim to construct a statistical tool for economic measurement based on the Burns and Mitchell’s (1946) pioneering coincident and leading macroeconomic indicators. The objective was to compile a set of macroeconomic indicators that are determined by a narrow amount of shared determinants and by 155 individualistic factors that are not correlated across the indicators tested. Stock and

Watson (1989) thus designed an aggregated unobservable factor in the form of index which is coincidental to the fluctuation of the business cycle. The index is composed through a dynamic factor model of four coincident variables, i.e. industrial production

(total), personal income (total less transfer payment, in 1982 dollars), manufacturing and trade sales (total, in 1982 dollars) and employees-hour in non-agricultural establishments. The Stock and Watson (1993) Leading Index was a revised and enhanced index to the 1989 predecessor after the original version performed unsatisfactory in detecting the turning point to the 1990 economic downturn. After updating the series for over 14 years, the original authors discontinued the series & recommended the CFNAI as its most suitable replacement (Stock & Watson, n.d.).

In reference to the official website (Federal ReserveMalaya Bank of Chicago, 2013), the CFNAI which is primarily designed to gauge the performance of the U.S. economic activity, is composed of 85 weighted averageof of variables. The data is updated on a monthly basis. Various studies conducted by well-established institutions (i.e. Federal

Reserve Bank of Chicago, Princeton University and the Harvard University) have proven the consistency of the CFNAI in imparting timely signals to the turning points of the economy. It was also shown to be a reliable indicator to changes in inflationary pressure. The movement of the CFNAI through the years follows closely the fluctuation of the business cycle and the oscillation of the inflationary pressure. The ultimate aim of theUniversity monthly index is to make available for researchers and market participants an

“objective real-time statistical measure” which is composed with an extensive series of monthly economy-related data that coincides with the U.S. economic performance.

156

The macroeconomic indicators used for the construction of the CFNAI comprised of four general classifications as follows (Federal Reserve Bank of Chicago,

2013):

“1) Production and income (23 series); 2) Employment, unemployment, and

hours (24 series); 3) Personal consumption and housing (15 series); and 4) Sales,

orders, and inventories (23 series).”

Further reading on the studies of the index formulation and its reliability tests can be found in Stock & Watson (1999), Fisher (2000), Evans, Liu & Pham-Kanter

(2002) and Brave (2009).

3.4 Specifications for Bear Markets and Bull MarketsMalaya Before examining the predictive significanceof of the list test variables on the proposed models for bear markets identification, the research herein begins by specifying the objective definitions for bear markets as outlined in previous studies.

Since there is yet to be a consensus on the universal definition for bear markets and vice versa (see Gonzalez, Powell, Shi & Wilson, 2005; Candelon, Piplak & Straetmans

2008) the research therefore adopts a more inclusive framework and examines all models applied by Chen (2009) 23 with an addition of another B-B variant and two innovatedUniversity methods. These models for the bear market identification are considerably comprehensive and covers across the parametric approach, semi-parametric approach and non-parametric approach.

23 Chen (2009) also included another approach called the “naïve moving average”. In the study, it was noted that the use of different specifications for the modelling of bull and bear markets does not result in any significant difference in the predictive power tests on variables. 157

3.4.1 Parametric Approach: Markov-switching (Filtered Probabilities & Smoothed Probabilities)

Consider 푝푡 as the nominal monthly Standard & Poor 500 (S&P 500 hereafter) closing price index at time t.

푟푡 = 100 ∗ (푙푛푝t − 푙푛푝t−1) (1)

24 Therefore, 푟푡can be interpreted as the compounded stock return of S&P 500 at time t. Consider the following Markov-switching model of stock returns with two regimes i.e. two respective mean and variance for different regimes:

2 푟푡 = μst + 휀푡, 휀푡 ~ 푖. 푑. 푑. 푁(0, 휎푠푡), (2)

2 where μst and 휎푠푡 are the respective regime dependent mean and variance. The two unobserved regimes are denoted by a latent processMalaya of 푠푡 that takes the dummy values of either 0 or 1. In this case, let 푠푡 = 0 to representof the bear market and 푠푡 = 1 the bear market.

A Markov-switching model can be specified in higher order by increasing the autoregressive lags i.e. MS(r). Perez-Quiros & Timmermann (2000) employed an arbitrary MS(1) 25 whereas Maheu & McCurdy (2000) tested up to MS(10) in their bull and bear markets models. No autoregressive lag is applied in the research nonetheless as the main aim here is to investigate the predictive efficacy of test variables on the list of Universityproposed ex-post models. Thus, the definition for bull market based on the Markov-

24 The compounded stock return is expressed the ∆ 푝t in percentage – thus the multiplication by 100 in 3.1.1.01 for the percentage denominator 25 An example of MS(r) model specification: A MS(1) or Markov-switching model AR (1) can be expressed as: 2 푦푡 = 훼0,푠푡 + 훽푠푡푋푡−1 + 휖푡 휀푡 ~ 푖. 푑. 푑. 푁(0, 휎푠푡) (Perez-Quiros & Timmermann, 2000) or 훼0 + 훽푡−1 + 휖푡, 푠푡 = 0 2 푦푡 = { 휀푡 ~ 푖. 푑. 푑. 푁(0, 휎푠푡), (Kuan, 2002) 훼0 + 훼0 + 훽푡−1 + 휖푡, 푠푡 = 1 where, 휇 = 훼 /(1 − 훽), 푠 = 0 0 0 푡 휇1 = 훼0 + 훼1/(1 − 훽), 푠푡 = 1 158 switching model is kept as parsimonious as possible to avoid the risk of over- specification. The justification for the use of parsimonious model is in conformation with Chen (2009).

a. Transition Probabilities

In accordance to Hamilton (1989), the switching of regime in the succeeding order of the Markov process is assumed to be dependent on the prior regime, 푠푡−1. Since the switching of state variable is a latent process, a two-regime Markov-switching model therefore is inferred to be conditioned by a Markov chain probabilistic value in the form of a fixed transition probability matrix as follows:

푝00 1 − 푝11 Malaya 푃 = [ 00 11 ], (3) 1 − 푝 푝 of where

00 푝 = P(st = 0|st−1 = 0), (4)

11 푝 = P(st = 1|st−1 = 1), (5) where 푝00 represent the probability of the system remaining in regime ‘0’, given that the preceding regime is also regime ‘0’ while 푝11 denote the probability of the system remainingUniversity in regime ‘1’, given that the preceding regime was also regime ‘1’. 00 Therefore, the 1 − 푝 represents the probability that the 푟푡will switch from regime ‘0’ in period 푡 − 1 to regime ‘1’ in the subsequently at 푡. Likewise, 1 − 푝11denotes the probability of changing of regime from regime ‘1’ in the prior period of 푡 − 1 to regime

‘0’ in the next period of 푡. Therefore, the two parameters of the transition probability matrix (푝00 and 푝11) dictate the fluctuation of the state variable i.e. the S&P 500 return.

159 b. Probabilities for Parameters Estimation

The vector parameters and the transition probabilities of the given model are summarised as:

00 11 휃 = (휇st, 휎st, 푝 , 푝 )′, (6)

These parameters are used for the maximum likelihood estimation of the model.

As noted, the state variables in a Markov-switching model are assumed to be latent. Therefore the estimation of parameters for the model requires the probabilities of each state to be estimated first. However, the estimation of state changing probabilities requires the optimal forecasts or the “conditional expectations” to be estimated first.

Therefore, let Ω푡 = {푟푡, 푟푡−1, … , 푟1} represent the complete set of data on the observed logarithm of returns up to time 푡 which denotes the Malayainformation at time 푡. Therefore, Ω푡 is the information set that encompasses theof complete observation of the research. The optimal forecasts of 푠푡 = 푗, 푗 = 0; 1 are derived based on the numerous information sets that emerge at different junctures throughout the observation. The probabilities for optimal forecasts are in the following:

Prediction probabilities, estimated based on the past information preceding to time t:

푃(푠푡 = 푗|Ω푡−1; 휃) (7) FilteredUniversity probabilities, estimated based on the current information and information prior to time t:

푃(푠푡 = 푗|Ω푡; 휃) (8)

Smoothed probabilities, estimated based on the complete information from the observation:

160

푃(푠푡 = 푗|Ω푇; 휃) (9)

which is subsequently used to estimate the likelihood of the state switching variable 푠푡.

Given that the density of 푟푡 conditional on Ω푡−1 for being in state 푠푡 = 푗, 푗 = 0; 1 follows a normal distribution assumption26:

2 1 −(푟푡−휇푗) 푓(푟푡| 푠푡 = 푗, Ω푡−1; 휃) = 푒푥푝 { 2 }, (10) 2 휎푗 √2휋휎푗

In reference to the prediction probabilities of (7), the density of 푟푡 conditional on

Ω푡−1 alone can be computed recursively based on (10) with the summation of the probabilities-weighted densities for the two switching states:

( | ) ∑1 | ( | ) 푓 푟푡 Ω푡−1; 휃 = 푗=0 푃(푠푡 = 푗 Ω푡−1; 휃)푓 푟푡 푠푡 = 푗,Malaya Ω푡−1; 휃 (11) The conditional probabilities, 푃(푠푡 = 푗|Ω푡−1; 휃) is interpreted as the probabilities to be in state 푗 at time 푡 basedof on the 푡 − 1 information (Ω), given the vector of parameters of 휃.

The filtered probabilities of 푠푡 = 푗, 푗 = 0; 1 are:

푃(푠푡 = 푗|Ω푡−1; 휃)푓(푟푡|푠푡 = 푗, Ω푡−1; 휃) 푃(푠푡 = 푗|Ω푡; 휃) = 푓(푟푡|Ω푡−1; 휃)

(12)

UniversityThe association of the filtered probabilities and the prediction probabilities based on the Bayes theorem can be expressed as:

푃(푠푡 = 푗|Ω푡; 휃) = 푝0푗 푃(푠푡 = 0|Ω푡; 휃) + 푝1푗 푃(푠푡 = 1|Ω푡; 휃) (13)

26 Perez-Quiros & Timmermann (2000) nevertheless noted that the Gaussian assumption in Markov- switching model has its drawbacks since the combination of a numerous normal samples could spuriously accommodate a considerable extent of “densities with nonzero skewness and fat tails”. 161 where 푝0푗 = 푃(푠푡+1 = 푗|푠푡 = 0) and 푝1푗 = 푃(푠푡+1 = 푗|푠푡 = 1) are the transition probabilities. Therefore, equation (10) to equation (13) complete the filtering probability. The approximated conditional log-likelihood function is derived as a by- product and expressed as:

푇 푙(푟푡|Ω푡−1; 휃) = ∑푡−1 푙푛 푓(푟푡|Ω푡−1; 휃) (14)

To obtain smoothed probabilities Kim (1994) and Kuan (2002) showed that:

푝푗푘푃(푠푡 = 푗|Ω푡; 휃) 푃(푠푡 = 푗|푠푡+1 = 푘, Ω푇; 휃) = 푃(푠푡 = 푗|푠푡+1 = 푘, Ω푡; 휃) = 푃(푠푡+1 = 푘|Ω푡; 휃)

(15) for 푗, 푘 = 0,1. The smoothed probabilities thus can beMalaya denoted as: 푃(푠푡 = 푗|Ω푇; 휃)

= 푃(푠푡+1 = 0|Ω푇; 휃) 푃(푠푡 = 푗|푠푡+1 = 0, Ωof푇; 휃) + 푃(푠푡+1 = 0|Ω푇; 휃) 푃(푠푡 = 푗|푠푡+1

= 0, Ω푇; 휃)

푝푗0푃(푠푡+1 = 0|Ω푇; 휃) 푝푗1푃(푠푡+1 = 1|Ω푇; 휃) = 푃(푠푡 = 푗|Ω푡; 휃) ∗ { + } 푃(푠푡+1 = 0|Ω푡; 휃) 푃(푠푡+1 = 1|Ω푡; 휃)

(16)

It is simplified by Chen (2009) as:

University푇 푃(푠푡 = 푗|푦 ) (17) where 푗 = {0,1}

This part of the research benefited enormously from works by Chen (2009),

Chin, Mital & Chua (2000), Engel & Hamilton (1990), Hamilton (1989), Kuan (2002),

Maheu & McCurdy (2000) and Perez-Quiros & Timmermann (2000).

162

3.4.2 Semi-Parametric Approach: Markov-switching (Dichotomised Smoothed Probabilities)

In reference to the 3.4.1 Section, the research proposes to dichotomise the smoothed probabilities output extracted from Markov-switching model. The modest innovation is simply named as “dichotomised smoothed probabilities”. The dichotomised smoothed probabilities can be expressed as:

푇 1 (푏푒푎푟 푚푎푟푘푒푡)푖푓 푃(푠푡 = 푗|푦 ) > 0.5, 퐷푡 { 푇 (18) 0 (푏푢푙푙 푚푎푟푘푒푡)푖푓 푃(푠푡 = 푗|푦 ) < 0.5

Justification to this innovation is to enable the research to examine the Markov- switching results in parallel with other semi-parametric and non-parametric models using the same test statistic produced with the probit analysis.

Malaya 3.4.3 Semi-Parametric Approach: Naïveof Moving Average The “naïve moving average” approach averages out the returns of stock market over 푡 periods in a forward moving manner. The bull and bear markets are thus determined by the sequences of mean derived by continuously removing the first observation of a predetermined subset size and replacing it next to the observation, which trail the original series. The process is repeated to the last observation of the complete data seriesUniversity (Chen 2009). 푘 Consider 푟̅푡 as the moving average of the last 푘 observations of the series of returns,

r + r + ⋯ + r 푟̅푘 = 푡−1 푡−1 푡−푘 푡 푘

(19)

163

The dummy variable 퐷푡 is denoted as:

푘 1 (푏푒푎푟 푚푎푟푘푒푡)푖푓 푟̅푡 < 0, 퐷푡 { 푘 (20) 0 (푏푢푙푙 푚푎푟푘푒푡)푖푓 푟̅푡 > 0,

Therefore, if the average for the preceding k periods is positive, the run is identified as bull market, otherwise, if the rolling average of k is negative, the period would be marked as bear market. The naïve moving average approach has the advantage of smothering volatility in determining the bull / bear regime compared to the various variant of B-B models. The model also removes the need for arbitrary assignment of threshold values for sharp decline to trigger a regime switch as proposed by Lunde &

Timmermann (2004). The detailed description of the Lunde & Timmermann’s B-B algorithm is featured in Section 3.4.5 in the following.

Malaya

3.4.4 Semi-parametric Approach: Naïveof Moving Average Negative Return

The research reckons that naïve moving average model (as discussed above) has its shortcoming where it is susceptible to outlier-caused bias (both positive and negative) that could result in the misspecification of regimes. The research thus seeks to improve the model with two enhancements to address the issue.

Firstly, for negative return filtering, consider equation (19), the rules are overriddenUniversity by the algorithm as follows:

1훼 (푏푒푎푟 푚푎푟푘푒푡)푖푓 훼 = {푎, 푏, 푐} 퐷푡 { 푘 (21) 0 (푏푢푙푙 푚푎푟푘푒푡)푖푓 푟̅푡 > 0,

Where,

̅ 푘 1푎 = {푟푡 < 0, 푟푡 < 0} (22)

164

푘 1푏 = {푟̅푡 < 0, 푟푡 > 0} (23)

푘 1푐 = {푟̅푡 > 0, 푟푡 < 0} (24)

and the priority for switching rule of 1훼 (푏푒푎푟 푚푎푟푘푒푡) > 0 (푏푢푙푙 푚푎푟푘푒푡), i.e. the identification of 0 (bull market) is overwritten if it contradicts with any of the rule set for 1 (bear market).

Thus, algorithm (21) assigns higher priority for bear market detection compared to the original naïve moving average model. Such proposal is justified because stock markets are more inclined to experience prolonged gradual upward trends with occasional short and sharp reversals. Thus bear markets regimes are more temporaneous and difficult to detect with a rigid algorithm. The enhancements introduced by the research reduce the chance of positive bias caused by outliers and allowMalaya the model to be more sensitive for bear markets identification by detecting the combination of both the point negative

푘 return (-) 푟푡 and the moving average negativeof return (-)푟̅푡 .

Secondly, the algorithm is complemented with a smoothed states filter to eliminate indiscriminate point of negative return (-) 푟푡 in the observation simply by imposing a rule of a minimum of 6 consecutive runs of negative return i.e. {푛 (−)푟푡+휏 ≥

6} at time 푡 before a streak of bull market is allowed to switch to bear market. The 6 consecutive runs is an arbitrary number that is in concurrence with other models such as theUniversity 6-month window for both the naïve moving average model, the B-B models and the JLS models as follows.

165

3.4.5 Non-parametric Approach: Lunde & Timmermann’s B-B Algorithm (2004) Variant

The B-B algorithm that is originally used on the GDP time series to determine the heterogeneous states for business cycles (naturally, also the identification of turning points in the economy, peaks and troughs), was first adapted by Pagan and Sossounov

(2003) for the study on the stock market. Candelon Piplak & Straetmans (2008) underscored that the main characteristic of the nonparametric method of B-B algorithm is in adopting the “local maxima and minima” approach to determine the bull and bear markets with either the stock market index or returns time series.

There are two more prominent variants to the B-B algorithm which are adapted for stock market. The first is the aforementioned variant introduced by Pagan and Sossounov (2003) in which the algorithm was subsequentlyMalaya modified by Lunde & Timmermann (2004) and later adopted in a study by Mahue, McCurdy & Song (2009). The construction of the Pagan & Sossounovof (2003 ) B-B variant is encapsulated in the steps as follows: 1) Determine the turning points of peaks and troughs with an 8-month timeframe; 2) Impose switching of phases through the removal of the lesser adjoining peaks and the greater of adjoining troughs; 3) Remove phases that are fewer than 4 months except for changes which are over 20%27; 4) Remove cycles that are not more than 16 months. Maheu, McCurdy & Song (2009) recommended a shorter 6-month timeframe. The use of the 6-month window also allows the results of the variant to be examinedUniversity vis-à-vis the Candelon Piplak & Straetmans (2008) variant which is explained in the following.

27 Citing the October 1987 crash as the primary example, Pagan & Sossounov (2003) pointed that the phase of the stock market decline only lasted for 3 months of which then after, the market began to recover. Setting the elimination rule for phases lasting less than 4 months would have filtered out the aforementioned crash. However, setting the threshold for phase switching at 3 months minimum would result in many spurious cycles. Thus, another additional restriction was imposed as such that the removal of phases that are fewer than 4 months can be overruled should the stock market falls by 20% or more within a month. 166

It is important to note that the 20% rule for decline is an arbitrary value for the algorithm. The in-depth justification can be found in the study by Pagan & Sossounov

(2003). The 20% rule is also a convenient figure that is commonly cited as the threshold for sharp declines by market participants (Candelon, Piplak & Straetmans, 2008.; Lunde

& Timmermann, 2004). Lunde & Timmermann (2004) nonetheless also suggested that other combinations of thresholds could be more sensitive for the identification regime switching i.e. a 15% surge for bull market vis-a-vis a 15% decline for bear market or a

15% increase for bull market vis-a-vis a 10% decrease for bear market. The research thus chooses the 15% threshold for both the peak and throng turning points as recommended.

To determine local minimum or maximum with the 6-month timeframe rule, consider a local maximum at time 푡 , therefore 푃푚푎푥 = 푃 , where at the time of 푡 , the 0 푡0 Malaya푡0 0

푃푡0is the tracked price of stock of the stochastic process. The “time-stopping variables” for a run of bull market is expressed in the offollowing:

푚푎푥 푚푎푥 휏푚푎푥(푃푡0 , 푡0| 퐼푡0 = 0) = 푖푛푓{푡0 + 휏 ∶ 푃푡0+휏 ≥ 푃푡0 } (25)

푚푎푥 푚푎푥 휏푚푖푛(푃푡0 , 푡0| 퐼푡0 = 0) = 푖푛푓{푡0 + 휏 ∶ 푃푡0+휏 ≤ 0.85푃푡0 } (26)

Next, a filter is imposed on phase switching if one of the scenarios as below occurs:

If 휏University푚푎푥 < 휏푚푖푛, the run of bull market remains, the new value of the peak is computed, 푚푎푥 푃푡0+휏푚푎푥 = 푃푡0+휏푚푎푥 and the peak time prior at 푡0 is replaced with set 퐼푡0+1 =

⋯ 퐼푡0+휏푚푎푥 = 1. Return to rule (25) and (26).

If 휏푚푎푥 > 휏푚푖푛, a trough will commence at time 푡0 + 휏푚푖푛, hence the bear market

to would have occurred from 푡0 + 1 푡0 + 휏푚푖푛, 퐼푡0+1 = ⋯ 퐼푡0+휏푚푖푛 = 1. The value of

167

푚푖푛 푃푡0+휏푚푖푛 = 푃푡0+휏푚푖푛 is computed and the time at 푡0 is identified as a peak. Next rule

(27) and (28) is applied for bear market.

Vice versa, the “time-stopping variables” for a run of bear market is denoted in the following:

푚푖푛 푚푖푛 휏푚푖푛(푃푡0 , 푡0| 퐼푡0 = 1) = 푖푛푓{푡0 + 휏 ∶ 푃푡0+휏 ≤ 푃푡0 } (27)

푚푖푛 푚푖푛 휏푚푎푥(푃푡0 , 푡0| 퐼푡0 = 1) = 푖푛푓{푡0 + 휏 ∶ 푃푡0+휏 ≥ 1.15푃푡0 } (28)

If one of the scenarios as below occurs:

If 휏푚푖푛 < 휏푚푎푥, the run of bear market remains, the new value of the trough is computed,

푚푖푛 푃푡0+휏푚푖푛 = 푃푡0+휏푚푖푛 and the trough time prior at 푡0 is replaced with set 퐼푡0+1 =

⋯ 퐼 = 1. Return to rule (27) and (28). 푡0+휏푚푖푛 Malaya If 휏푚푖푛 > 휏푚푎푥, a peak will commence at of time 푡0 + 휏푚푎푥. Hence a run of bull market to would have occurred from 푡0 + 1 푡0 + 휏푚푎푥, 퐼푡0+1 = ⋯ 퐼푡0+휏푚푖푛 = 0. The value of

푚푎푥 푃푡0+휏푚푎푥 = 푃푡0+휏푚푎푥is computed and the time at 푡0 is recorded as a trough. Return to rule (25) and (26) for bull market.

For the Lunde & & Timmermann (2004) model as explained above, the nominal binary representations for both regimes of bear markets and bull markets are reversed from the original study (i.e. in the literature, bear = 0; bull = 1 were changed to bear = 1; bullUniversity = 0) to be standardised with other regime switching models used in the research.

168

3.4.6 Non-parametric Approach: - Candelon, Piplak & Straetmans’s B-B Algorithm (2008) Variant

The Candelon, Piplak & Straetmans (2008) B-B variant which was also used in the study by Chen (2009) is more simplistic in relative to other variants. The focus of algorithm is only for the identification of local minimum and the local maximum of

푟푡within a rolling 6 months window based on the original B-B algorithm.

Thus, a local maximum is identified at time 푡 whenever:

푟푚푎푥 = {푟푡 > 푟푡±6} (29)

Likewise, a local minimum is identified at time 푡 whenever:

푟푚푖푛 = {푟푡 < 푟푡±6} (30)

Once the local maximum / minimum i.e. the Malayaturning points in the time series are ascertained, the 푟푚푎푥 to 푟푚푖푛 period is identifiedof as bear regime (퐷푡 = 1) and the 푟푚푖푛 to 푟푚푎푥 period is identified as bull regime (퐷푡 = 0), where 퐷푡 is a binary dummy variable to denote the alternating regimes in series.

3.5 Specification of Crashes and Rebounds

The past results of stock market crashes (based on the S&P 500 Index) identified with theUniversity JLS model are well documented by Feigenbaum (2001), Johansen (2004), Johansen & Sornette (2010), Sornette & Cauwels (2014) and Liberatore (2011a, 2011b). Updates of the subsequent identifications can be obtained from the Financial Crisis Observatory website (at http://www.er.ethz.ch/financial-crisis-observatory.html). The site is co- founded and headed by Didier Sornette, the co-pioneer of the JLS model. Stationarity is not a required assumption for the scale-invariant approach (Chainais, Reidi, & Abry,

169

2005) which is an advantage compared to most econometric methodologies. Thus the output of the JLS models can be derived without the need for data transformation from the original form (i.e. S&P 500 Index) into logarithm or stock returns.

3.5.1 Scale-invariant Approach: JLS Model and JLS “Negative Bubbles” Model

The JLS “negative bubbles” model is an innovation from the original Johansen-Ledoit-

Sornette (JLS) model by Johansen, Ledoit & Sornette (1999). Yan, Woodard & Sornette

(2012) who coined the term “negative bubbles” for the innovated JLS model and defined the concept as a reflection / mirror image for market bubbles fuelled by

“positive feedback mechanisms”. The model identify the periods of ± 200 days (about 6 ½ months) before and after the bottom of a troughMalaya in stock market. These are the transient periods that sandwich the turning points of a throng. The pre and post periods of turning points are not symmetrical in mostof cases.

The complex procedure of the JLS model derivation is not illustrated in-depth in the research. As suggested by Bree & Joseph (2010) and shown in studies by Yan,

Woodard & Sornette (2012) and Jiang, Zhou, Sornette, Woodard, Bastiaense &

Cauwels (2010), the simplest form of the JLS model which includes all of its technical assumptions could be epitomised in the form of LPPL. UniversityThe equation of LPPL is represented as follows: 훽 푦푡 = 퐴 + 퐵(푡푐 − 푡) {1 + 퐶푐표푠(휔 log(푡푐 − 푡) + ∅)} (31)

푦푡 > 0 is the price (index), or the log of the price, at time t; A > 0 is the value that 푦푡 would be if the bubble were to last until the critical time 푡푐; B < 0 is the increase in 푦푡 over the time unit before the crash, if C were to be close to zero; C is the magnitude of the fluctuations around the exponential growth, as a proportion; 170

푡푐 > 0 is the critical time; t < 푡푐 is any time into the bubble, preceding 푡푐; 훽 is the exponent of the power law growth; 휔 is the frequency of the fluctuations during the bubble; 0 ≤ ∅ ≤ 2휋 is a shift parameter.

The identification of dates for rebounds is an inverse to the modelling process of the basic JLS model. Day (d) is denoted as the local minimum of price for a window of

± 200 days of which a rebound – Rbd would ensue.

푅푏푑 = {푑|푃푑 = min{푃푥} , ∀푥 ∈ |푑 − 200, 푑 + 200|} (32)

The research calibrates the ± 200 days (about 6 ½ months) proposed by Yan,

Woodard & Sornette (2012) to ± 6 months in conformation to the rule-of-thumb threshold requirement for regime-switching of the other B-B models used. The complete sampling period for predictability tests is composedMalaya of the dating results of the JLS “negative bubbles” model and the dates of peaks and the corresponding ± 6 months windows of the JLS model. The last day of(d) of local minimum of price for a window was documented on 9 March 2009 (09-03-2009). Up until the final month of research observation (i.e. June 2014), no stock market crash was recorded in the Financial Crisis

Observatory website.

3.6 Predictive Regression Model for In-sample and Out-of-sample Test for Parametric Approaches University A predictive regression model is used to evaluate the significance of test variables in predicting the bear markets regime derived from the Markov-switching model. For the predictive regression model, consider:

푄0,푡+푘 =∝ +훽푥푡 + 푒푡 (33)

171 where 푥푡 is the test variable to predict the bear markets. Equation (33) represents the in- sample test for the null hypothesis which attests that the test variable is not significant in predicting the bear markets for a set of predetermined forecasting horizons ahead, i.e.

훽 = 0 against the alternative hypothesis, 훽 ≠ 0. Therefore, the forecastability of the ex-

2 post model with the use of 푥푡 is measured by the t-statistic and the 푅 (goodness of fit) corresponding to the 훽̂.

Next, the research also investigates the predictability of the bear markets with the out-of-sample test using the identical set of test variables. The restricted model and unrestricted model are shown in the following:

Restricted model:

푄0,푡+푘 =∝ +푢1,푡 Malaya (34) Unrestricted model: of 푄0,푡+푘 =∝ +훽푥푡 + 푢1,푡 (35)

Consider model (34) and model (35), should the value derived from the out-of- sample test is positive, it would indicate that the unrestricted model has a better predictive capability compared to the restricted model. As vis-à-vis comparison, consider the mean square prediction error (MSPE) is used as a measure of forecasting performance. If the MSPE (unrestricted) < MSPE (restricted), it would indicate that the unrestrictedUniversity model performs better than the restricted model. Thus a conclusion can be drawn that the predictive power of test variables (푥푡) is significant. The research uses the improved test statistic of ‘MSPE-adj’ proposed by Clark & West (2007) to compare the output of the restricted model and the unrestricted model.

172

For the MSPE-adj test, the overall sample of T observations is divided into in- sample and out-of-sample factions. The total in-sample observations is denoted as R,

푡 = 1, … , 푅 and 푃 is the total out-of-sample observations, 푡=푅+1,…,푅+푃 .

Therefore, 푅 + 푃 = 푇. The estimation procedure is recursive. The forecasting errors are

1 2 denoted as 푢̂푡+푘 and 푢̂푡+푘 for the restricted and unrestricted models respectively.

̂1 Likewise, the predictions of 푄0,푡+푘 from the two models are represented as 푄푡+푘 and

̂2 푄푡+푘. Clark & West (2007) MSPE-adj statistic is represented as follows:

푃푓̅ 푀푆푃퐸 − 퐴푑푗 = √ (36) √푉̂ where,

̅ −1 ̂ 푓 = 푃 ∑ 푓푡+푘푡 Malaya (37) where,

̂ 1 2 2 2 ̂1 ̂2 of2 푓푡+푘 = (푢̂푡+푘) − [(푢̂푡+푘) − (푄푡+푘 − 푄푡+푘) (38)

̂ ̅ and 푉̂ is the sample variance of (푓푡+푘 − 푓).

The MSPE-adj statistic is an “approximately normal test for equal predictive accuracy in nested models”. The MSPEs for both restricted and unrestricted model are equal for the null hypothesis while the MSPE for restricted model has a larger value compared to the unrestricted model for the alternative hypothesis. Null hypothesis is rejectedUniversity if the test statistic shows an acceptable positive threshold value. The asymptotic distribution for the statistic is the standard normal distribution (Clark & West, 2007;

Chen, 2009).

173

3.7 Probit Model for In-sample and Out-of-sample Test for Semi-parametric, Non-parametric and Scale-invariant Approaches

For the two B-B variants and the JLS “negative bubbles” model, a binary variable is obtained: 퐷푡 = 1 indicates a bear market while 퐷푡 = 0 indicates a bull market. Consider the axiom of the probit model in the following:

푃(퐷푡+푘 = 1) = 퐹(∝ +훽푥푡) (39)

To measure the in-sample fit, the research adopts the Estrella and Mishkin

(1998) approach to compute the pseudo 푅2 which was first suggested by Estrella

(1998). Let 퐿푢 represent the maximised probit likelihood value, and let 퐿푐 represent the maximised likelihood value restricted to the constraint that all coefficients are zero except for the constant. The measures of fit are expressedMalaya as follows: 2 푙표푔퐿푢 푝푠푒푢푑표 푅푀푐푓푎푑푑푒푛 = 1 − ( ) (40) 푙표푔퐿푐 of and

−(2/푇)푙표푔퐿푐 2 푙표푔퐿푢 푝푠푒푢푑표 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ = 1 − ( ) (41) 푙표푔퐿푐

A low value of the pseudo- 푅2 indicates ‘‘no fit”, while a high pseudo- 푅2 = 1 signifies ‘‘perfect fit”. To examine the prediction of the out-of-sample probit model, the research uses the quadratic probability score (QPS) proposed by Diebold & Rudebusch (1989).University The equation is illustrated in the following:

−1 2 푄푃푆 = 푇 ∑푡 2[푃(퐷푡+푘 = 1) − 퐷푡+푘] (42)

The QPS ranges from 0 to 2, with the value of 0 equivalent to perfect accuracy.

174

3.8 Concluding Remark

Apart from the description of test variables, most sections in the chapter are dedicated to the elaboration of the ex-post models used for the research and the methods used for testing the selected test variables on the output of these models in an ex-ante fashion.

These models and methods are expressed concisely in the form of mathematical notations. The sections that describe the axiomatic specifications of the models for stock market declines are self-explanatory. Thus, the focus in this section is to underscore the novelties introduced by the research in terms of the use of variables and the innovations to some methodologies.

Firstly, the use of the Shiller’s “cyclically adjusted price earnings ratio” or

CAPE as a test variable across all models proposed for the research is unprecedented.

Secondly, the past studies by Shiller mostly investigatesMalaya this set of variables from the perspective of behavioural finance. In theseof past studies, CAPE was proposed as an important market indicator for bubbles and harbinger to financial crisis (Campbell &

Shiller, 1998; Shiller, 2005). The findings conformed with one of the latest studies (at the time of writing) by Keimling (2016) which found encouraging result in predicting stock market returns using CAPE as the sole independent variable.

The combination of the Shiller’s Financial Variables and the Estrella &

Mishkin’s Financial Variables provides a comprehensive spectrum of market fundamentalsUniversity for the examination of stock market declines for the research. The Shiller’s Financial Variables in general represent variables of the market’s aggregated microvaluation. Estrella & Mishkin’s Financial Variables complements the former variables set with market fundamentals from the macroanalysis aspect. These two sets of variables as discussed previously were identified in the earlier studies as the most suitable indicators for recession and financial crises (e.g. Estrella & Mishkin, 1998; Qi,

175

2001; and Shiller, 2005). Selected variables from the lists such as dividend yield, 3- month Treasury bill rate, inflation rate etc. were tested as predictors in a number of studies that are related to this research, namely Chang (2009), Chauvet & Potter (2000) and Chen (2009).

The Shiller’s Financial Variables and the Estrella & Mishkin’s Financial

Variables are combined in the research and regrouped into self-explanatory categories entitled Stock Market Fundamentals, Financial Market Fundamentals, Industrial

Indicators, Market Sentiments, and Indexes of Leading Indicators. The categories are labelled to best reflect the variables’ representation accordingly. The ex-post models for the identification of stock market declines are also grouped in accordance to their respective statistical characteristics i.e. parametric, semi-parametric, non-parametric or scale-invariant. Malaya The research follows the Clark & ofWest (2007) recursive estimation to achieve the most stringent and parsimonious requirements (at the time of writing) for the nested model prediction. The approach which introduced the MSPE-adj statistic requires the adherence of the research to a set of conditions in dividing the complete observation for the in-sample and out-of-sample portions. The procedure is meticulously repeated for each forecasting horizon (i.e. k = 1, k = 3, k = 6, k = 9, k = 12) as such that different horizons yield different set of sampling portions. The sampling portions determined for eachUniversity horizon are applied uniformly on all models for the subsequent in-sample and out- of-sample tests.

Most importantly, the research also contributes to the extension of literature on the technical aspects through the introductions of the naïve moving average negative return model, the dichotomised smoothed probabilities of Markov-switching model and the integrated identifications of crashes and rebounds for stock market via the JLS

176 model (and its variant of JLS “negative bubbles” model). The mathematical annotations and description to these models are shown in the above. The results of these innovations are discussed in the following chapters.

The aim of the naïve moving average negative return model is to reduce the volatility of regime changing compared to the original naïve moving average used by

Chen (2009). On another note, the introduction of the dichotomised smoothed probabilities of Markov-switching model allows the output from a Markov-switching model to be examined with tests that are meant for non-parametric approach. This allows the adjusted output to be analysed with the probit model and enables a side-by- side comparison of results with other semi-parametric and non-parametric models used in the research. The exploratory integrated identifications of crashes and rebounds with the combined JLS models is another modest contributionMalaya of the research. It enables a parallel comparison of the shorter and sharper type of stock market declines (crashes) with the more gradual and prolonged typeof (bear markets) in testing the predictive significance of the selected market fundamentals.

University

177

CHAPTER 4

RESULTS

4.0 Introduction

Results in this chapter are segmented into five sections. The first section herein includes the introduction of the chapter and is followed by Section 4.1 - “Descriptive Statistics” that summarises the statistics of variables for the research. The corresponding charts on the historical movement of the respective S&P 500 index and RT100, pairing with the list of test variables proposed for the research are featured in Appendix B. The data sources of variables are enclosed in Appendix A. DiscussionsMalaya on descriptive statistics that are presented in Table 4.1, 4.2, 4.3 andof 4.4 are consolidated and featured after Table 4.3.

Section 4.2 tests the stochastic assumptions of the research’s variables. Tables in this section illustrate the results for the tests of Augmented Dickey-Fuller, Phillips-

Perron and Elliott-Rothenberg-Stock DF-GLS that are used to diagnose the problem of unit root in time series. Test variables of the research which serve the proxies to the market fundamentals that are not stationary are transformed and subsequently retested.

UniversityResults of models for stock market declines identification are featured in section

4.3. These models as aforementioned in the earlier part of the research, include two variants of the Markov-switching model (i.e. regime switching identified with the filtered probabilities and another with the dichotomised smoothed probabilities – an innovation introduced by the research), two variants of naïve moving average (i.e. the basic model and the naïve moving average negative return – another innovation 178 introduced by the research), two variants of the B-B algorithm (i.e. the Candelon, Piplak

& Straetmans (2008) variant and the Lunde & Timmermann (2004) variant and the combined results of two JLS models (i.e. the JLS model for crashes and the JLS

“negative bubbles” model for rebounds).

Section 4.4 explains in detail how the research derives the observations timeframe for out-of-sample tests for the research using the approach introduced by

Clark & West (2007). The in-sample predictability test results for predicting bear stock markets (and predicting bubble induced-crashes) of all the research’s models and algorithms as recapped above are illustrated in tables in section 4.5. Each of the tables is followed by a brief summary on the values of results. Section 4.6 which ensues vis-à-vis features the out-of-sample predictability test results and short elaborations. Malaya of

University

179

4.1 Descriptive Statistics

Table 4.1: Summary Statistics of Variables

Variable Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Prob. SP500 592.329 350.250 1947.090 67.070 525.697 0.625 1.898 65.550 0.000 RT100 0.543 0.831 11.352 -22.804 3.684 -1.018 7.056 486.421 0.000 DVD 12.616 12.010 37.380 2.900 8.517 0.797 2.809 60.844 0.000 EAR 30.271 19.500 103.120 5.130 25.702 1.167 3.247 130.187 0.000 RDVD 21.376 20.680 37.380 15.730 4.649 1.257 4.152 180.718 0.000 REAR 48.560 40.560 101.890 7.600 20.022 1.038 3.191 102.742 0.000 CAPE 19.443 19.300 44.200 6.640 8.374 Malaya0.779 3.460 62.290 0.000 CPI 129.648 133.800 238.343 33.100 62.628 0.000 1.767 35.942 0.000 TB3M 5.125 5.070 16.300 0.010 3.228 0.523 3.682 36.820 0.000 TN5Y 6.413 6.300 15.930 0.620 of3.070 0.430 3.287 19.404 0.000 TN10Y 6.771 6.550 15.320 1.530 2.787 0.628 3.335 39.956 0.000 S10Y3M 1.646 1.740 4.420 -2.650 1.284 -0.469 2.707 22.819 0.000 S5Y3M 1.289 1.350 4.330 -2.250 1.009 -0.370 3.163 13.582 0.001 M1 906.019 822.100 2834.800 174.200 603.059 0.919 3.596 88.110 0.000 M2 3804.528 3263.600 11351.300 492.100 2853.499 0.868 2.811 72.104 0.000 RM1 646.410 644.100 1192.600 458.600 139.097 1.555 6.032 445.689 0.000 RM2 2570.090 2407.100 4775.600 1486.700 811.636 0.915 3.004 79.091 0.000 ISMI 46.529 46.200 66.500 24.600 6.306 -0.053 4.009 24.314 0.000 COPE 33239.650 31165.000 59181.000 15793.000 10343.370 0.360 2.121 30.481 0.000 NPHP 1399.526 1425.000 2419.000 513.000 416.557 -0.089 2.462 7.589 0.022 CMUO 2.981 2.900 32.100 -26.700 9.608 0.107 3.317 3.465 0.177 PMI 52.354 53.000 72.100 29.400 6.528 -0.547 4.192 61.883 0.000 MSCI 84.539 87.000 112.000 51.700 12.499 -0.286 2.450 14.854 0.001 RGDPG 2.863 3.000 16.500 -8.200 3.371 -0.131 4.939 90.435 0.000 CDLI 71.040 67.400University 108.100 43.500 19.740 0.259 1.642 49.892 0.000 CFNAI -0.002 0.080 2.680 -4.990 1.004 -1.138 6.356 388.384 0.000

180

4.2 Unit Root Tests for Stationarity and Data Transformation

Table 4.2: Unit Root Tests: Pre-transformed Variables

Variable ADF PP DF-GLS t-Statistic Prob. Adj. t-Stat Prob. t-Statistic SP500 1.1013 0.997 0.763 0.993 2.167 RT100 -18.318 0.000*** -18.346 0.000*** -17.178** DVD 0.960 0.996 2.297 1.000 1.803 EAR -1.086 0.723 -0.680 0.849 -0.178 RDVD 0.198 0.972 1.186 0.998 0.178 REAR -2.555 0.103 -1.933 0.317 -2.093 CAPE -1.243 0.657 -1.310 0.627 -1.247 CPI 0.962 0.996 1.100 0.998 2.570 TB3M -1.905 0.330 -1.765 0.398 -1.921 TN5Y -1.176 0.686 -1.311 0.626 -1.191 TN10Y -1.116 0.711 -1.250 0.654 -1.072 S5Y3M -4.936 0.000*** -4.425 0.000*** -4.205*** S10Y3M -4.237 0.001*** -3.944 0.002*** -3.122*** M1 3.459 1.000 5.832 1.000 4.152 M2 6.428 1.000 Malaya11.697 1.000 7.059 RM1 1.987 1.000 3.213 1.000 2.561 RM2 3.534 1.000 3.726 1.000 5.964 ISMI -5.232 0.000***of -6.405 0.000*** -3.404*** COPE -1.970 0.300 -2.323 0.165 -0.361 NPHP -2.323 0.165 -2.563 0.101 -1.824 CMUO -4.391 0.000*** -4.182 0.001*** -4.339*** PMI -5.829 0.000*** -5.258 0.000*** -3.276*** MSCI -3.745 0.004*** -3.591 0.006*** -2.994*** RGDPG -6.343 0.000*** -8.390 0.000*** -4.771*** CDLI -0.852 0.803 -0.694 0.864 0.452 CFNAI -6.985 0.000*** -12.609 0.000*** -6.904*** Test statistics of ADF, PP and DF-GLS denotes for Augmented Dickey-Fuller, Phillips- Perron and Elliott-Rothenberg-Stock DF-GLS, respectively. In each test, the null hypothesis suggests the presence of unit root in the series. The higher negative value indicates the stronger the rejection of the null hypothesis at some level of significance. Test critical values for ADF and PP are -3.44 (1%), -2.87 (5%) and -2.57 (10%). Test criticalUniversity values for DF-GLS are -2.58 (1%), -1.95 (5%) and -1.62 (10%). Lags in the ADF and DF-GLS tests are chosen with Schwarz information criterion.

* Denotes significance at the 10% level. ** Denotes significance at the 5% level. *** Denotes significance at the 1% level.

181

Table 4.3: Unit Root Tests: Transformed Variables for Stationarity

Variable ADF PP DF-GLS t-Statistic Prob. Adj. t-Stat Prob. t-Statistic SP500 - - - - - RT100 -18.318 0.000*** -18.346 0.000*** -17.178*** DVD (g) -5.146 0.000*** -4.659 0.000*** -4.141*** EAR (g) -6.603 0.000*** -9.224 0.000*** -5.195*** RDVD (g) -5.040 0.000*** -7.214 0.000*** -4.154*** REAR (g) -6.699 0.000*** -9.266 0.000*** -5.247*** CAPE (∆) -19.070 0.000*** -19.285 0.000*** -2.111** INFL -3.017 0.034** -12.339 0.000*** -1.635* TB3M (∆) -5.813 0.000*** -16.722 0.000*** -1.633** TN5Y (∆) -17.510 0.000*** -16.070 0.000*** -17.453*** TN10Y (∆) -17.480 0.000*** -17.310 0.000*** -16.972** S5Y3M -4.936 0.000*** -4.425 0.000*** -4.205*** S10Y3M -4.237 0.001*** -3.944 0.002*** -3.122*** M1 (g) -5.312 0.000*** -22.390 0.000*** -2.471** M2 (g) -6.276 0.000*** -14.182 0.000*** -6.156*** RM1 (g) -5.539 0.000*** -19.187 0.000*** -3.138*** RM2 (g) -11.746 0.000*** Malaya-11.855 0.000*** -11.749*** ISMI -5.232 0.000*** -6.405 0.000*** -3.404*** COPE (∆) -7.655 0.000*** -37.761 0.000*** -7.643*** NPHP (∆) -26.599 0.000*** -26.472 0.000*** -2.061** CMUO -4.391 0.000***of -4.182 0.001*** -4.339*** PMI -5.829 0.000*** -5.258 0.000*** -3.276*** MSCI -3.745 0.004*** -3.591 0.006*** -2.994*** RGDPG -6.343 0.000*** -8.390 0.000*** -4.771*** CDLI (∆) -5.735 0.000*** -13.743 0.000*** -5.629*** CFNAI -6.985 0.000*** -12.609 0.000*** -6.904*** Test statistics of ADF, PP and DF-GLS denotes for Augmented Dickey-Fuller, Phillips- Perron and Elliott-Rothenberg-Stock DF-GLS, respectively. In each test, the null hypothesis suggests the presence of unit root in the series. The higher negative value indicates the stronger the rejection of the null hypothesis at some level of significance. Test critical values for ADF and PP are -3.44 (1%), -2.87 (5%) and -2.57 (10%). Test critical values for DF-GLS are -2.58 (1%), -1.95 (5%) and -1.62 (10%). Lags in the ADF and DF-GLS tests are chosen with Schwarz information criterion. NoteUniversity that INFL is transformed from CPI (i.e. % change of CPI) * Denotes significance at the 10% level. ** Denotes significance at the 5% level. *** Denotes significance at the 1% level.

182

Table 4.1 shows the summary statistics of S&P 500, RT100 and all the test variables used for the research. The Jarque-Bera test is a test for goodness-of-fit that combines skewness and kurtosis. The results of both pre-transformed (Table 4.1) and transformed variables

(Table 4.4) show that the null hypothesis for normality for all variables (except CMUO) is rejected. The exhibition of non-normality in economic time series is prevalent. Bai & Ng

(2005) noted that kurtosis is easily distorted by sample size as such that “sample kurtosis measure is generally unreliable and should always be viewed with caution”. Ensuring stationarity of variables on the other hand is crucial as data that has unit root problem could have varying means and variances over time that could cause spurious modelling.

Data series of the research that are diagnosed with unit root problem are either de-trended with first difference (i.e. the change between t and t-1),Malaya or transformed into growth rate (i.e. log(xt) – log(xt-1) according to appropriateness,of with the exception for CPI which is transformed into inflation rate as illustrated in Table 4.2. Table 4.3 shows the statistically significant stationarised variables (all greater than 1% critical value for at least one test -

Augmented Dickey-Fuller, Phillips-Perron or Elliott-Rothenberg-Stock DF-GLS) that are used as the test variables for the subsequent probit model in the research.

Lags in the ADF and DF-GLS tests are chosen with Schwarz information criterion

(SIC). Koehler & Murphee (1988) demonstrated that the SIC is more reliable compared to theUniversity Akaike information criterion (AIC) as SIC brings about lower order models for forecasting. It is noteworthy that the series of S&P 500 is not a test variable for the research. The series in its original form is used for the JLS models where stationarity is not a prerequisite. Table 4.4 in the following shows the summary statistics of the transformed variables that are free of unit root problem.

183

Table 4.4: Summary Statistics of Transformed Variables

Variable Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Prob. SP500 593.215 355.140 1947.090 67.070 525.738 0.622 1.896 65.285 0.000 RT100 0.541 0.824 11.352 -22.804 3.687 -1.015 7.043 482.795 0.000 DVD (g) 0.451 0.491 1.842 -2.316 0.597 -1.361 7.698 696.397 0.000 EAR(g) 0.519 0.723 70.386 -52.841 5.545 1.729 73.654 118218.4 0.000 RDVD (g) 0.102 0.095 2.292 -2.996 0.650 -0.536 6.022 242.915 0.000 REAR (g) 0.170 0.282 70.334 -51.788 5.512 2.084 74.037 119629.6 0.000 CAPE (∆) 0.009 0.060 3.600 -3.980 0.753 -0.797 7.853 616.537 0.000 INFL 0.339 0.295 1.774 -1.953 0.361 Malaya-0.097 6.770 336.120 0.000 TB3M (∆) -0.007 0.000 2.610 -4.620 0.471 -1.681 26.482 13293.890 0.000 TN5Y (∆) -0.005 -0.010 1.860 -2.030 0.353 -0.375 8.529 735.461 0.000 TN10Y (∆) -0.003 -0.010 1.610 -1.760 0.305 -0.406 8.001 606.332 0.000 S5Y3M 1.290 1.350 4.330 -2.250 of1.010 -0.373 3.163 13.760 0.001 S10Y3M 1.647 1.740 4.420 -2.650 1.285 -0.472 2.708 23.046 0.000 M1 (g) 0.491 0.472 5.755 -3.374 0.721 1.538 15.106 3685.762 0.000 M2 (g) 0.554 0.527 2.763 -0.561 0.376 0.911 6.539 374.245 0.000 RM1 (g) 0.143 0.076 6.592 -3.090 0.817 1.811 14.586 3481.422 0.000 RM2 (g) 0.206 0.183 3.032 -1.517 0.497 0.875 6.642 385.751 0.000 ISMI 46.542 46.250 66.500 24.600 6.305 -0.057 4.017 24.721 0.000 COPE (∆) 60.540 104.000 13291.000 -10906.000 2631.428 -0.011 5.390 134.943 0.000 NPHP (∆) 0.032 1.000 326.000 -419.000 82.150 -0.304 5.683 178.094 0.000 CMUO 2.980 2.850 32.100 -26.700 9.616 0.108 3.312 3.386 0.184 PMI 52.371 53.000 72.100 29.400 6.521 -0.552 4.214 63.526 0.000 MSCI 84.522 87.000 112.000 51.700 12.503 -0.282 2.449 14.699 0.001 RGDPG 2.867 3.000 16.500 -8.200 3.373 -0.134 4.940 90.497 0.000 CDLI (∆) 0.104 0.200 1.600 -3.000 0.539 -1.445 7.719 723.639 0.000 CFNAI -0.002 0.080University 2.680 -4.990 1.005 -1.137 6.347 386.157 0.000

184

4.3 Findings of Models for Stock Market Declines

Results of models for the identification of stock market declines (and vice versa) in this section are illustrated in charts. The section begins with Figure 4.1 that shows the time series of the Markov-switching’s filtered probabilities for bear markets. Figure

4.2 to Figure 4.8 exhibit charts of the timeframes for stock market declines identified with the respective models as elaborated earlier against the back drop of the S&P 500 index time series (S&P 500). The identified periods of stock market declines are depicted in grey scale (except stated otherwise i.e. the JLS “negative bubbles” rebounds). Discussions on the results are featured in the subsequent chapter.

4.3.1 Markov-switching Summary Outputs Malaya Table 4.5: Summary ofof Markov-switching Results Parameters of Markov-switching Model ퟎퟎ ퟏퟏ 흁ퟎ 흁ퟏ 𝝈ퟎ 𝝈ퟏ 𝝆 𝝆 LogLik Estimates 1.133 -1.226 2.740 5.135 0.95 0.86 -1487.064 (0.00) (0.04) (0.00) (0.00) (0.00) (0.56) Expected duration of 0 (bull markets): 21.04 time periods Expected duration of 1 (bear markets): 7.02 time periods

Table 4.5 illustrates the summary of outputs for the Markov-switching model. Figure

4.1 exhibits three line charts in sequence i.e. the time series of RT100, the filtered probabilityUniversity of the Markov-switching model and the smoothed probability of the model.

The results are almost parallel to that of Chen (2009) who investigated the S&P 500 from February 1957 to December 2007 (compared to the research’s period of observation which is from April 1967 to June 2014).

185

Chen (2009) denoted the bear market regime as “0” and bull market regime as

“1”. Contrarily, the research denotes the bear market regime as “1” and bull market regime as “0” to standardise with other models of stock market declines. Thus, the

00 estimates of휇0 , 휎0 and 푝 of the research are parallel to Chen’s (2009) 휇1 , 휎1 and

푝11and vice versa.

Comparing the estimates of the research with Chen’s (2009) in bracket [ ], the results read as 휇0 = 1.133 [1.13]; 휇1 = -1.226 [1.26]; 휎0 = 2.740 [2.53]; 휎1 = 5.135

[5.17]; 푝00 = 0.95 [0.95]; 푝11 = 0.86 [0.85]; LogLik = -1487.064 [-1581.22]. The duration of bear market is about 1/3 of the duration of bull market i.e. the expected duration of bear regime is 7.02 time periods compared to the expected duration of bull regime, which is 21.04 time periods. Malaya of

University

186

Malaya of

Figure 4.1: Markov-switching Results University

187

4.3.2 Markov-switching Filtered Probabilities Result

Malaya of

UniversityFigure 4.2: Markov-switching Filtered Probabilities Bear Markets

188

4.3.3 Markov-switching Smoothed Probabilities Result

Malaya of

UniversityFigure 4.3: Markov-switching Smoothed Probabilities Bear Markets

189

4.3.4 Markov-switching Dichotomised Smoothed Probabilities Result

Malaya of

Figure 4.4: S&PUniversity 500 Index vs. Markov-switching Dichotomised Smoothed Probabilities Bear Markets

190

4.3.5 Naïve Moving Average Result

Malaya of

UniversityFigure 4.5: S&P 500 Index vs. Naïve Moving Average Bear Markets

191

4.3.6 Naïve Moving Average Negative Return Result

Malaya of

Figure University4.6: S&P 500 Index vs. Naïve Moving Average Negative Return Bear Markets

192

4.3.7 Lunde & Timmermann’s B-B Algorithm Result

Malaya of

Figure University4.7: S&P 500 Index vs. Lunde & Timmermann’s B-B Algorithm Bear Markets

193

4.3.8 Candelon, Piplak & Straetmans’ B-B Algorithm Result

Malaya of

Figure 4.8:University S&P 500 Index vs. Candelon, Piplak & Straetmans’ B-B Algorithm Bear Markets

194

4.3.9 JLS’ Crashes Result

Malaya of

FigureUniversity 4.9: S&P 500 Index vs. JLS’ Crashes (Turning Points and Limited Windows)

195

4.3.10 JLS’ Negative Bubbles Rebounds Result

Malaya of

Figure 4.10: S&PUniversity 500 Index vs. JLS’ Negative Bubbles Rebounds (Turning Points and Limited Windows)

196

Figure 4.1, 4.2 and 4.3 show the Markov-switching results of the research. The outputs of the filtered and smoothed probabilities illustrated in charts are fuzzy and the switching of the heterogeneous regimes is not easy to observe visually. Figure 4.4, 4.5, 4.6, 4.7 and 4.8 of the semi-parametric and non-parametric approaches on the other hand illustrate a more discernible contrast of dual-regime durations. Likewise for the scale-invariant outputs that are presented in table 4.9 and 4.10.

As illustrated earlier in Table 4.5, the Markov-switching model has identified 7.02 time periods of bear regime (compared to 21.04 times of bull regime) over the research timeline. The sensitivity level of the semi-parametric and non-parametric models based on the number of times the bear regimes is identified, is listed in sequence in the followings: 1) naïve moving average (27 times); 2) Candelon, PiplakMalaya & Straetmans’ B-B algorithm (18 times); 3) naïve moving average negative return (14 times); 4) Markov-switching dichotomised smoothed probabilities (13 times);of 5) Lunde & Timmermann’s B-B algorithm (7 times). The JLS model shows 6 crashes while the JLS “negative bubbles” model shows

13 rebounds.

Although the results across models show varying degree of sensitivity, it is important to note that the indication of greater sensitivity does not imply superiority. The primary objective of studies in heterogeneous regimes switching is to identify distinctly the latent long-run structure of the market. A more sensitive model (i.e. one with less rigid specification)University is conceivably more ideal for relationship test between multiple macro variables as oppose to test on a single variable as the former would yield higher statistical significance compared to the later. The specification of models used in this area of study in the past studies thus were discernibly calibrated in congruence to the motivation of studies

(Ho, Estrada Ruiz & Yap, 2017). 197

4.4 Observations for Out-of-sample Test

Table 4.6: In-sample Observations & Out-of-sample Observations for Out-of-sample Test (Clark & West, 2007)

In-sample Observations (R) Out-of-sample Observations (P) Start Date End data Total Obs Start Date End data Total Obs k = 1 1967.05 1982.12 189 1983.01 2014.05 377 k = 3 1967.07 1982.11 188 1982.12 2014.03 376 k = 6 1967.10 1982.10 187 1982.11 2013.12 374 k = 9 1968.01 1982.09 186 1982.10 2013.09 372 k = 12 1968.04 1982.08 185 1982.09 2013.06 370 *Ratio of P/R ≈ 2

As elaborated earlier, the research employed the ‘MSPE-adj’ statistics for the nested model which was proven by Clark & West (2007) to be as equally accurate as other MSPE models. On top of that, the improved model is more parsimonious, more efficient, has equal or better power and can be more easily interpreted. TheMalaya test requires the population sample of T observations to be divided into an in-sampleof and an out-of-sample section. The total in-sample observations is denoted as R, 푡 = 1, … , 푅 and 푃 is the total out-of-sample observations, 푡=푅+1,…,푅+푃; and therefore, 푅 + 푃 = 푇. The estimation procedure is recursive.

The ratio of P/R ≈ 2 effectively yields one portion of observations for the modeling of models for predictive test and two remaining portions of observations for out-of-sample test. The MSPE-adj statistics is applicable exclusively on the testing of parametric models. Nonetheless,University the apportioning of the 푅 + 푃 = 푇 observations for all test horizon as illustrated in Table 4.6 is applied for all other semi-parametric and non-parametric models for predictive test using the probit model. As a recap, the output derived from the probit model is then evaluated with the quadratic probability score (QPS) (see Chen, 2009 and

Diebold & Rudebusch, 1989).

198

As for the in-sample predictability test, the P/R ratio is not applied. Full observation is used to estimate the model parameters for the predictive regression model. The in-sample analysis which is a back-testing scheme repeats the same observation for predictability test, adjusted for test horizons.

4.5 In-sample Predictability Test Results

The overall sample of observations for the data set is denoted as T as explained previously, while k (1, 3, 6, 9, 12) as featured in Table 4.7 to Table 4.14 denotes the forecasting horizons. As a recap to Chapter 3, the in-sample predictability test results are derived from a predictive regression model denoted as 푃(퐷푡+푘 =Malaya 1) = 퐹(∝ +훽푥푡), where 퐷푡+푘 = 1 represents accordingly either the switching probabilities (for parametric models i.e. filtered or smoothed for Markov-switching models) ofor a dummy variable (for semi-parametric and non-parametric models) to which a regime is 퐷푡+푘 = 1 for bear stock market and 퐷푡+푘 = 0 for bull stock market.

Entries in bold reflect a significance level of 5%; 0.000 reflects the value is lesser than 0.0005. Table 4.8 and Table 4.9 show the in-sample predictability test results for parametric models. Table 4.10 to Table 4.11 illustrate the comparable results for semi- parametric models followed by the results for non-parametric models in Table 4.12 to Table 4.13.University In-sample predictability test results for the LPPL models are shown in Table 4.14.

199

4.5.1 Markov-switching Filtered Probabilities In-sample Predictability Test Results

Table 4.7: In-sample Predictability Test Results for Parametric Model: Markov-switching Filtered Probabilities

휷̂ t-stat p-value 푹ퟐ DVD (g) k = 1 0.114 5.807 0.000 0.056 k = 3 0.086 4.290 0.000 0.032 k = 6 0.054 2.664 0.008 0.013 k = 9 0.038 1.844 0.066 0.006 k = 12 0.022 1.087 0.277 0.002 EAR (g) k = 1 0.012 5.599 0.000 0.053 k = 3 0.013 6.098 0.000 0.062 k = 6 0.008 3.777 0.000 0.025 k = 9 0.003 1.491 0.137 0.004 k = 12 0.005 2.075 0.038 0.008 RDVD (g) k = 1 0.117 6.504 0.000 0.070 k = 3 0.106 5.825 0.000 0.057 k = 6 0.083 4.488 Malaya0.000 0.035 k = 9 0.069 3.700 0.000 0.024 k = 12 0.058 3.072 0.002 0.017 REAR (g) of k = 1 0.012 5.768 0.000 0.056 k = 3 0.013 6.362 0.000 0.067 k = 6 0.009 4.049 0.000 0.028 k = 9 0.004 1.735 0.083 0.005 k = 12 0.005 2.333 0.020 0.010 CAPE (∆) k = 1 0.117 7.648 0.000 0.094 k = 3 0.094 6.027 0.000 0.061 k = 6 0.085 5.432 0.000 0.050 k = 9 0.046 2.849 0.005 0.014 k = 12 0.041 2.530 0.012 0.011 TB3M (∆) k = 1 0.081 3.200 0.001 0.018 k = 3 0.099 3.914 0.000 0.027 k = University6 0.006 0.238 0.812 0.000 k = 9 -0.003 -0.102 0.919 0.000 k = 12 0.027 1.054 0.292 0.002

200

Table 4.7, continued (i)

휷̂ t-stat p-value 푹ퟐ TN5Y (∆) k = 1 0.053 1.561 0.119 0.004 k = 3 0.092 2.712 0.007 0.013 k = 6 -0.075 -2.194 0.029 0.009 k = 9 -0.041 -1.209 0.227 0.003 k = 12 -0.015 -0.448 0.654 0.000 TN10Y (∆) k = 1 0.038 0.970 0.332 0.002 k = 3 0.079 1.991 0.047 0.007 k = 6 -0.118 -2.998 0.003 0.016 k = 9 -0.056 -1.412 0.159 0.004 k = 12 -0.034 -0.844 0.399 0.001 S5Y3M k = 1 0.011 0.929 0.353 0.002 k = 3 0.033 2.811 0.005 0.014 k = 6 0.058 4.980 0.000 0.042 k = 9 0.091 8.034 0.000 0.104 k = 12 0.103 9.149 0.000 0.131 S10Y3M k = 1 0.011 1.192 Malaya0.234 0.003 k = 3 0.029 3.134 0.002 0.017 k = 6 0.050 5.416 0.000 0.050 k = 9 0.073 8.193of 0.000 0.108 k = 12 0.083 9.347 0.000 0.136 M1 (g) k = 1 -0.010 -0.624 0.533 0.001 k = 3 0.010 0.584 0.559 0.001 k = 6 0.017 1.028 0.304 0.002 k = 9 0.028 1.620 0.106 0.005 k = 12 0.009 0.555 0.579 0.001 M2 (g) k = 1 -0.080 -2.508 0.012 0.011 k = 3 -0.021 -0.666 0.506 0.001 k = 6 -0.042 -1.305 0.192 0.003 k = 9 -0.009 -0.282 0.778 0.000 k = 12 -0.042 -1.308 0.191 0.003 RM1University (g) k = 1 0.004 0.285 0.776 0.000 k = 3 0.028 1.906 0.057 0.006 k = 6 0.033 2.238 0.026 0.009 k = 9 0.047 3.171 0.002 0.018 k = 12 0.034 2.268 0.024 0.009

201

Table 4.7, continued (ii)

휷̂ t-stat p-value 푹ퟐ RM2 (g) k = 1 -0.012 -0.514 0.608 0.000 k = 3 0.043 1.772 0.077 0.006 k = 6 0.029 1.170 0.243 0.002 k = 9 0.063 2.599 0.010 0.012 k = 12 0.047 1.908 0.057 0.007 INFL k = 1 0.052 1.557 0.120 0.004 k = 3 0.006 0.192 0.848 0.000 k = 6 -0.022 -0.655 0.512 0.001 k = 9 -0.002 -0.074 0.941 0.000 k = 12 0.032 0.939 0.348 0.002 RGDPG k = 1 0.030 8.799 0.000 0.121 k = 3 0.031 9.213 0.000 0.131 k = 6 0.022 6.393 0.000 0.068 k = 9 0.017 4.850 0.000 0.041 k = 12 0.012 3.386 0.001 0.020 ISMI k = 1 0.009 4.870 Malaya0.000 0.040 k = 3 0.005 2.643 0.008 0.012 k = 6 0.001 0.415 0.678 0.000 k = 9 -0.003 -1.698of 0.090 0.005 k = 12 -0.005 -2.797 0.005 0.014 COPE (∆) k = 1 0.000 1.757 0.079 0.005 k = 3 0.000 2.092 0.037 0.008 k = 6 0.000 1.476 0.141 0.004 k = 9 0.000 1.110 0.267 0.002 k = 12 0.000 0.421 0.674 0.000 NPHP (∆) k = 1 0.000 2.235 0.026 0.009 k = 3 0.000 2.951 0.003 0.015 k = 6 0.000 3.402 0.001 0.020 k = 9 0.001 3.908 0.000 0.027 k = 12 0.000 3.249 0.001 0.019 CMUOUniversity k = 1 0.009 7.775 0.000 0.097 k = 3 0.007 5.505 0.000 0.051 k = 6 0.003 2.071 0.039 0.008 k = 9 -0.002 -1.375 0.170 0.003 k = 12 -0.006 -4.399 0.000 0.034

202

Table 4.7, continued (iii)

휷̂ t-stat p-value 푹ퟐ PMI k = 1 0.016 9.291 0.000 0.133 k = 3 0.013 7.047 0.000 0.081 k = 6 0.007 3.825 0.000 0.025 k = 9 0.003 1.595 0.111 0.005 k = 12 0.000 0.040 0.968 0.000 MSCI k = 1 0.008 9.041 0.000 0.127 k = 3 0.007 7.149 0.000 0.083 k = 6 0.005 4.755 0.000 0.039 k = 9 0.003 2.836 0.005 0.014 k = 12 0.001 0.909 0.364 0.001 CDLI (∆) k = 1 0.250 12.624 0.000 0.220 k = 3 0.236 11.711 0.000 0.196 k = 6 0.208 9.985 0.000 0.151 k = 9 0.171 7.982 Malaya0.000 0.103 k = 12 0.125 5.642 0.000 0.054 CFNAI k = 1 0.124 11.450of 0.000 0.189 k = 3 0.109 9.795 0.000 0.146 k = 6 0.075 6.430 0.000 0.069 k = 9 0.049 4.136 0.000 0.030 k = 12 0.024 2.020 0.044 0.007

Note: Entries in bold indicate significance at the 5% level.

The in-sample predictability test for predicting bear stock markets of the Markov- switching model with filtered probabilities shows that variables that are significant for all horizons (i.e. k = 1, 3, 6, 9, 12) include RGDPG, NPHP (∆), CDLI (∆) and CFNAI. Overall,

CDLIUniversity (∆) has the highest 푅2across all horizons compared to other variables with the values

2 2 2 2 2 of 푅푘=1 = 0.220, 푅푘=3 = 0.196, 푅푘=6 = 0.151, 푅푘=9 = 0.103 and 푅푘=12 = 0.054.

203

4.5.2 Markov-switching Smoothed Probabilities In-sample Predictability Test Results

Table 4.8: In-sample Predictability Test Results for Parametric Model: Markov-switching Smoothed Probabilities

휷̂ t-stat p-value 푹ퟐ DVD (g) k = 1 0.121 5.264 0.000 0.047 k = 3 0.090 3.873 0.000 0.026 k = 6 0.062 2.642 0.008 0.012 k = 9 0.041 1.742 0.082 0.005 k = 12 0.020 0.846 0.398 0.001 EAR (g) k = 1 0.017 6.867 0.000 0.077 k = 3 0.015 5.929 0.000 0.059 k = 6 0.007 2.824 0.005 0.014 k = 9 0.005 2.109 0.035 0.008 k = 12 0.007 2.703 Malaya0.007 0.013 RDVD (g) k = 1 0.137 6.569 0.000 0.071 k = 3 0.120 5.684of 0.000 0.054 k = 6 0.100 4.673 0.000 0.038 k = 9 0.086 4.001 0.000 0.028 k = 12 0.066 3.035 0.003 0.016 REAR (g) k = 1 0.017 7.122 0.000 0.083 k = 3 0.015 6.227 0.000 0.065 k = 6 0.008 3.104 0.002 0.017 k = 9 0.006 2.403 0.017 0.010 k = 12 0.008 2.987 0.003 0.016 CAPE (∆) k = 1 0.146 8.253 0.000 0.108 k = 3 0.123 6.820 0.000 0.076 k = 6 0.101 5.551 0.000 0.052 k = 9 0.069 3.731 0.000 0.024 k = University12 0.061 3.245 0.001 0.019 TB3M (∆) k = 1 0.111 3.762 0.000 0.024 k = 3 0.078 2.621 0.009 0.012 k = 6 0.028 0.952 0.342 0.002 k = 9 0.007 0.218 0.827 0.000 k = 12 -0.007 -0.229 0.819 0.000

204

Table 4.8, continued (i)

휷̂ t-stat p-value 푹ퟐ TN5Y (∆) k = 1 0.068 1.712 0.087 0.005 k = 3 0.039 0.968 0.334 0.002 k = 6 -0.054 -1.361 0.174 0.003 k = 9 -0.036 -0.911 0.363 0.001 k = 12 -0.047 -1.183 0.237 0.003 TN10Y (∆) k = 1 0.043 0.925 0.356 0.002 k = 3 0.007 0.142 0.887 0.000 k = 6 -0.096 -2.080 0.038 0.008 k = 9 -0.061 -1.325 0.186 0.003 k = 12 -0.067 -1.435 0.152 0.004 S5Y3M k = 1 0.036 2.564 0.011 0.012 k = 3 0.066 4.841 0.000 0.040 k = 6 0.103 7.748 0.000 0.097 k = 9 0.137 10.768 0.000 0.173 k = 12 0.154 12.359 0.000 0.216 S10Y3M k = 1 0.032 2.907 Malaya0.004 0.015 k = 3 0.056 5.272 0.000 0.047 k = 6 0.085 8.223 0.000 0.108 k = 9 0.110 11.039of 0.000 0.180 k = 12 0.123 12.610 0.000 0.223 M1 (g) k = 1 0.004 0.182 0.856 0.000 k = 3 0.031 1.589 0.113 0.004 k = 6 0.033 1.705 0.089 0.005 k = 9 0.033 1.660 0.097 0.005 k = 12 0.025 1.257 0.209 0.003 M2 (g) k = 1 -0.063 -1.693 0.091 0.005 k = 3 -0.010 -0.279 0.780 0.000 k = 6 -0.026 -0.683 0.495 0.001 k = 9 -0.009 -0.230 0.819 0.000 k = 12 -0.062 -1.646 0.100 0.005 RM1University (g) k = 1 0.024 1.417 0.157 0.004 k = 3 0.052 3.031 0.003 0.016 k = 6 0.054 3.156 0.002 0.018 k = 9 0.060 3.478 0.001 0.021 k = 12 0.051 2.935 0.003 0.015

205

Table 4.8, continued (ii)

휷̂ t-stat p-value 푹ퟐ RM2 (g) k = 1 0.022 0.781 0.435 0.001 k = 3 0.069 2.439 0.015 0.010 k = 6 0.061 2.148 0.032 0.008 k = 9 0.087 3.080 0.002 0.017 k = 12 0.049 1.704 0.089 0.005 INFL k = 1 0.038 0.974 0.330 0.002 k = 3 -0.001 -0.036 0.972 0.000 k = 6 -0.006 -0.158 0.875 0.000 k = 9 0.025 0.635 0.526 0.001 k = 12 0.047 1.198 0.232 0.003 RGDPG k = 1 0.042 10.987 0.000 0.176 k = 3 0.039 10.194 0.000 0.156 k = 6 0.030 7.364 0.000 0.088 k = 9 0.020 4.773 0.000 0.039 k = 12 0.010 2.387 0.017 0.010 ISMI k = 1 0.010 4.454 Malaya0.000 0.034 k = 3 0.005 2.183 0.029 0.008 k = 6 -0.001 -0.336 0.737 0.000 k = 9 -0.005 -2.348of 0.019 0.010 k = 12 -0.008 -3.604 0.000 0.023 COPE (∆) k = 1 0.000 2.251 0.025 0.009 k = 3 0.000 2.245 0.025 0.009 k = 6 0.000 1.554 0.121 0.004 k = 9 0.000 1.058 0.291 0.002 k = 12 0.000 0.462 0.644 0.000 NPHP (∆) k = 1 0.000 2.866 0.004 0.014 k = 3 0.001 3.656 0.000 0.023 k = 6 0.001 4.355 0.000 0.033 k = 9 0.001 3.970 0.000 0.028 k = 12 0.001 3.154 0.002 0.018 CMUOUniversity k = 1 0.011 7.613 0.000 0.093 k = 3 0.007 4.713 0.000 0.038 k = 6 0.001 0.711 0.477 0.001 k = 9 -0.004 -2.951 0.003 0.015 k = 12 -0.009 -6.165 0.000 0.064

206

Table 4.8, continued (iii)

휷̂ t-stat p-value 푹ퟐ PMI k = 1 0.020 9.881 0.000 0.148 k = 3 0.015 7.203 0.000 0.085 k = 6 0.008 3.720 0.000 0.024 k = 9 0.002 1.048 0.295 0.002 k = 12 -0.002 -0.705 0.481 0.001 MSCI k = 1 0.010 9.145 0.000 0.129 k = 3 0.007 6.881 0.000 0.078 k = 6 0.004 4.009 0.000 0.028 k = 9 0.002 1.697 0.090 0.005 k = 12 -0.001 -0.484 0.628 0.000 CDLI (∆) k = 1 0.328 14.811 0.000 0.280 k = 3 0.309 13.622 0.000 0.248 k = 6 0.266 11.178 0.000 0.183 k = 9 0.206 8.280 0.000 0.110 k = 12 0.164 6.392 0.000 0.069 CFNAI k = 1 0.157 12.680 Malaya0.000 0.222 k = 3 0.134 10.432 0.000 0.162 k = 6 0.089 6.576 0.000 0.072 k = 9 0.052 3.761of 0.000 0.025 k = 12 0.019 1.373 0.170 0.003

Note: Entries in bold indicate significance at the 5% level.

The in-sample predictability test for predicting bear stock markets of the Markov- switching model with smoothed probabilities shows that variables that are significant for all horizons (i.e. k = 1, 3, 6, 9, 12) include EAR (g), RDVD (g), REAR (g), CAPE (∆),

S5Y3M, S10Y3M, RGDPG, NPHP (∆) and CDLI (∆). Overall, CDLI (∆) has the highest 푅2acrossUniversity the k = 1, k = 3 and k = 6 horizons compared to other variables with the values of 2 2 2 2 푅푘=1 = 0.280, 푅푘=3 = 0.248, 푅푘=6 = 0.151; while S10Y3M has the highest 푅 across the k =

2 2 9 and k = 12 horizons of 푅푘=9 = 0.180 and 푅푘=12 = 0.223.

207

4.5.3 Markov-switching Dichotomised Smoothed Probabilities Predictability Test Results

Table 4.9: In-sample Predictability Test Results for Semi-parametric Model: Markov- switching Dichotomised Smoothed Probabilities

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 DVD (g) k = 1 -0.354 -3.633 0.000 0.022 0.023 k = 3 -0.262 -2.694 0.007 0.012 0.013 k = 6 -0.202 -2.050 0.040 0.007 0.007 k = 9 -0.149 -1.503 0.133 0.004 0.004 k = 12 -0.059 -0.591 0.555 0.001 0.001 EAR (g) k = 1 -0.232 -7.112 0.000 0.139 0.142 k = 3 -0.123 -6.005 0.000 0.087 0.089 k = 6 -0.029 -2.632 0.009 0.012 0.013 k = 9 -0.025 -2.063 0.039 0.008 0.008 k = 12 -0.026 -2.116 0.034 0.008 0.008 RDVD (g) k = 1 -0.407 -4.353 0.000 0.033Malaya 0.034 k = 3 -0.370 -3.969 0.000 0.027 0.028 k = 6 -0.302 -3.242 0.001 0.018 0.019 k = 9 -0.359 -3.815 0.000of 0.026 0.027 k = 12 -0.248 -2.654 0.008 0.012 0.013 REAR (g) k = 1 -0.251 -7.508 0.000 0.153 0.157 k = 3 -0.133 -6.393 0.000 0.097 0.100 k = 6 -0.031 -2.827 0.005 0.014 0.014 k = 9 -0.029 -2.407 0.016 0.010 0.011 k = 12 -0.030 -2.424 0.015 0.011 0.011 CAPE (∆) k = 1 -0.562 -6.575 0.000 0.084 0.087 k = 3 -0.466 -5.612 0.000 0.059 0.061 k = 6 -0.379 -4.739 0.000 0.041 0.042 k = 9 -0.285 -3.666 0.000 0.024 0.025 k = 12 -0.272 -3.499 0.001 0.022 0.022 TB3MUniversity (∆) k = 1 -0.393 -3.325 0.001 0.019 0.020 k = 3 -0.291 -2.482 0.013 0.010 0.011 k = 6 -0.161 -1.377 0.169 0.003 0.003 k = 9 -0.044 -0.369 0.712 0.000 0.000 k = 12 0.096 0.763 0.445 0.001 0.001

208

Table 4.9, continued (i)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 TN5Y (∆) k = 1 -0.225 -1.399 0.162 0.003 0.003 k = 3 -0.217 -1.314 0.189 0.003 0.003 k = 6 0.142 0.862 0.388 0.001 0.001 k = 9 0.086 0.518 0.604 0.000 0.000 k = 12 0.318 1.831 0.067 0.006 0.006 TN10Y (∆) k = 1 -0.115 -0.614 0.539 0.001 0.001 k = 3 -0.116 -0.606 0.545 0.001 0.001 k = 6 0.330 1.703 0.089 0.005 0.005 k = 9 0.179 0.918 0.359 0.001 0.002 k = 12 0.375 1.842 0.066 0.006 0.006 S5Y3M k = 1 -0.079 -1.373 0.170 0.003 0.003 k = 3 -0.215 -3.662 0.000 0.023 0.024 k = 6 -0.361 -5.968 0.000 0.064 0.066 k = 9 -0.552 -8.403 0.000 0.138 0.142 k = 12 -0.656 -9.280 0.000 0.179Malaya 0.185 S10Y3M k = 1 -0.068 -1.505 0.132 0.004 0.004 k = 3 -0.173 -3.790 0.000 0.025 0.026 k = 6 -0.290 -6.178 0.000of 0.068 0.071 k = 9 -0.432 -8.494 0.000 0.139 0.144 k = 12 -0.503 -9.369 0.000 0.177 0.183 M1 (g) k = 1 -0.018 -0.234 0.815 0.000 0.000 k = 3 -0.096 -1.217 0.224 0.003 0.003 k = 6 -0.125 -1.490 0.136 0.004 0.004 k = 9 -0.168 -1.938 0.053 0.007 0.007 k = 12 -0.068 -0.815 0.415 0.001 0.001 M2 (g) k = 1 0.239 1.550 0.121 0.004 0.004 k = 3 0.037 0.239 0.811 0.000 0.000 k = 6 -0.055 -0.343 0.732 0.000 0.000 k = 9 -0.130 -0.803 0.422 0.001 0.001 k = University12 0.172 1.084 0.278 0.002 0.002

209

Table 4.9, continued (ii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 RM1 (g) k = 1 -0.066 -0.968 0.333 0.002 0.002 k = 3 -0.150 -2.186 0.029 0.008 0.009 k = 6 -0.183 -2.428 0.015 0.011 0.011 k = 9 -0.302 -3.661 0.000 0.025 0.026 k = 12 -0.194 -2.447 0.014 0.011 0.011 RM2 (g) k = 1 -0.017 -0.150 0.881 0.000 0.000 k = 3 -0.194 -1.704 0.088 0.005 0.005 k = 6 -0.246 -2.065 0.039 0.007 0.008 k = 9 -0.483 -3.719 0.000 0.025 0.026 k = 12 -0.251 -2.017 0.044 0.007 0.008 INFL k = 1 -0.126 -0.771 0.441 0.001 0.001 k = 3 -0.024 -0.148 0.882 0.000 0.000 k = 6 -0.026 -0.156 0.876 0.000 0.000 k = 9 -0.133 -0.795 0.427 0.001 0.001 k = 12 -0.164 -0.967 0.334 0.002Malaya 0.002 RGDPG k = 1 -0.198 -8.999 0.000 0.169 0.174 k = 3 -0.215 -9.362 0.000 0.189 0.194 k = 6 -0.127 -6.706 0.000of 0.084 0.087 k = 9 -0.077 -4.261 0.000 0.032 0.033 k = 12 -0.032 -1.805 0.071 0.006 0.006 ISMI k = 1 -0.042 -4.566 0.000 0.037 0.038 k = 3 -0.021 -2.312 0.021 0.009 0.010 k = 6 0.001 0.101 0.920 0.000 0.000 k = 9 0.023 2.414 0.016 0.010 0.011 k = 12 0.030 3.310 0.001 0.016 0.020 COPE (∆) k = 1 0.000 -2.492 0.013 0.011 0.011 k = 3 0.000 -2.391 0.017 0.010 0.010 k = 6 0.000 -1.676 0.094 0.005 0.005 k = 9 0.000 -1.024 0.306 0.002 0.002 k = University12 0.000 -0.236 0.813 0.000 0.000

210

Table 4.9, continued (iii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 NPHP (∆) k = 1 -0.002 -2.367 0.018 0.010 0.010 k = 3 -0.002 -2.849 0.004 0.014 0.015 k = 6 -0.003 -4.538 0.000 0.037 0.039 k = 9 -0.003 -3.945 0.000 0.028 0.029 k = 12 -0.002 -2.526 0.012 0.011 0.012 CMUO k = 1 -0.046 -6.889 0.000 0.090 0.092 k = 3 -0.029 -4.557 0.000 0.037 0.038 k = 6 -0.007 -1.148 0.251 0.002 0.002 k = 9 0.014 2.286 0.022 0.009 0.009 k = 12 0.035 5.363 0.000 0.053 0.055 PMI k = 1 -0.084 -8.558 0.000 0.141 0.145 k = 3 -0.066 -7.045 0.000 0.092 0.094 k = 6 -0.036 -3.997 0.000 0.028 0.029 k = 9 -0.008 -0.883 0.377 0.001 0.001 k = 12 0.011 1.223 0.221 0.003Malaya 0.003 MSCI k = 1 -0.038 -7.592 0.000 0.107 0.110 k = 3 -0.029 -5.956 0.000 0.063 0.065 k = 6 -0.017 -3.575 0.000of 0.022 0.023 k = 9 -0.007 -1.456 0.146 0.004 0.004 k = 12 0.004 0.778 0.437 0.001 0.001 CDLI (∆) k = 1 -1.413 -9.693 0.000 0.218 0.224 k = 3 -1.229 -9.258 0.000 0.186 0.191 k = 6 -1.037 -8.514 0.000 0.144 0.149 k = 9 -0.751 -6.962 0.000 0.086 0.089 k = 12 -0.580 -5.427 0.000 0.051 0.053 CFNAI k = 1 -0.671 -9.275 0.000 0.187 0.192 k = 3 -0.596 -8.658 0.000 0.155 0.159 k = 6 -0.400 -6.555 0.000 0.079 0.081 k = 9 -0.211 -3.637 0.000 0.023 0.024 k = University12 -0.032 -0.532 0.595 0.000 0.001

Note: Entries in bold indicate significance at the 5% level.

211

The in-sample predictability test for predicting bear stock markets of the Markov- switching model with dichotomised smoothed probabilities shows that variables that are significant for all horizons (i.e. k = 1, 3, 6, 9, 12) include EAR (g), RDVD (g), REAR (g),

CAPE (∆), NPHP (∆) and CDLI (∆). Overall, CDLI (∆) has the highest value for both the pseudo- 푅2(Mcfadden) and pseudo- 푅2(Diebold & Rudebusch) across the k = 1, k = 3 and k

2 = 6 horizons compared to other variables with the values of 푅푀푐푓푎푑푑푒푛,푘=1 = 0.218,

2 2 2 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=1 = 0.224; 푅푀푐푓푎푑푑푒푛,푘=3 = 0.186, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=3 = 0.191;

2 2 and 푅푀푐푓푎푑푑푒푛,푘=6 = 0.144, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=6 = 0.149; while S10Y3M has the

2 2 2 highest 푅 across the k = 9 horizon of 푅푀푐푓푎푑푑푒푛,푘=9 = 0.139 and 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=9 =

2 2 0.144. S5Y3M has the highest 푅 across the k = 12 horizon of 푅푀푐푓푎푑푑푒푛,푘=12 = 0.179 and

2 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=12 = 0.185. Malaya of

University

212

4.5.4 Naïve Moving Average In-sample Predictability Test Results

Table 4.10: In-sample Predictability Test Results for Semi-parametric Model: Naïve Moving Average

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 DVD (g) k = 1 -0.053 -0.578 0.563 0.000 0.001 k = 3 0.006 0.063 0.949 0.000 0.000 k = 6 0.025 0.271 0.786 0.000 0.000 k = 9 -0.007 -0.075 0.940 0.000 0.000 k = 12 -0.126 -1.376 0.169 0.003 0.003 EAR (g) k = 1 -0.110 -4.252 0.000 0.047 0.059 k = 3 -0.052 -3.381 0.001 0.022 0.027 k = 6 -0.012 -1.095 0.274 0.002 0.002 k = 9 0.014 1.287 0.198 0.003 0.003 k = 12 0.008 0.743 0.458 0.001 0.001 RDVD (g) k = 1 -0.170 -2.007 0.045 0.006 0.007 k = 3 -0.108 -1.279 0.201 0.002Malaya 0.003 k = 6 -0.155 -1.824 0.068 0.005 0.006 k = 9 -0.224 -2.627 0.009 0.010 0.012 k = 12 -0.233 -2.686 0.007of 0.010 0.013 REAR (g) k = 1 -0.125 -4.608 0.000 0.054 0.067 k = 3 -0.058 -3.622 0.000 0.025 0.031 k = 6 -0.015 -1.345 0.179 0.003 0.003 k = 9 0.011 1.011 0.312 0.002 0.002 k = 12 0.006 0.581 0.561 0.000 0.001 CAPE (∆) k = 1 -0.873 -8.810 0.000 0.142 0.175 k = 3 -0.661 -7.609 0.000 0.096 0.119 k = 6 -0.309 -4.104 0.000 0.025 0.031 k = 9 -0.156 -2.126 0.034 0.006 0.008 k = 12 -0.024 -0.325 0.745 0.000 0.000 TB3M (∆) k = University1 -0.074 -0.650 0.516 0.001 0.001 k = 3 -0.099 -0.861 0.389 0.001 0.001 k = 6 0.141 1.192 0.233 0.002 0.003 k = 9 -0.029 -0.250 0.803 0.000 0.000 k = 12 0.031 0.272 0.785 0.000 0.000

213

Table 4.10, continued (i)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 TN5Y (∆) k = 1 -0.022 -0.144 0.885 0.000 0.000 k = 3 -0.002 -0.011 0.991 0.000 0.000 k = 6 0.341 2.178 0.029 0.007 0.009 k = 9 0.207 1.327 0.185 0.003 0.003 k = 12 0.108 0.689 0.491 0.001 0.001 TN10Y (∆) k = 1 0.056 0.313 0.754 0.000 0.000 k = 3 -0.037 -0.204 0.839 0.000 0.000 k = 6 0.267 1.471 0.141 0.003 0.004 k = 9 0.230 1.258 0.208 0.002 0.003 k = 12 0.320 1.707 0.088 0.004 0.005 S5Y3M k = 1 -0.017 -0.319 0.750 0.000 0.000 k = 3 -0.063 -1.154 0.248 0.002 0.002 k = 6 -0.112 -2.049 0.040 0.006 0.007 k = 9 -0.229 -4.131 0.000 0.025 0.031 k = 12 -0.282 -5.022 0.000 0.038Malaya 0.047 S10Y3M k = 1 -0.102 -2.393 0.017 0.008 0.010 k = 3 -0.131 -3.095 0.002 0.014 0.017 k = 6 -0.163 -3.826 0.000of 0.021 0.026 k = 9 -0.200 -4.623 0.000 0.031 0.039 k = 12 -0.232 -5.292 0.000 0.041 0.052 M1 (g) k = 1 0.026 0.352 0.725 0.000 0.000 k = 3 0.042 0.564 0.573 0.000 0.001 k = 6 -0.080 -1.008 0.313 0.001 0.002 k = 9 -0.021 -0.269 0.788 0.000 0.000 k = 12 0.071 0.933 0.351 0.001 0.002 M2 (g) k = 1 0.203 1.399 0.162 0.003 0.003 k = 3 0.268 1.848 0.065 0.005 0.006 k = 6 0.118 0.812 0.417 0.001 0.001 k = 9 0.222 1.532 0.126 0.003 0.004 k = University12 0.482 3.262 0.001 0.016 0.019

214

Table 4.10, continued (ii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 RM1 (g) k = 1 -0.036 -0.553 0.580 0.000 0.001 k = 3 -0.013 -0.205 0.838 0.000 0.000 k = 6 -0.139 -1.907 0.057 0.005 0.007 k = 9 -0.117 -1.627 0.104 0.004 0.005 k = 12 0.005 0.073 0.942 0.000 0.000 RM2 (g) k = 1 -0.041 -0.377 0.706 0.000 0.000 k = 3 0.024 0.223 0.824 0.000 0.000 k = 6 -0.115 -1.027 0.304 0.002 0.002 k = 9 -0.122 -1.085 0.278 0.002 0.002 k = 12 0.142 1.284 0.199 0.002 0.003 INFL k = 1 0.065 0.436 0.663 0.000 0.000 k = 3 -0.022 -0.152 0.880 0.000 0.000 k = 6 0.221 1.459 0.145 0.003 0.004 k = 9 0.444 2.813 0.005 0.012 0.015 k = 12 0.234 1.512 0.130 0.003Malaya 0.004 RGDPG k = 1 -0.101 -5.787 0.000 0.050 0.063 k = 3 -0.065 -3.907 0.000 0.022 0.028 k = 6 -0.010 -0.599 0.549of 0.001 0.001 k = 9 0.008 0.508 0.611 0.000 0.000 k = 12 0.008 0.461 0.645 0.000 0.000 ISMI k = 1 0.019 2.203 0.028 0.007 0.009 k = 3 0.031 3.513 0.000 0.018 0.022 k = 6 0.045 4.874 0.000 0.036 0.045 k = 9 0.044 4.751 0.000 0.034 0.043 k = 12 0.027 2.985 0.003 0.013 0.016 COPE (∆) k = 1 0.000 -0.945 0.345 0.001 0.002 k = 3 0.000 -1.021 0.307 0.001 0.002 k = 6 0.000 -0.339 0.735 0.000 0.000 k = 9 0.000 -0.161 0.872 0.000 0.000 k = University12 0.000 0.179 0.858 0.000 0.000

215

Table 4.10, continued (iii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 NPHP (∆) k = 1 -0.002 -2.897 0.004 0.012 0.015 k = 3 -0.002 -2.273 0.023 0.007 0.009 k = 6 -0.001 -1.978 0.048 0.006 0.007 k = 9 -0.001 -1.640 0.101 0.004 0.005 k = 12 0.001 1.136 0.256 0.002 0.002 CMUO k = 1 -0.010 -1.813 0.070 0.005 0.006 k = 3 0.003 0.588 0.556 0.000 0.001 k = 6 0.017 2.963 0.003 0.013 0.016 k = 9 0.022 3.735 0.000 0.020 0.025 k = 12 0.021 3.649 0.000 0.019 0.024 PMI k = 1 -0.013 -1.632 0.103 0.004 0.005 k = 3 0.005 0.647 0.518 0.001 0.001 k = 6 0.030 3.413 0.001 0.017 0.022 k = 9 0.033 3.749 0.000 0.021 0.026 k = 12 0.020 2.337 0.020 0.008Malaya 0.010 MSCI k = 1 -0.025 -5.453 0.000 0.043 0.054 k = 3 -0.016 -3.573 0.000 0.018 0.023 k = 6 -0.002 -0.350 0.727of 0.000 0.000 k = 9 0.001 0.119 0.905 0.000 0.000 k = 12 0.006 1.343 0.179 0.003 0.003 CDLI (∆) k = 1 -1.113 -8.626 0.000 0.130 0.161 k = 3 -0.896 -7.614 0.000 0.095 0.118 k = 6 -0.401 -3.952 0.000 0.022 0.028 k = 9 -0.216 -2.150 0.032 0.007 0.008 k = 12 -0.063 -0.613 0.540 0.001 0.001 CFNAI k = 1 -0.259 -4.645 0.000 0.032 0.040 k = 3 -0.114 -2.118 0.034 0.006 0.008 k = 6 0.044 0.785 0.432 0.001 0.001 k = 9 0.105 1.859 0.063 0.005 0.006 k = University12 0.098 1.775 0.076 0.005 0.006

Note: Entries in bold indicate significance at the 5% level.

216

The in-sample predictability test for the naïve moving average model shows that variables that are significant for all horizons (i.e. k = 1, 3, 6, 9, 12) include S10Y3M and

ISMI. Overall, CAPE (∆) has the highest pseudo- 푅2(Mcfadden) and pseudo- 푅2(Diebold

& Rudebusch) across the k = 1 and k = 3 horizons compared to other variables with the

2 2 2 values of 푅푀푐푓푎푑푑푒푛,푘=1 = 0.142, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=1 = 0.175; and 푅푀푐푓푎푑푑푒푛,푘=3 =

2 2 0.096, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=3 = 0.119; while ISMI has the highest 푅 across the k = 6 and

2 2 k = 9 horizons of 푅푀푐푓푎푑푑푒푛,푘=6 = 0.036, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=6 = 0.045; and

2 2 2 푅푀푐푓푎푑푑푒푛,푘=9 = 0.034, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=9 = 0.043. S10Y3M has the highest 푅 across

2 2 the k = 12 horizon of 푅푀푐푓푎푑푑푒푛,푘=12 = 0.041 and 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=12 = 0.052. Malaya

University

217

4.5.5 Naïve Moving Average Negative Return In-sample Predictability Test Results

Table 4.11: In-sample Predictability Test Results for Semi-parametric Model: Naïve Moving Average Negative Return

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 DVD (g) k = 1 -0.178 -1.951 0.051 0.005 0.007 k = 3 -0.114 -1.243 0.214 0.002 0.003 k = 6 -0.173 -1.887 0.059 0.005 0.006 k = 9 -0.230 -2.494 0.013 0.009 0.011 k = 12 -0.312 -3.342 0.001 0.016 0.020 EAR (g) k = 1 -0.103 -4.106 0.000 0.045 0.055 k = 3 -0.050 -3.287 0.001 0.021 0.026 k = 6 -0.005 -0.461 0.645 0.000 0.000 k = 9 0.012 1.122 0.262 0.002 0.002 k = 12 0.003 0.312 0.755 0.000 0.000 RDVD (g) k = 1 -0.222 -2.605 0.009 0.010 0.012 k = 3 -0.152 -1.790 0.074 0.005Malaya 0.006 k = 6 -0.281 -3.262 0.001 0.015 0.019 k = 9 -0.370 -4.177 0.000 0.026 0.032 k = 12 -0.360 -4.018 0.000of 0.024 0.030 REAR (g) k = 1 -0.111 -4.314 0.000 0.049 0.060 k = 3 -0.053 -3.422 0.001 0.022 0.028 k = 6 -0.006 -0.631 0.528 0.001 0.001 k = 9 0.010 0.925 0.355 0.001 0.002 k = 12 0.002 0.208 0.836 0.000 0.000 CAPE (∆) k = 1 -0.679 -7.639 0.000 0.099 0.121 k = 3 -0.614 -7.121 0.000 0.085 0.104 k = 6 -0.396 -5.098 0.000 0.040 0.049 k = 9 -0.177 -2.399 0.016 0.008 0.010 k = 12 -0.057 -0.785 0.433 0.001 0.001 TB3M (∆) k = University1 -0.104 -0.891 0.373 0.001 0.001 k = 3 -0.121 -1.044 0.297 0.002 0.002 k = 6 -0.043 -0.375 0.708 0.000 0.000 k = 9 -0.064 -0.557 0.578 0.000 0.001 k = 12 0.158 1.279 0.201 0.002 0.003

218

Table 4.11, continued (i)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 TN5Y (∆) k = 1 0.023 0.147 0.883 0.000 0.000 k = 3 -0.080 -0.511 0.609 0.000 0.000 k = 6 0.166 1.064 0.287 0.002 0.002 k = 9 0.133 0.846 0.398 0.001 0.001 k = 12 0.221 1.376 0.169 0.003 0.003 TN10Y (∆) k = 1 0.056 0.313 0.754 0.000 0.000 k = 3 -0.037 -0.204 0.839 0.000 0.000 k = 6 0.267 1.471 0.141 0.003 0.004 k = 9 0.230 1.258 0.208 0.002 0.003 k = 12 0.320 1.707 0.088 0.004 0.005 S5Y3M k = 1 -0.017 -0.319 0.750 0.000 0.000 k = 3 -0.063 -1.154 0.248 0.002 0.002 k = 6 -0.112 -2.049 0.040 0.006 0.007 k = 9 -0.229 -4.131 0.000 0.025 0.031 k = 12 -0.282 -5.022 0.000 0.038Malaya 0.047 S10Y3M k = 1 -0.034 -0.796 0.426 0.001 0.001 k = 3 -0.066 -1.555 0.120 0.003 0.004 k = 6 -0.106 -2.478 0.013of 0.009 0.011 k = 9 -0.187 -4.304 0.000 0.027 0.034 k = 12 -0.226 -5.137 0.000 0.039 0.049 M1 (g) k = 1 0.026 0.352 0.725 0.000 0.000 k = 3 0.042 0.564 0.573 0.000 0.001 k = 6 -0.080 -1.008 0.313 0.001 0.002 k = 9 -0.021 -0.269 0.788 0.000 0.000 k = 12 0.071 0.933 0.351 0.001 0.002 M2 (g) k = 1 0.203 1.399 0.162 0.003 0.003 k = 3 0.268 1.848 0.065 0.005 0.006 k = 6 0.118 0.812 0.417 0.001 0.001 k = 9 0.222 1.532 0.126 0.003 0.004 k = University12 0.482 3.262 0.001 0.016 0.019

219

Table 4.11, continued (ii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 RM1 (g) k = 1 -0.036 -0.553 0.580 0.000 0.001 k = 3 -0.013 -0.205 0.838 0.000 0.000 k = 6 -0.139 -1.907 0.057 0.005 0.007 k = 9 -0.117 -1.627 0.104 0.004 0.005 k = 12 0.005 0.073 0.942 0.000 0.000 RM2 (g) k = 1 -0.041 -0.377 0.706 0.000 0.000 k = 3 0.024 0.223 0.824 0.000 0.000 k = 6 -0.115 -1.027 0.304 0.002 0.002 k = 9 -0.122 -1.085 0.278 0.002 0.002 k = 12 0.142 1.284 0.199 0.002 0.003 INFL k = 1 0.099 0.665 0.506 0.001 0.001 k = 3 0.029 0.198 0.843 0.000 0.000 k = 6 0.223 1.460 0.144 0.003 0.004 k = 9 0.451 2.845 0.004 0.012 0.015 k = 12 0.185 1.195 0.232 0.002Malaya 0.003 RGDPG k = 1 -0.095 -5.521 0.000 0.046 0.057 k = 3 -0.058 -3.489 0.001 0.018 0.022 k = 6 -0.023 -1.417 0.156of 0.003 0.004 k = 9 0.007 0.399 0.690 0.000 0.000 k = 12 0.025 1.521 0.128 0.003 0.004 ISMI k = 1 0.026 2.921 0.004 0.012 0.015 k = 3 0.040 4.440 0.000 0.030 0.036 k = 6 0.052 5.401 0.000 0.045 0.056 k = 9 0.048 5.020 0.000 0.039 0.048 k = 12 0.030 3.310 0.001 0.016 0.020 COPE (∆) k = 1 0.000 -0.874 0.382 0.001 0.001 k = 3 0.000 -1.183 0.237 0.002 0.002 k = 6 0.000 0.058 0.954 0.000 0.000 k = 9 0.000 -0.101 0.920 0.000 0.000 k = University12 0.000 0.383 0.701 0.000 0.000

220

Table 4.11, continued (iii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 NPHP (∆) k = 1 -0.001 -1.714 0.087 0.004 0.005 k = 3 -0.001 -2.168 0.030 0.007 0.008 k = 6 -0.001 -1.492 0.136 0.003 0.004 k = 9 -0.001 -1.207 0.228 0.002 0.003 k = 12 0.000 0.024 0.981 0.000 0.000 CMUO k = 1 -0.010 -1.700 0.089 0.004 0.005 k = 3 0.002 0.272 0.786 0.000 0.000 k = 6 0.013 2.268 0.023 0.007 0.009 k = 9 0.014 2.352 0.019 0.008 0.010 k = 12 0.012 2.037 0.042 0.006 0.008 PMI k = 1 -0.009 -1.124 0.261 0.002 0.002 k = 3 0.011 1.348 0.178 0.003 0.003 k = 6 0.028 3.186 0.001 0.015 0.019 k = 9 0.034 3.732 0.000 0.021 0.026 k = 12 0.027 3.082 0.002 0.014Malaya 0.018 MSCI k = 1 -0.024 -5.179 0.000 0.039 0.049 k = 3 -0.015 -3.255 0.001 0.015 0.019 k = 6 -0.003 -0.657 0.511of 0.001 0.001 k = 9 0.002 0.366 0.714 0.000 0.000 k = 12 0.009 1.944 0.052 0.006 0.007 CDLI (∆) k = 1 -1.034 -8.227 0.000 0.118 0.144 k = 3 -0.834 -7.202 0.000 0.085 0.104 k = 6 -0.402 -3.935 0.000 0.023 0.028 k = 9 -0.233 -2.311 0.021 0.008 0.009 k = 12 -0.093 -0.899 0.369 0.001 0.001 CFNAI k = 1 -0.248 -4.473 0.000 0.030 0.037 k = 3 -0.118 -2.204 0.028 0.007 0.009 k = 6 0.048 0.861 0.389 0.001 0.001 k = 9 0.084 1.465 0.143 0.003 0.004 k = University12 0.128 2.201 0.028 0.007 0.009

Note: Entries in bold indicate significance at the 5% level.

221

The in-sample predictability test for predicting bear stock markets of the naïve moving average negative return model shows that only ISMI is significant for all horizons

(i.e. k = 1, 3, 6, 9, 12). Overall, CDLI (∆) has the highest value for both the pseudo-

푅2(Mcfadden) and pseudo- 푅2(Diebold & Rudebusch) across the k = 1 horizon and joint highest for k = 3 horizon along with CAPE (∆) compared to other variables with the values

2 2 2 of 푅푀푐푓푎푑푑푒푛,푘=1 = 0.118, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=1 = 0.144; and 푅푀푐푓푎푑푑푒푛,푘=3 = 0.085,

2 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=3 = 0.104 (same values of k = 3 horizon for both CDLI (∆) and

CAPE (∆)). CAPE (∆) also registered the highest pseudo- 푅2 for the k = 6 horizon of

2 2 푅푀푐푓푎푑푑푒푛,푘=6 = 0.040, and 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=6 = 0.049. Further, ISMI has the highest

2 2 2 푅 across the k = 9 horizon of 푅푀푐푓푎푑푑푒푛,푘=9= 0.039 and 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=9 = 0.048;

2 2 while S10Y3M has the highest 푅 across the k = 12 Malaya horizon, i.e. 푅푀푐푓푎푑푑푒푛,푘=12 = 0.039 2 and 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=12 = 0.049. of

University

222

4.5.6 Lunde & Timmermann’s B-B Algorithm In-sample Predictability Test Results

Table 4.12: In-sample Predictability Test Results for Non-parametric Model: Lunde & Timmermann’s B-B Algorithm

223

Table 4.12, continued (i)

224

Table 4.12, continued (ii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 RM1 (g) k = 1 -0.036 -0.553 0.580 0.000 0.001 k = 3 -0.013 -0.205 0.838 0.000 0.000 k = 6 -0.139 -1.907 0.057 0.005 0.007 k = 9 -0.117 -1.627 0.104 0.004 0.005 k = 12 0.005 0.073 0.942 0.000 0.000 RM2 (g) k = 1 -0.041 -0.377 0.706 0.000 0.000 k = 3 0.024 0.223 0.824 0.000 0.000 k = 6 -0.115 -1.027 0.304 0.002 0.002 k = 9 -0.122 -1.085 0.278 0.002 0.002 k = 12 0.142 1.284 0.199 0.002 0.003 INFL k = 1 0.099 0.665 0.506 0.001 0.001 k = 3 0.029 0.198 0.843 0.000 0.000 k = 6 0.223 1.460 0.144 0.003 0.004 k = 9 0.451 2.845 0.004 0.012 0.015 k = 12 0.185 1.195 0.232 0.002Malaya 0.003 RGDPG k = 1 -0.095 -5.521 0.000 0.046 0.057 k = 3 -0.058 -3.489 0.001 0.018 0.022 k = 6 -0.023 -1.417 0.156of 0.003 0.004 k = 9 0.007 0.399 0.690 0.000 0.000 k = 12 0.025 1.521 0.128 0.003 0.004 ISMI k = 1 0.026 2.921 0.004 0.012 0.015 k = 3 0.040 4.440 0.000 0.030 0.036 k = 6 0.052 5.401 0.000 0.045 0.056 k = 9 0.048 5.020 0.000 0.039 0.048 k = 12 0.030 3.310 0.001 0.016 0.020 COPE (∆) k = 1 0.000 -0.874 0.382 0.001 0.001 k = 3 0.000 -1.183 0.237 0.002 0.002 k = 6 0.000 0.058 0.954 0.000 0.000 k = 9 0.000 -0.101 0.920 0.000 0.000 k = University12 0.000 0.383 0.701 0.000 0.000

225

Table 4.12, continued (iii)

Note: Entries in bold indicate significance at the 5% level.

226

The in-sample predictability test for predicting bear stock markets of the Lunde &

Timmermann’s B-B algorithm shows that only ISMI is significant for all horizons (i.e. k =

1, 3, 6, 9, 12). Overall, CDLI (∆) has the highest value for both the pseudo- 푅2(Mcfadden) and pseudo- 푅2(Diebold & Rudebusch) across the k = 1 horizon and joint highest for k = 3

2 horizon with CAPE (∆) compared to other variables with the values of 푅푀푐푓푎푑푑푒푛,푘=1 =

2 2 2 0.118,푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=1 = 0.144; and 푅푀푐푓푎푑푑푒푛,푘=3 = 0.085, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=3

= 0.104 (same values of k = 3 for both CDLI (∆) and CAPE (∆)). Further, ISMI has the

2 2 highest 푅 across both the k = 6 and k = 9 horizons of 푅푀푐푓푎푑푑푒푛,푘=6 = 0.045,

2 2 2 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=6 = 0.056; and 푅푀푐푓푎푑푑푒푛,푘=9 = 0.039, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=9 =

2 2 0.048; while S10Y3M has the highest 푅 across the k = 12 horizon, i.e. 푅푀푐푓푎푑푑푒푛,푘=12 =

2 0.039 and 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=12 = 0.049. Malaya of

University

227

4.5.7 Candelon, Piplak & Straetman’s B-B Algorithm In-sample Predictability Test Results

Table 4.13: In-sample Predictability Test Results for Non-parametric Model: Candelon, Piplak & Straetman’s B-B Algorithm

Pseudo- Pseudo- 2 2 훽̂ z-stat p-value 푅푀푐푓푎푑푑푒푛 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ DVD (g) k = 1 -0.035 -0.376 0.707 0.000 0.000 k = 3 0.010 0.102 0.919 0.000 0.000 k = 6 -0.007 -0.078 0.938 0.000 0.000 k = 9 -0.044 -0.478 0.633 0.000 0.000 k = 12 -0.127 -1.361 0.173 0.003 0.003 EAR (g) k = 1 -0.058 -3.484 0.001 0.025 0.030 k = 3 -0.032 -2.382 0.017 0.010 0.012 k = 6 0.005 0.537 0.592 0.000 0.001 k = 9 0.008 0.762 0.446 0.001 0.001 k = 12 -0.007 -0.676 0.499 0.001 0.001 RDVD (g) k = 1 -0.219 -2.560 0.011 0.009Malaya 0.012 k = 3 -0.151 -1.759 0.079 0.004 0.005 k = 6 -0.188 -2.197 0.028 0.007 0.009 k = 9 -0.220 -2.533 0.011of 0.009 0.012 k = 12 -0.236 -2.684 0.007 0.011 0.013 REAR (g) k = 1 -0.066 -3.828 0.000 0.030 0.037 k = 3 -0.037 -2.630 0.009 0.012 0.015 k = 6 0.003 0.290 0.772 0.000 0.000 k = 9 0.005 0.531 0.595 0.000 0.001 k = 12 -0.009 -0.854 0.393 0.001 0.001 CAPE (∆) k = 1 -0.721 -8.245 0.000 0.114 0.138 k = 3 -0.408 -5.232 0.000 0.042 0.051 k = 6 -0.191 -2.567 0.010 0.010 0.012 k = 9 -0.072 -0.977 0.329 0.001 0.002 k = 12 -0.054 -0.725 0.468 0.001 0.001 TB3MUniversity (∆) k = 1 0.079 0.674 0.500 0.001 0.001 k = 3 0.120 1.019 0.308 0.002 0.002 k = 6 0.098 0.834 0.404 0.001 0.001 k = 9 0.025 0.225 0.822 0.000 0.000 k = 12 0.059 0.530 0.596 0.000 0.001

228

Table 4.13, continued (i)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 TN5Y (∆) k = 1 0.217 1.396 0.163 0.003 0.003 k = 3 0.166 1.066 0.287 0.002 0.002 k = 6 0.290 1.859 0.063 0.005 0.006 k = 9 0.128 0.825 0.409 0.001 0.001 k = 12 0.083 0.534 0.593 0.000 0.001 TN10Y (∆) k = 1 0.301 1.673 0.094 0.004 0.005 k = 3 0.235 1.298 0.194 0.002 0.003 k = 6 0.348 1.916 0.055 0.005 0.007 k = 9 0.208 1.139 0.255 0.002 0.002 k = 12 0.183 1.001 0.317 0.001 0.002 S5Y3M k = 1 -0.159 -2.911 0.004 0.012 0.015 k = 3 -0.171 -3.114 0.002 0.014 0.017 k = 6 -0.173 -3.142 0.002 0.014 0.018 k = 9 -0.240 -4.282 0.000 0.028 0.034 k = 12 -0.241 -4.277 0.000 0.028Malaya 0.034 S10Y3M k = 1 -0.151 -3.515 0.000 0.018 0.022 k = 3 -0.161 -3.732 0.000 0.020 0.025 k = 6 -0.155 -3.604 0.000of 0.019 0.023 k = 9 -0.196 -4.469 0.000 0.030 0.037 k = 12 -0.198 -4.481 0.000 0.030 0.037 M1 (g) k = 1 0.076 0.989 0.323 0.001 0.002 k = 3 0.057 0.733 0.464 0.001 0.001 k = 6 -0.049 -0.621 0.535 0.001 0.001 k = 9 0.001 0.015 0.988 0.000 0.000 k = 12 0.105 1.373 0.170 0.003 0.003 M2 (g) k = 1 0.434 2.945 0.003 0.013 0.015 k = 3 0.412 2.826 0.005 0.011 0.014 k = 6 0.309 2.129 0.033 0.006 0.008 k = 9 0.468 3.208 0.001 0.015 0.018 k = University12 0.776 5.082 0.000 0.040 0.048

229

Table 4.13, continued (ii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 RM1 (g) k = 1 -0.054 -0.819 0.413 0.001 0.001 k = 3 -0.056 -0.828 0.408 0.001 0.001 k = 6 -0.162 -2.175 0.030 0.007 0.009 k = 9 -0.127 -1.761 0.078 0.005 0.006 k = 12 0.008 0.110 0.913 0.000 0.000 RM2 (g) k = 1 -0.062 -0.575 0.566 0.000 0.001 k = 3 -0.031 -0.286 0.775 0.000 0.000 k = 6 -0.116 -1.025 0.306 0.002 0.002 k = 9 -0.049 -0.440 0.660 0.000 0.000 k = 12 0.231 2.105 0.035 0.007 0.008 INFL k = 1 0.176 1.187 0.235 0.002 0.002 k = 3 0.402 2.589 0.010 0.010 0.012 k = 6 0.541 3.417 0.001 0.017 0.021 k = 9 0.425 2.678 0.007 0.011 0.013 k = 12 0.232 1.485 0.137 0.003Malaya 0.004 RGDPG k = 1 -0.062 -3.723 0.000 0.020 0.025 k = 3 -0.037 -2.247 0.025 0.007 0.009 k = 6 0.012 0.720 0.471of 0.001 0.001 k = 9 0.019 1.148 0.251 0.002 0.002 k = 12 0.031 1.885 0.060 0.005 0.006 ISMI k = 1 0.044 4.788 0.000 0.035 0.042 k = 3 0.054 5.685 0.000 0.050 0.062 k = 6 0.047 4.932 0.000 0.037 0.046 k = 9 0.036 3.908 0.000 0.023 0.028 k = 12 0.014 1.636 0.102 0.004 0.005 COPE (∆) k = 1 0.000 -0.708 0.479 0.001 0.001 k = 3 0.000 -0.674 0.500 0.001 0.001 k = 6 0.000 0.208 0.836 0.000 0.000 k = 9 0.000 0.356 0.722 0.000 0.000 k = University12 0.000 0.333 0.739 0.000 0.000

230

Table 4.13, continued (iii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 NPHP (∆) k = 1 -0.001 -2.157 0.031 0.007 0.008 k = 3 -0.001 -2.130 0.033 0.007 0.008 k = 6 0.000 -0.713 0.476 0.001 0.001 k = 9 0.000 0.285 0.776 0.000 0.000 k = 12 0.000 0.153 0.878 0.000 0.000 CMUO k = 1 -0.003 -0.480 0.632 0.000 0.000 k = 3 0.012 2.069 0.039 0.006 0.008 k = 6 0.026 4.385 0.000 0.029 0.035 k = 9 0.023 3.817 0.000 0.022 0.027 k = 12 0.015 2.505 0.012 0.009 0.011 PMI k = 1 0.009 1.067 0.286 0.002 0.002 k = 3 0.029 3.276 0.001 0.016 0.020 k = 6 0.032 3.640 0.000 0.020 0.025 k = 9 0.025 2.901 0.004 0.013 0.015 k = 12 0.013 1.566 0.117 0.004Malaya 0.004 MSCI k = 1 -0.014 -3.220 0.001 0.015 0.018 k = 3 -0.006 -1.463 0.144 0.003 0.004 k = 6 0.000 -0.068 0.946of 0.000 0.000 k = 9 0.000 0.065 0.948 0.000 0.000 k = 12 0.002 0.476 0.634 0.000 0.000 CDLI (∆) k = 1 -0.846 -7.286 0.000 0.087 0.106 k = 3 -0.529 -5.014 0.000 0.038 0.047 k = 6 -0.180 -1.787 0.074 0.005 0.006 k = 9 -0.093 -0.910 0.363 0.001 0.001 k = 12 -0.069 -0.660 0.509 0.001 0.001 CFNAI k = 1 -0.114 -2.105 0.035 0.006 0.008 k = 3 0.009 0.169 0.866 0.000 0.000 k = 6 0.127 2.222 0.026 0.007 0.009 k = 9 0.117 2.045 0.041 0.006 0.008 k = University12 0.087 1.525 0.127 0.003 0.004

Note: Entries in bold indicate significance at the 5% level.

231

The in-sample predictability test for predicting bear stock markets of the Candelon,

Piplak & Straetman’s B-B algorithm shows that variables that are significant for all horizons (i.e. k = 1, 3, 6, 9, 12) include S5Y3M, S10Y3M and M2 (g). Overall, CAPE (∆) has the highest pseudo- 푅2(Mcfadden) and pseudo- 푅2(Diebold & Rudebusch) across the k

2 = 1 and k = 3 horizons compared to other variables with the values of 푅푀푐푓푎푑푑푒푛,푘=1=

2 2 0.114, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=1 = 0.138; and 푅푀푐푓푎푑푑푒푛,푘=3 = 0.042,

2 2 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=3 = 0.051. ISMI has the highest 푅 across the k = 6 horizon of

2 2 푅푀푐푓푎푑푑푒푛,푘=6 = 0.037 and 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=6 = 0.046; while S10Y3M has the highest

2 2 2 푅 across the k = 9 and k = 12 horizons of 푅푀푐푓푎푑푑푒푛,푘=9 = 0.030, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=9 =

2 2 0.037; and 푅푀푐푓푎푑푑푒푛,푘=12 = 0.030, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=Malaya12 = 0.037.

University

232

4.5.8 JLS and JLS “Negative Bubbles” Integrated Identifications In-sample Predictability Test Results

Table 4.14: In-sample Predictability Test Results for Scale-invariant Model: JLS and JLS “Negative Bubbles” (6-month Window)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 DVD (g) k = 1 0.218 1.015 0.310 0.009 0.011 k = 3 0.232 1.090 0.276 0.011 0.012 k = 6 0.193 0.884 0.377 0.007 0.008 k = 9 -0.303 -1.384 0.166 0.016 0.018 k = 12 -0.650 -2.693 0.007 0.065 0.074 EAR (g) k = 1 -0.084 -1.601 0.110 0.027 0.031 k = 3 -0.115 -1.983 0.047 0.054 0.062 k = 6 -0.126 -2.343 0.019 0.066 0.076 k = 9 -0.141 -2.661 0.008 0.134 0.154 k = 12 -0.165 -3.072 0.002 0.226 0.257 RDVD (g) k = 1 0.166 0.797 0.426 0.005Malaya 0.006 k = 3 0.031 0.152 0.879 0.000 0.000 k = 6 0.031 0.135 0.892 0.000 0.000 k = 9 -0.046 -0.209 0.834of 0.000 0.000 k = 12 -0.370 -1.661 0.097 0.024 0.027 REAR (g) k = 1 -0.087 -1.625 0.104 0.028 0.033 k = 3 -0.129 -2.141 0.032 0.062 0.072 k = 6 -0.132 -2.440 0.015 0.071 0.082 k = 9 -0.145 -2.709 0.007 0.137 0.157 k = 12 -0.174 -3.138 0.002 0.233 0.265 CAPE (∆) k = 1 -0.407 -2.733 0.006 0.064 0.074 k = 3 -0.196 -1.222 0.222 0.012 0.014 k = 6 -0.501 -2.516 0.012 0.059 0.068 k = 9 -0.537 -2.828 0.005 0.079 0.091 k = 12 -0.398 -2.472 0.013 0.055 0.064 TB3MUniversity (∆) k = 1 0.075 0.326 0.744 0.001 0.001 k = 3 -0.261 -0.852 0.394 0.006 0.007 k = 6 -0.538 -1.505 0.132 0.021 0.024 k = 9 -0.178 -0.615 0.539 0.003 0.004 k = 12 0.622 1.616 0.106 0.022 0.026

233

Table 4.14, continued (i)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 TN5Y (∆) k = 1 0.058 0.180 0.857 0.000 0.000 k = 3 -0.468 -1.081 0.280 0.010 0.011 k = 6 0.023 0.056 0.955 0.000 0.000 k = 9 -0.089 -0.235 0.814 0.000 0.001 k = 12 0.487 1.418 0.156 0.017 0.020 TN10Y (∆) k = 1 0.128 0.319 0.750 0.001 0.001 k = 3 -0.678 -1.242 0.214 0.013 0.015 k = 6 0.218 0.437 0.662 0.002 0.002 k = 9 0.044 0.101 0.919 0.000 0.000 k = 12 0.622 1.616 0.106 0.022 0.026 S5Y3M k = 1 -0.792 -5.167 0.000 0.255 0.289 k = 3 -0.866 -5.386 0.000 0.286 0.324 k = 6 -0.378 -3.246 0.001 0.088 0.101 k = 9 -0.231 -2.059 0.040 0.035 0.040 k = 12 -0.257 -2.074 0.038 0.036Malaya 0.042 S10Y3M k = 1 -0.574 -4.768 0.000 0.209 0.238 k = 3 -0.609 -4.912 0.000 0.226 0.257 k = 6 -0.270 -2.865 0.004of 0.068 0.079 k = 9 -0.139 -1.584 0.113 0.021 0.024 k = 12 -0.133 -1.461 0.144 0.018 0.021 M1 (g) k = 1 -0.880 -3.104 0.002 0.090 0.103 k = 3 -1.217 -3.570 0.000 0.122 0.141 k = 6 -0.156 -0.745 0.456 0.005 0.005 k = 9 0.231 1.406 0.160 0.018 0.021 k = 12 0.173 1.217 0.224 0.013 0.015 M2 (g) k = 1 -0.283 -0.794 0.428 0.005 0.006 k = 3 -0.420 -1.159 0.246 0.012 0.014 k = 6 0.074 0.229 0.819 0.000 0.001 k = 9 0.481 1.583 0.113 0.021 0.025 k = University12 0.482 3.262 0.001 0.016 0.019

234

Table 4.14, continued (ii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 RM1 (g) k = 1 -0.763 -3.082 0.002 0.086 0.099 k = 3 -1.435 -3.952 0.000 0.170 0.194 k = 6 -0.241 -1.239 0.215 0.013 0.015 k = 9 0.191 1.462 0.144 0.019 0.022 k = 12 0.156 1.362 0.173 0.016 0.019 RM2 (g) k = 1 -0.267 -0.954 0.340 0.008 0.009 k = 3 -0.480 -1.699 0.089 0.026 0.030 k = 6 -0.149 -0.556 0.578 0.003 0.003 k = 9 0.353 1.672 0.095 0.024 0.028 k = 12 0.358 1.745 0.081 0.026 0.030 INFL k = 1 -0.351 -0.835 0.404 0.006 0.007 k = 3 -0.064 -0.153 0.879 0.000 0.000 k = 6 -0.908 -2.011 0.044 0.036 0.041 k = 9 -0.714 -1.746 0.081 0.026 0.030 k = 12 -0.178 -0.424 0.672 0.001Malaya 0.002 RGDPG k = 1 0.019 0.625 0.532 0.003 0.004 k = 3 -0.070 -1.780 0.075 0.027 0.031 k = 6 -0.220 -3.968 0.000of 0.163 0.187 k = 9 -0.187 -3.878 0.000 0.164 0.187 k = 12 -0.146 -3.596 0.000 0.126 0.145 ISMI k = 1 0.058 2.493 0.013 0.057 0.065 k = 3 0.053 2.290 0.022 0.047 0.054 k = 6 0.015 0.698 0.485 0.004 0.005 k = 9 -0.041 -1.975 0.048 0.033 0.038 k = 12 -0.111 -3.861 0.000 0.160 0.183 COPE (∆) k = 1 0.000 -0.052 0.959 0.000 0.000 k = 3 0.000 -0.291 0.771 0.001 0.001 k = 6 0.000 -2.121 0.034 0.039 0.046 k = 9 0.000 -1.273 0.203 0.014 0.016 k = University12 0.000 -2.495 0.013 0.053 0.061

235

Table 4.14, continued (iii)

Pseudo- Pseudo- ퟐ ퟐ 휷̂ z-stat p-value 푹푴풄풇풂풅풅풆풏 푹푫풊풆풃풐풍풅 & 푹풖풅풆풃풖풔풄풉 NPHP (∆) k = 1 -0.002 -1.586 0.113 0.021 0.025 k = 3 -0.002 -1.477 0.140 0.018 0.021 k = 6 -0.002 -1.194 0.232 0.012 0.014 k = 9 -0.003 -1.809 0.071 0.027 0.032 k = 12 -0.002 -1.463 0.144 0.018 0.021 CMUO k = 1 0.058 3.885 0.000 0.146 0.167 k = 3 0.027 1.901 0.057 0.031 0.036 k = 6 -0.026 -1.565 0.118 0.021 0.024 k = 9 -0.052 -2.804 0.005 0.075 0.087 k = 12 -0.055 -3.089 0.002 0.097 0.112 PMI k = 1 0.033 1.796 0.073 0.028 0.033 k = 3 0.004 0.221 0.825 0.000 0.000 k = 6 -0.076 -3.061 0.002 0.084 0.097 k = 9 -0.122 -4.338 0.000 0.211 0.240 k = 12 -0.112 -4.385 0.000 0.227Malaya 0.258 MSCI k = 1 0.021 2.181 0.029 0.041 0.047 k = 3 0.012 1.282 0.200 0.014 0.016 k = 6 -0.002 -0.230 0.818of 0.000 0.001 k = 9 -0.008 -0.858 0.391 0.006 0.007 k = 12 -0.008 -0.874 0.382 0.006 0.007 CDLI (∆) k = 1 -1.110 -4.167 0.000 0.156 0.179 k = 3 -1.631 -4.654 0.000 0.243 0.275 k = 6 -1.784 -4.730 0.000 0.306 0.346 k = 9 -1.375 -4.066 0.000 0.260 0.295 k = 12 -1.052 -3.848 0.000 0.189 0.216 CFNAI k = 1 0.120 1.107 0.268 0.010 0.012 k = 3 -0.156 -1.356 0.175 0.015 0.018 k = 6 -0.750 -4.110 0.000 0.185 0.212 k = 9 -0.769 -4.124 0.000 0.217 0.247 k = University12 -0.636 -3.966 0.000 0.194 0.221

Note: Entries in bold indicate significance at the 5% level.

236

The in-sample predictability test for predicting bubble induced-crashes (6-month window) of the JLS model and JLS “negative bubbles” model shows that variables that are significant for all horizons (i.e. k = 1, 3, 6, 9, 12) include S5Y3M and CDLI (∆). Overall,

S5Y3M has the highest pseudo- 푅2(Mcfadden) and pseudo- 푅2 (Diebold & Rudebusch) across the k = 1 and k = 3 horizons compared to other variables with the values of

2 2 2 푅푀푐푓푎푑푑푒푛,푘=1 = 0.255, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=1 = 0.289; and 푅푀푐푓푎푑푑푒푛,푘=3 = 0.286,

2 2 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=3 = 0.324. CDLI (∆) has the highest 푅 across the k = 6 and k = 9

2 2 2 horizons of 푅푀푐푓푎푑푑푒푛,푘=6 = 0.306, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=6 = 0.346; and 푅푀푐푓푎푑푑푒푛,푘=9 =

2 2 0.260, 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=9 = 0.295; while CFNAI has the highest 푅 across the k = 12

2 2 horizon of 푅푀푐푓푎푑푑푒푛,푘=12 = 0.194 and 푅퐷푖푒푏표푙푑 & 푅푢푑푒푏푢푠푐ℎ,푘=12 = 0.221.

Malaya

4.6 Out-of-sample Predictability Test Resultsof

Table 4.15 to Table 4.16 illustrates the out-of-sample predictability test results for predicting stock market declines of parametric models. As a recap, a MSPE-adj of higher value denotes that the specific variable has better predictive power compared to that of the restricted model. Recalling the in depth description in Chapter 3, a restricted predictive regressive model is a control model where only a constant is included. Table 4.17 to Table

4.22 that follows shows the out-of-sample predictability test results for predicting stock marketUniversity declines of the semi-parametric and non-parametric models measured with QPS.

237

4.6.1 Markov-switching Filtered Probabilities Out-of-sample Predictability Test Results

Table 4.15: Out-of-sample Predictability Test Results for Parametric Model: Markov- switching Filtered Probabilities

Clark & West (2007)’s MSPE-adj k = 1 k = 3 k = 6 k = 9 k = 12 DVD (g) -15.561 -3.518 15.347** 15.194** 14.977** EAR (g) -15.880 -15.425 -14.961 14.488** 14.071** RDVD (g) -1.029 -0.918 -0.818 -0.825 -0.946 REAR (g) -0.834 -0.954 -1.086 1.337* 1.634** CAPE (∆) -2.103 -2.033 -1.921 -1.833 -1.854 TB3M (∆) -4.532 -4.522 4.400** 4.430** -4.660 TN5Y (∆) -4.932 -4.933 4.976** 5.287** 5.526** TN10Y (∆) -4.998 -4.969 5.000** 5.307** 5.539** S5Y3M -9.579 -9.629 -9.654 -9.648 -9.560 S10Y3M -9.425 -9.480 -9.507 -9.499 -9.406 M1 (g) -9.425 -9.480 -9.507 -9.499 -9.406 M2 (g) -13.815 -13.819 -13.865 -13.894 13.801** RM1 (g) -2.548 -2.360 -2.088 -1.919 -1.883 RM2 (g) -11.310 -10.791 -10.504Malaya -10.456 -10.182 INFL -15.301 -15.427 -15.487 -15.522 -15.441 RDGDP -7.047 -7.073 -7.085 -7.037 -6.844 ISMI -6.049 -6.085of 6.112** 6.076** 5.894** COPE (∆) -6.090 -6.130 -6.127 6.110** -5.948 NPHP (∆) -7.393 -8.137 -9.196 -15.045 -0.417 CMUO -7.643 -7.665 -7.665 7.595** 7.372** PMI -6.045 -6.081 -6.108 6.072** 5.889** MSCI -6.028 -6.064 -6.091 -6.055 -5.873 CDLI (∆) -4.654 -4.539 -4.449 -4.398 -4.401 CFNAI -6.242 -6.429 -6.685 -7.236 -7.617 Clark & West (2007)’s MSPE-adj statistic critical values are 1.282 (10%) and 1.645 (5%). Bold entries indicate significant critical values.

* Denotes significance at the 10% level. ** Denotes significance at the 5% level. University

238

Evidence of the out-of-sample predictability test for predicting bear stock markets of the Markov-switching model using the extracted filtered probabilities shows that DVD

(g), TN5Y (∆), TN10Y (∆) and ISMI are the most consistent leading variables across the k

= 6, k = 9 and k = 12 horizons compared to other variables at 5% level of significance. On the other hand, variables which exhibit the highest predictive power is the DVD (g) for the k = 6, k = 9 and k = 12 horizons followed by EAR (g) for the k = 9 and k = 12 horizons. It is noteworthy that M2 (g) also has a very high MSPE-adj value for the k = 12 horizon. Other variables which significantly predict stock market declines in the out-of-sample test at 5% level of significance include TB3M (∆) (for the k = 6 and k = 9 horizons), COPE (∆) (for the k = 9 horizon), CMUO (for the k = 9 and k = 12 horizons) and PMI (for the k = 9 and k = 12 horizons). Malaya of

University

239

4.6.2 Markov-switching Smoothed Probabilities Out-of-sample Predictability Test Results

Table 4.16: Out-of-sample Predictability Test Results for Parametric Model: Markov- switching Smoothed Probabilities

Clark & West (2007)’s MSPE-adj k = 1 k = 3 k = 6 k = 9 k = 12 DVD (g) -17.525 3.946** 17.130** 16.922** 16.732** EAR (g) -17.993 -17.283 -16.592 15.953** 15.492** RDVD (g) -0.501 -0.355 -0.253 -0.277 0.423 REAR (g) -0.305 -0.391 -0.521 0.789 1.116 CAPE (∆) -1.588 -1.481 -1.363 -1.288 -1.339 TB3M (∆) -4.120 -4.071 3.929** 3.971** 4.257** TN5Y (∆) -4.549 4.513** 4.545** 4.890** 5.194** TN10Y (∆) -4.621 4.551** 4.571** 4.913** 5.209** S5Y3M -10.023 -10.076 -10.074 -10.048 -9.976 S10Y3M -10.190 -10.236 -10.231 -10.207 -10.143 M1 (g) -15.147 -15.117 -15.135 -15.170 15.138** M2 (g) -11.893 -11.908 -11.842 -11.736 -11.577 RM1 (g) -2.044 -1.815 -1.531 -1.375 -1.368 RM2 (g) -12.162 -11.442 -11.037Malaya -10.978 -10.696 INFL -17.153 -17.286 -17.330 -17.390 -17.398 RDGDP -7.533 -7.566 -7.555 -7.485 -7.299 ISMI -6.522 -6.569of 6.575** 6.518** 6.340** COPE (∆) -6.562 -6.614 -6.589 -6.552 6.394** NPHP (∆) -7.888 -8.659 -9.742 -16.714 0.105 CMUO -8.145 -8.172 8.145** 8.054** 7.839** PMI -6.517 -6.564 -6.570 6.513** 6.335** MSCI -6.500 -6.548 -6.554 -6.497 -6.319 CDLI (∆) -4.250 -4.090 -3.981 -3.938 -3.980 CFNAI -5.986 -6.158 -6.428 -7.061 7.555** Clark & West (2007)’s MSPE-adj statistic critical values are 1.282 (10%) and 1.645 (5%). Bold entries indicate significant critical values.

* Denotes significance at 10% level. ** Denotes significance at 5% level.

University

240

Evidence of the out-of-sample predictability test for predicting bear stock markets of the Markov-switching model using the extracted smoothed probabilities shows that DVD

(g), TN5Y (∆), TN10Y (∆) are the most consistent leading variables across the k = 3, k = 6, k = 9 and k = 12 horizons compared to other variables at 5% level of significance. These are followed by TB3M (∆), ISMI and CMUO for the k = 6, k = 9 and k = 12 horizons. On the other hand, variables which exhibit the highest predictive power is the DVD (g) for the k =

6, k = 9 and k = 12 horizons followed by EAR (g) for the k = 9 and k = 12 horizons. It is noteworthy that M2 (g) also has a very high MSPE-adj value for the k = 12 horizon, while

TN10Y (∆) displays the highest predictive power for the k = 3 horizon, albeit not registering a double digit MSPE-adj value compared to the aforementioned variables. Other variables which significantly predict stockMalaya market declines in the out-of- sample test at 5% level of significance include COPE (∆) (for the k = 9 horizon) and PMI (for the k = 9 and k = 12 horizons). of

University

241

4.6.3 Markov-switching Dichotomised Smoothed Probabilities Out-of-sample Predictability Test Results

Table 4.17: Out-of-sample Predictability Test Results for Semi-parametric Model: Markov-switching Dichotomised Smoothed Probabilities

Quadratic Probability Score (QPS) k = 1 k = 3 k = 6 k = 9 k = 12 DVD (g) 0.318 0.333 0.348 0.345 0.355 EAR (g) 0.260 0.252 0.307 0.467 0.543 RDVD (g) 0.293 0.298 0.310 0.582 0.312 REAR (g) 0.250 0.240 0.282 0.387 0.487 CAPE (∆) 0.262 0.281 0.304 0.334 0.324 TB3M (∆) 0.307 0.312 0.317 0.321 0.328 TN5Y (∆) 0.314 0.314 0.317 0.321 0.327 TN10Y (∆) 0.314 0.315 0.314 0.320 0.325 S5Y3M 0.310 0.303 0.294 0.275 0.263 S10Y3M 0.308 0.306 0.298 0.283 0.270 M1 (g) 0.348 0.353 0.324 0.334 0.327 M2 (g) 0.412 0.445 0.432 0.458 0.426 RM1 (g) 0.321 0.322 0.315 0.322 0.321 RM2 (g) 0.337 0.337 0.336Malaya 0.341 0.357 INFL 0.313 0.314 0.318 0.321 0.325 RDGDP 0.248 0.248 0.293 0.311 0.323 ISMI 0.306 0.314of 0.317 0.326 0.340 COPE (∆) 0.317 0.311 0.316 0.321 0.325 NPHP (∆) 0.312 0.316 0.309 0.320 0.331 CMUO 0.281 0.301 0.317 0.321 0.320 PMI 0.262 0.285 0.311 0.326 0.370 MSCI 0.283 0.305 0.339 0.336 0.331 CDLI (∆) 0.215 0.221 0.267 0.300 0.307 CFNAI 0.245 0.264 0.303 0.317 0.329 The quadratic probability score (QPS) proposed by Diebold and Rudebusch (1989) ranges from 0 to 2 with a score of 0 corresponding to perfect accuracy.

The out-of-sample predictability test for predicting bear stock markets of the

Markov-switchingUniversity model with dichotomised smoothed probabilities shows that all values are lower than 0.600 which suggests that the variables proposed by the research have good predictive power. Evidence shows that CDLI (∆) has the highest predictive power (lowest

QPS score) for the horizons of k = 1 (QPS = 0.215), k = 3 (QPS = 0.221) and k = 6 (QPS =

242

0.267) while S5Y3M has the highest predictive power for both the k = 9 horizon (QPS =

0.275) and the k = 12 horizon (QPS = 0.263) compared to the rest of variables in the list.

4.6.4 Naïve Moving Average Out-of-sample Predictability Test Results

Table 4.18: Out-of-sample Predictability Test Results for Semi-parametric Model: Naïve Moving Average

Quadratic Probability Score (QPS) k = 1 k = 3 k = 6 k = 9 k = 12 DVD (g) 0.508 0.531 0.561 0.550 0.524 EAR (g) 0.521 0.507 0.547 0.549 0.607 RDVD (g) 0.489 0.596 0.658 0.675 0.641 REAR (g) 0.466 0.492 0.547 0.560 0.645 CAPE (∆) 0.368 0.408 0.457 0.479 0.491 TB3M (∆) 0.462 0.468 0.475 0.487 0.488 TN5Y (∆) 0.462 0.467 0.468 0.484 0.503 TN10Y (∆) 0.461 0.467 0.468 0.485 0.503 S5Y3M 0.448 0.446 0.445Malaya 0.446 0.447 S10Y3M 0.440 0.435 0.433 0.446 0.455 M1 (g) 0.543 0.485 0.478 0.509 0.494 M2 (g) 0.778 0.695of 0.564 0.487 0.479 RM1 (g) 0.468 0.523 0.467 0.523 0.525 RM2 (g) 0.497 0.488 0.477 0.487 0.490 INFL 0.460 0.465 0.467 0.472 0.480 RDGDP 0.433 0.453 0.477 0.487 0.486 ISMI 0.450 0.445 0.435 0.439 0.454 COPE (∆) 0.461 0.467 0.483 0.492 0.487 NPHP (∆) 0.460 0.467 0.476 0.485 0.486 CMUO 0.461 0.477 0.485 0.491 0.482 PMI 0.462 0.470 0.465 0.470 0.481 MSCI 0.424 0.462 0.540 0.567 0.581 CDLI (∆) 0.364 0.381 0.453 0.498 0.514 CFNAI 0.447 0.467 0.476 0.482 0.479 The quadratic probability score (QPS) proposed by Diebold and Rudebusch (1989) ranges fromUniversity 0 to 2 with a score of 0 corresponding to perfect accuracy.

The out-of-sample predictability test for predicting bear stock markets of the naïve moving average model shows that all values are lower than 0.800 (generally, most values are in the range of 0.400 to 0.500) which suggests that the variables proposed by the

243 research have good predictive power. Evidence shows that CDLI (∆) has the highest predictive power (lowest QPS score) for both the horizons of k = 1 (QPS = 0.364), k = 3

(QPS = 0.381) while S10Y3M has the highest predictive power for the k = 6 horizon (QPS

= 0.433) and joint highest with S5Y3M for the k = 9 horizon (QPS = 0.446). Further,

S5Y3M is the best predictive variable compared to the rest in the list for the k = 12 horizon

(QPS = 0.447).

4.6.5 Naïve Moving Average Negative Return Out-of-sample Predictability Test Results

Table 4.19: Out-of-sample Predictability Test Results for Semi-parametric Model: Naïve Moving Average Negative Return

Quadratic Probability Score (QPS) k = 1 k = 3 k =Malaya 6 k = 9 k = 12 DVD (g) 0.577 0.595 0.576 0.556 0.533 EAR (g) 0.585 0.616 0.598 0.590 0.632 RDVD (g) 0.615 0.750of 0.707 0.675 0.619 REAR (g) 0.581 0.646 0.633 0.629 0.675 CAPE (∆) 0.446 0.439 0.514 0.539 0.509 TB3M (∆) 0.501 0.510 0.518 0.515 0.509 TN5Y (∆) 0.498 0.506 0.511 0.519 0.518 TN10Y (∆) 0.496 0.504 0.508 0.520 0.520 S5Y3M 0.484 0.482 0.500 0.482 0.488 S10Y3M 0.469 0.466 0.500 0.488 0.501 M1 (g) 0.557 0.510 0.541 0.515 0.506 M2 (g) 0.735 0.734 0.497 0.488 0.470 RM1 (g) 0.493 0.541 0.581 0.541 0.542 RM2 (g) 0.520 0.510 0.532 0.534 0.519 INFL 0.493 0.498 0.508 0.496 0.499 RDGDP 0.484 0.502 0.523 0.515 0.512 ISMI 0.476 0.467 0.474 0.478 0.491 COPEUniversity (∆) 0.500 0.516 0.528 0.520 0.510 NPHP (∆) 0.501 0.508 0.520 0.515 0.510 CMUO 0.517 0.539 0.549 0.527 0.509 PMI 0.512 0.517 0.509 0.501 0.504 MSCI 0.568 0.640 0.689 0.640 0.606 CDLI (∆) 0.397 0.460 0.578 0.551 0.554 CFNAI 0.498 0.518 0.519 0.508 0.502 The quadratic probability score (QPS) proposed by Diebold and Rudebusch (1989) ranges from 0 to 2 with a score of 0 corresponding to perfect accuracy. 244

The out-of-sample predictability test for predicting bear stock markets of the naïve moving average negative return model shows that all values are lower than 0.750

(generally, most values are in the range of 0.400 to 0.500) which suggests that the variables proposed by the research have good predictive power. Evidence shows that CDLI (∆) has the highest predictive power (lowest QPS score) for the k = 1 horizon (QPS = 0.397) and

CAPE (∆) for the k = 3 horizon (QPS = 0.439). ISMI has the highest predictive power for both the horizons of k = 6 (QPS = 0.474) and k = 9 (QPS = 0.478) while M2 (g) is the best predictive variable for the k = 12 horizon (QPS = 0.470).

Malaya

University

245

4.6.6 Lunde & Timmermann’s B-B Algorithm Out-of-sample Predictability Test Results

Table 4.20: Out-of-sample Predictability Test Results for Non-parametric Model: Lunde & Timmermann’s B-B Algorithm

Quadratic Probability Score (QPS) k = 1 k = 3 k = 6 k = 9 k = 12 DVD (g) 0.449 0.494 0.483 0.474 0.475 EAR (g) 0.403 0.510 0.567 0.615 0.641 RDVD (g) 0.641 0.786 0.712 0.564 0.608 REAR (g) 0.398 0.543 0.620 0.626 0.693 CAPE (∆) 0.393 0.392 0.409 0.443 0.466 TB3M (∆) 0.430 0.438 0.450 0.462 0.470 TN5Y (∆) 0.430 0.438 0.448 0.472 0.475 TN10Y (∆) 0.429 0.439 0.451 0.480 0.477 S5Y3M 0.418 0.419 0.427 0.411 0.444 S10Y3M 0.402 0.403 0.418 0.404 0.441 M1 (g) 0.482 0.449 0.450 0.461 0.479 M2 (g) 0.703 0.596 0.516 0.490 0.467 RM1 (g) 0.417 0.488 0.466 0.488 0.561 RM2 (g) 0.447 0.440 0.449Malaya 0.468 0.496 INFL 0.431 0.432 0.443 0.456 0.474 RDGDP 0.394 0.420 0.443 0.465 0.481 ISMI 0.418 0.412of 0.399 0.390 0.409 COPE (∆) 0.428 0.437 0.462 0.463 0.486 NPHP (∆) 0.431 0.437 0.448 0.459 0.474 CMUO 0.433 0.452 0.463 0.464 0.468 PMI 0.435 0.454 0.456 0.458 0.469 MSCI 0.428 0.494 0.579 0.619 0.703 CDLI (∆) 0.331 0.352 0.429 0.471 0.580 CFNAI 0.412 0.437 0.454 0.458 0.462 The quadratic probability score (QPS) proposed by Diebold and Rudebusch (1989) ranges from 0 to 2 with a score of 0 corresponding to perfect accuracy.

The out-of-sample predictability test for predicting bear stock markets of the Lunde & Timmermann’sUniversity B-B algorithm shows that all values are lower than 0.710 (generally, most values are in the range of 0.400 to 0.500) which suggests that the variables proposed by the research have good predictive power. Evidence shows that CDLI (∆) has the highest predictive power (lowest QPS score) for both the horizons of k = 1 (QPS = 0.331), k = 3

246

(QPS = 0.352) while ISMI has the highest predictive power for the horizons of k = 6 (QPS

= 0.399), k = 9 (QPS = 0.390) and k = 12 (QPS = 0.409).

4.6.7 Candelon, Piplak & Straetmans’ B-B Algorithm Out-of-sample Predictability Test Results

Table 4.21: Out-of-sample Predictability Test Results for Non-parametric Model: Candelon, Piplak & Straetman’s B-B Algorithm

Quadratic Probability Score (QPS) k = 1 k = 3 k = 6 k = 9 k = 12 DVD (g) 0.567 0.580 0.558 0.541 0.523 EAR (g) 0.577 0.597 0.580 0.578 0.617 RDVD (g) 0.621 0.760 0.695 0.656 0.608 REAR (g) 0.576 0.631 0.615 0.616 0.659 CAPE (∆) 0.422 0.467 0.519 0.537 0.510 TB3M (∆) 0.500 0.510 0.521 0.517 0.510 TN5Y (∆) 0.497 0.505 0.514 0.521 0.518 TN10Y (∆) 0.494 0.503 0.513Malaya 0.522 0.521 S5Y3M 0.472 0.473 0.498 0.480 0.490 S10Y3M 0.448 0.450 0.498 0.487 0.502 M1 (g) 0.568 0.511of 0.548 0.517 0.505 M2 (g) 0.767 0.766 0.499 0.485 0.460 RM1 (g) 0.493 0.545 0.592 0.545 0.543 RM2 (g) 0.522 0.509 0.537 0.535 0.515 INFL 0.493 0.488 0.503 0.499 0.505 RDGDP 0.487 0.505 0.524 0.515 0.507 ISMI 0.461 0.453 0.467 0.471 0.490 COPE (∆) 0.500 0.517 0.530 0.521 0.511 NPHP (∆) 0.502 0.512 0.523 0.517 0.511 CMUO 0.516 0.536 0.539 0.518 0.506 PMI 0.504 0.506 0.506 0.501 0.503 MSCI 0.561 0.630 0.689 0.644 0.612 CDLI (∆) 0.416 0.474 0.570 0.546 0.551 CFNAI 0.499 0.513 0.516 0.507 0.500 The quadratic probability score (QPS) proposed by Diebold and Rudebusch (1989) ranges fromUniversity 0 to 2 with a score of 0 corresponding to perfect accuracy.

The out-of-sample predictability test for predicting bear stock markets of the

Candelon, Piplak & Straetman’s B-B algorithm shows that all values are lower than 0.770

(generally, most values are in the range of 0.400 to 0.600) which suggests that the variables

247 proposed by the research have good predictive power. Evidence shows that CDLI (∆) has the highest predictive power (lowest QPS score) for the k = 1 horizon (QPS = 0.416) and

S10Y3M for the k = 3 horizon (QPS = 0.450). ISMI has the highest predictive power for both the horizons of k = 6 (QPS = 0.467) and k = 9 (QPS = 0.471) while M2 (g) is the best predictive variable for the k = 12 horizon (QPS = 0.460).

4.6.8 JLS and JLS “Negative Bubbles” Integrated Identifications Out-of-sample Predictability Test Results

Table 4.22: Out-of-sample Predictability Test Results for Scale-invariant Model: JLS and JLS “Negative Bubbles” (6-month Window)

Quadratic Probability Score (QPS) k = 1 k = 3 k = 6 k = 9 k = 12 DVD (g) 0.579 0.630 0.725 0.744 0.679 EAR (g) 1.136 1.130 0.812Malaya 0.405 0.459 RDVD (g) 0.638 0.637 0.584 0.596 0.494 REAR (g) 1.039 1.038 0.712 0.382 0.544 CAPE (∆) 0.392 0.475of 0.541 0.540 0.460 TB3M (∆) 0.568 0.538 0.517 0.550 0.548 TN5Y (∆) 0.563 0.537 0.531 0.527 0.556 TN10Y (∆) 0.567 0.532 0.531 0.533 0.547 S5Y3M 0.457 0.478 0.542 0.464 0.582 S10Y3M 0.511 0.503 0.551 0.591 0.637 M1 (g) 0.553 0.523 0.547 0.538 0.520 M2 (g) 0.602 0.644 0.525 0.530 0.556 RM1 (g) 0.511 0.592 0.521 0.592 0.581 RM2 (g) 0.648 0.703 0.571 0.606 0.645 INFL 0.543 0.547 0.525 0.540 0.582 RDGDP 0.557 0.541 0.452 0.393 0.414 ISMI 0.603 0.554 0.541 0.499 0.385 COPE (∆) 0.558 0.546 0.507 0.513 0.486 NPHP (∆) 0.543 0.531 0.524 0.509 0.527 CMUOUniversity 0.501 0.533 0.524 0.489 0.460 PMI 0.571 0.545 0.491 0.489 0.419 MSCI 0.650 0.627 0.541 0.525 0.511 CDLI (∆) 0.447 0.356 0.305 0.329 0.371 CFNAI 0.549 0.537 0.402 0.333 0.343 The quadratic probability score (QPS) proposed by Diebold and Rudebusch (1989) ranges from 0 to 2 with a score of 0 corresponding to perfect accuracy.

248

The out-of-sample predictability test for predicting bubble induced-crashes (6- month window) of the JLS model and JLS “negative bubbles” model shows that generally, most values are in the range of 0.400 to 0.600 which suggests that the variables proposed by the research have good predictive power although variables of EAR (g) and REAR (g) exceeded 1.0 for some horizons. Evidence shows that CAPE (∆) has the highest predictive power (lowest QPS score) for the k = 1 horizon (QPS = 0.392). CDLI (∆) has the highest predictive power for the horizons of k = 3 (QPS = 0.356), k = 6 (QPS = 0.305) and k = 9

(QPS = 0.329) while CFNAI is the best predictive variable for the k = 12 horizon (QPS =

0.343).

4.7 Concluding Remark Malaya The chapter begins with the descriptive statistics of test variables that are selected for the research. The illustrations in charts can beof found in Appendix B where the time series movement of each test variable is paired with the movement of the S&P 500 index and next, paired with the movement of the S&P 500 return over the period of April 1967 to

June 2014. The visual inspection generally show that all stock market fundamentals (i.e.

DVD, EAR, RDVD, REAR and CAPE), selected industrial indicators (i.e. ISMI and

COPE) and selected variable from the category of indexes of leading indicators (namely

CDLI (∆)), track the movement of the S&P 500 index quite well.

UniversityFurther on, only four variables appear to be correlated with the S&P 500 return; the financial market fundamental of RGDPG, the industrial indicator of ISMI, the market sentiment of PMI and the index of leading indicator of CFNAI. It is important to note that 4 other variables – the financial market fundamentals of S5Y3M and S10Y3M and the industrial indicators of NPHP and CMUO seem to track the movement of the S&P 500 249 index closely only after year 2000s based on general observation. Further investigation on these prepositions is beyond the proposed scope of research here but such trends may vindicate a more comprehensive study in the future.

Subsequent to descriptive statistics and the testing of stochastic assumptions for the research’s variables and the rectification of the problem of unit root, this chapter furnishes a detailed explanation on how the data set is divided into the in-sample and out-of-sample segment for predictive regression (with probit) based on the Clark & West (2007).

Results of the modelled stock market declines (bull and bear markets, and bubble induced crashes and rebounds) are illustrated in charts at the back drop of the time series of the S&P 500 index. It is noteworthy that the presentation in charts, albeit not groundbreaking, is important but not found in previousMalaya literatures in related areas, e.g. Chang (2009), Chen (2009), Candelon, Piplakof & Straetmans (2008), Gonzalez, Powell, Shi, Wilson (2005), Maheu, McCurdy & Song (2009), Lunde & Timmermann (2004).

The graphical illustrations allow the research to portray clearly the efficacy of models in identifying stock market declines and enable easier visual comparisons across. In summary, the Markov-switching model produces three sets of outcomes for further analyses, i.e. the filtered probabilities and the smoothed probabilities (both parametric) and the research’s very own innovation, the dichotomised filtered probabilities (semi- parametric).University The charts also display the results of two other semi-parametric model, namely the naïve moving average model and the naïve moving average negative return model (another innovation of the research), followed by two variants of the B-B algorithm i.e. and the the

250

Lunde & Timmermann’s variant and the Candelon, Piplak & Straetmans’ variant. These are followed by the charts of the JLS’ crashes and the JLS’ rebounds.

Perceptibly, results that are derived from the parametric models particularly the

Markov-switching’s filtered probabilities are fuzzier compared to the rest. Charts in greyscale shades representing phases of stock market declines are generated for all models except those in which data are extracted from the Markov-switching (i.e. filtered probabilities and smoothed probabilities). These series of data which are in fractions instead of the binary 1 (for bear markets / crashes) or 0 (for bull markets / rebounds) are illustrated in time series line charts.

The general rule of thumb for out-of-sample analysis has been the use of two-thirds of the observations for modelling and one-third of theMalaya remaining observations for out-of- sample predictive analysis. However, in of most academic literatures, the division of observations are arbitrary (e.g. P/R = 0.5 in Kumar. Moorthy & Perraudin, 2002 and P/R =

1 in Rapach & Wohar, 2006). Likewise, Chen (2009) followed Clark & West (2007) discretion by setting the P/R at 4 but noted that results for other combinations such as P/R =

0.5, or 1, or 2, or 3 should be comparable.

The research sets the P/R at 2 (i.e. the mid value between the recommended P/R of

0.5, 1, 2, 3 and 4) as the research reckons that the utilisation of only one-fifth of the observationsUniversity to estimate the model parameters for predicting the rest of the four-fifth remaining period could be inadequate. By setting P/R at 4 for example, when k = 1, the in- sample observations would have only included the period of 1967.05 to 1976.10 (9 years and 5 months) to predict the period from 1976.11 to 2014.05 (37 years and 6 months).

251

Examining each market fundamental of the research with in-sample predictability test, CDLI (∆) and CAPE (∆) are consistently found to be the best predictors for stock market declines across the shorter horizons i.e. the k = 1 and the k = 3 horizon. Based on the, 푅2 or Pseudo- 푅2 (accordingly to the type of models), both ISMI and CDLI (∆) generally have better goodness-of-fit for the middle term horizons of k = 6 and the k = 9 while S10Y3M in general also produce the best results for the longer horizon of k = 12.

Results for the JLS model and JLS “negative bubbles” model (integrated identifications) show S5Y3M to have the highest goodness-of-fit for the shorter horizons i.e. the k = 1 and the k = 3 horizon while CDLI (∆) is the best predictors of sharp stock declines for the middle term horizons of k = 6 and the k = 9 and CFNAI is the most reliable market fundamental for the longer term horizon of k = 12. Malaya Results for the out-of-sample test in general show that DVD (g) has the highest MSPE-adj value from the middle term horizonsof onwards, i.e. k = 6, k = 9 and k = 12 for the parametric models while CDLI (∆) has the lowest QPS value across most semi-parametric and non-parametric models (including the econophysics scale-invariant model of JLS), notwithstanding the length of horizon. Further discussion is presented in Chapter 5.

University

252

CHAPTER 5

DISCUSSION AND CONCLUSION

5.0 Introduction

The identification of stock market declines is an important and challenging task for both investors and researchers as billions of dollars could be lost when the market slumps.

The crude terms of “stock market crashes” and “bear markets” have been in in use by stock market traders from the onset. As reviewed in the earlier chapters, the occurrence of stock market crashes from the viewpoint of orthodox financialMalaya economics is deemed as statistical outliers (Johansen & Sornette, 1998b; 1998c; Mandelbrot 2004) while the run of bull and bear markets is dismissed as statistical artefactof (Bakshi & Madan, 1998) and random phenomena (Gonzalez, Powell, Shi & Wilson, 2005). The quest of finding an unequivocal definition for both the stock market crashes and the bear markets is still a work in progress, even among schools of thought that espouse the concepts.

This research continues the undertaking of literature extension in the area of stock market declines by examining various specifications of models from different schools of thought and then investigates the most significant and consistent predictive variables in predictingUniversity these specifications. The predictive variables employed for the research consist of well-studied market fundamentals commonly used to predict recessions and financial crises. This general objective of the research is dissected into specific objectives in Chapter

1. The justifications to examine stock market declines with market fundamentals which inextricably links the research with economic environment are meticulously elucidated in 253 the previous chapters. As a reiteration, Siegel (2008) encapsulated succinctly the relationship between stock market declines and the economy, showing evidence that nine over ten of recessions in the history are either preceded or succeeded by a stock market decline.

5.1 Recapitulation of Research

The central focus of the research is on stock market declines. The research is structured on the pillars of eight objectives contrived to fulfil the proposed scope of study. Prior to the pursuance of the research objectives as outlined in Chapter 1, it is noteworthy that the research also undertakes a meticulous exertion in bridgingMalaya the theoretical gaps on the worldview of stock market declines across various schools of thought. The literature review of the research features a meticulous examinationof on the etymology, history, scientific paradigms and the contention on various issues among differing views. Synthesis in these sections are critically examined to lay the foundation for the classification of literatures and the theoretical framework of the research. The latest advancements in methodologies from relevant schools of thought are screened where approaches that are compatible with the research are selected to structure the empirical framework.

Notwithstanding that this undertaking ostensibly embodies the requisite literature reviewUniversity which is standard for a thesis, the rigorousness of the reconciliatory pursuit goes beyond the customary synthesis and discussion of past findings to justify the need for the research. The contributions to the literature that include in the examination of the etymology of definitions, which are commonly used interchangeably to describe the

254 phenomena of stock market “declines” (e.g. crashes, cascades, drawdowns and bear markets) and the critical review on the diverse descriptions of subject in the literature.

The theoretical reconciliation begins with a meticulous study on the evolution of literatures over the course of history on the worldview of stock market declines across major schools of thought. This includes chronicling how the topic of stock market declines began from retrospective analyses and case studies on specific episodes to the espousal of falsification philosophy that entails perpetual theoretical debates across major schools of thought. Such debates, as the research discovered, were mainly centralised on two major disputes, i.e. on whether the behaviour of investors are completely rational even in the event of abrupt declines (crashes) and the veracity of market bubbles in reality.

As a recap to the earlier review of literature, ArthurMalaya (1995) asserted that endogenous crashes are caused by the market psychologyof which leads to herding, market bubble and ultimately panic selling (all of which do not conform with the concept of rationality in conventional financial theory). The intricacy of the market is further compounded by the complex interactions of investors, major institutional fund managers, particularly those who engaged in a guessing game of trading decisions among themselves which creates feedbacks in every trading action and reaction in the market. The feedback loop of actions in the market is a notable example of a self-organising system in the context of complex systems theory. These feedback loops entail herding and ultimately leads to bubble in the priceUniversity level of stocks.

Most studies on the subject of stock market crashes (or remotely through the rational bubbles and speculative bubbles schools of thought) espouse the inquiry paradigm of methodological individualism that is similar to that of Arthur (1995). Empirical

255 investigation on the endogenous or bubble-induced crashes is led by the cross disciplinary econophysics which grows to prominence from the mid-2000s. The section that covers this subject in the research highlights the emergence of the complex systems theory, which is the catalyst to the LPPL frontier (i.e. the JLS model and its variants) for the modelling of bubble-induced crashes in the financial markets. Independently and contemporaneously, literature that examine the switching regimes of bull and bear markets began to expand, albeit to a lesser extent. These studies that generally model the extended upward and downward trend of the stock market with Gaussian-based econometric methodologies are classified by the research as the cyclical market school of thought.

The research acknowledges that the schools of thought as outlined above transcend the perpetual theoretical deliberation on the behaviourMalaya of agents in the stock market. Studies with such approaches that focus solely on the aim of generalising the phenomenon of stock market declines axiomatically couldof in principal be congregated and examined across on the same platform. This leads to the contrivance of the research objectives that conjointly examine various approaches to ascertain the most powerful and consistent market fundamentals for ex-ante prediction on ex-post models for stock market declines.

5.2 Discussion

ReiteratingUniversity the scope of study as defined in Chapter 1, the principle investigation of the research is on the efficacy of market fundamentals in predicting stock market declines instead of models comparison. Modelling of stock market declines is a process of axiomatisation of the objective specifications that can be developed into functional algorithms to identify heterogeneous regimes in the stock market consistently. The

256 calibration of specifications therefore could be arbitrary. For example, various variants of the Markov-switching model can be found in the literature (e.g. see Chang, 2009) but the

Markov-switching model chosen for the research is the one that is the most parsimonious.

Likewise, there are also a number of permutations to the Brys-Boschan algorithm apart from the Candelon, Piplak & Straetmans (2008) variants and the Lunde & Timmermann

(2004) used for the research (e.g. see Pagan & Sossounov, 2003).

Similar to other time series models like GARCH (and its variants), permutations to the basic model are calibrated to achieve different objectives. Thus the argument of which model is better is subjective and rather irrelevant. Within the research for example, the

Brys-Boschan algorithm of the Lunde & Timmermann (2004) variant is premeditatedly specified to be rigid to obtain a more univocal identificationMalaya of regimes. As a result, the model is less sensitive and identifies less frequent switching of regimes compared to the Candelon, Piplak & Straetmans (2008). Thus,of it is unjustified to conclude that the Lunde & Timmermann (2004) variant is inferior.

On another note, the in-sample observations timeframe determined based on the

Clark & West (2007) criteria of P/R = 2 has enabled more test variables reflecting the market fundamentals to be included in the research (e.g. CDLI is only available from

1959.01; CMUO from 1959.01; NPHP from 1960.01; and CFNAI from 1967.03). Such calibration is also important in extending the literature by Chen (2009) which set the in- sampleUniversity period from 1957.02 to 1967.02 for the estimation of the model’s parameter for the predictive regression and probit model. The research’s in-sample period is set from 1967.05 to 1982.02 that allows the identification of the well-documented bear markets55 in the early

55 These periods of bear markets are identified by most models used for the research. Details of the identifications can be found in Chapter 4. 257 years, i.e. for the periods of 1966-1967, 1969-1970, 1973-1974, 1976-1978, 1980-1982.

The ensuing discussion on the empirical results follows the sequence of the research objectives as listed in Section 1.4 in Chapter 1.

Malaya of

University

258

Table 5.1: Summary of In-sample Results for all Models: Highest Goodness of Fit, 푅2 or Pseudo- 푅2 for All Horizons

Models / Horizons k = 1 k = 3 k = 6 k = 9 k = 12

Parametric Models Markov-switching (filtered CDLI (∆) CDLI (∆) CDLI (∆) CDLI (∆) CDLI (∆) probabilities) (푅2= 0.220) (푅2= 0.196) (푅2= 0.151) (푅2= 0.103) (푅2= 0.054) Markov-switching (smoothed CDLI (∆) CDLI (∆) CDLI (∆) S10Y3M S10Y3M probabilities) (푅2= 0.280) (푅2= 0.248) (푅2= 0.151) (푅2= 0.180) (푅2= 0.223) Semi-parametric Models Markov-switching (dichotomized CDLI (∆) CDLI (∆) CDLI (∆) S10Y3M S5Y3M 2 2 2 2 2 smoothed probabilities) (푝푅푀= 0.218) (푝푅푀= 0.186) (푝푅푀= 0.144) (푝푅푀= 0.139) (푝푅푀= 0.179) 2 2 Malaya2 2 2 (푝푅퐷&푅= 0.224) (푝푅퐷&푅= 0.191) (푝푅퐷&푅= 0.149) (푝푅퐷&푅= 0.144) (푝푅퐷&푅= 0.185) Naïve moving average CAPE (∆) CAPE (∆) ISMI ISMI S10Y3M 2 2 2 2 2 (푝푅푀= 0.142) (푝푅푀= 0.096) (푝푅푀= 0.036) (푝푅푀= 0.34) (푝푅푀= 0.041) 2 2 2 2 2 (푝푅퐷&푅= 0.175) (푝푅퐷&푅= 0.119) of( 푝푅퐷&푅= 0.045) (푝푅퐷&푅= 0.43) (푝푅퐷&푅= 0.052) Naïve moving average negative CDLI (∆) CDLI (∆), CAPE (∆) CAPE (∆) ISMI S10Y3M 2 2 2 2 2 return (푝푅푀= 0.118) (푝푅푀= 0. 085) (p푅푀= 0.040) (푝푅푀= 0.039) (푝푅푀= 0.039) 2 2 2 2 2 (푝푅퐷&푅= 0.144) (p푅퐷&푅= 0.104) (푝푅퐷&푅= 0.049) (푝푅퐷&푅= 0.048) (푝푅퐷&푅= 0.049) Non-parametric Models Lunde & Timmermann’s B-B CDLI (∆) CDLI (∆), CAPE (∆) ISMI ISMI S10Y3M 2 2 2 2 2 Algorithm (푝푅푀= 0.118) (푝푅푀= 0. 085) (푝푅푀= 0.045) (푝푅푀= 0.039) (푝푅푀= 0.039) 2 2 2 2 2 (푝푅퐷&푅= 0.144) (p푅퐷&푅= 0.104) (푝푅퐷&푅= 0.056) (푝푅퐷&푅= 0.048) (푝푅퐷&푅= 0.049) Candelon, Piplak & Straetman’s CAPE (∆) CAPE (∆) ISMI S10Y3M S10Y3M 2 2 2 2 2 B-B Algorithm (푝푅푀= 0.114) (푝푅푀= 0.042) (푝푅푀= 0.037) (푝푅푀= 0.030) (푝푅푀= 0.030) 2 2 2 2 2 (푝푅퐷&푅= 0.138) (푝푅퐷&푅= 0.051) (푝푅퐷&푅= 0.046) (푝푅퐷&푅= 0.037) (푝푅퐷&푅= 0.037) Scale-invariant Model JLS Model & JLS "negative S5Y3M S5Y3M CDLI (∆) CDLI (∆) CFNAI 2 2 2 2 2 bubble" Model (integrated (푝푅푀= 0.255) (푝푅푀= 0.286) (푝푅푀= 0.306) (푝푅푀= 0.260) (푝푅푀= 0.194) 2 2 2 2 2 identifications) (푝푅퐷&푅= 0.289) (푝푅퐷&푅= 0.324) (푝푅퐷&푅= 0.346) (푝푅퐷&푅= 0.295) (푝푅퐷&푅= 0.221) University 2 2 2 2 Note: 푝푅푀 denotes pseudo- 푅 (Mcfadden), and 푝푅퐷&푅 denotes pseudo- 푅 (Diebold & Rudebusch). All values are significant at the 5% level.

259

Table 5.2: Summary of Out-of-sample Results of all Models: Highest MSPE-adj or Lowest QPS for All Horizons

Models / Horizons k = 1 k = 3 k = 6 k = 9 k = 12

Parametric Models Markov-switching (filtered - - DVD (g) DVD (g) DVD (g) probabilities) (푀푆푃퐸푎푑푗= 15.347) (푀푆푃퐸푎푑푗= 15.194) (푀푆푃퐸푎푑푗= 14.977) Markov-switching (smoothed - TN10Y (∆) DVD (g) DVD (g) DVD (g) probabilities) (푀푆푃퐸푎푑푗= 4.551) (푀푆푃퐸푎푑푗= 17.130) (푀푆푃퐸푎푑푗= 16.922) (푀푆푃퐸푎푑푗= 16.732) Semi-parametric Models Markov-switching (dichotomized CDLI (∆) CDLI (∆) CDLI (∆) S5Y3M S5Y3M smoothed probabilities) (QPS= 0.215) (QPS= 0.221) (QPS=Malaya 0.267) (QPS= 0.275) (QPS= 0.263) Naïve moving average CDLI (∆) CDLI (∆) S10Y3M S10Y3M, S5Y3M S5Y3M (QPS= 0.364) (QPS= 0.381) (QPS= 0.433) (QPS= 0.446) (QPS= 0.447) Naïve moving average negative CDLI (∆) CAPE (∆) ISMI CDLI (∆) M2 (g) return (QPS= 0.397) (QPS= 0.439) of (QPS= 0.474) (QPS= 0.478) (QPS= 0.0470) Non-parametric Models Lunde & Timmermann’s B-B CDLI (∆) CDLI (∆) ISMI ISMI ISMI Algorithm (QPS= 0.331) (QPS= 0.352) (QPS= 0.399) (QPS= 0.390) (QPS= 0.409) Candelon, Piplak & Straetman’s CDLI (∆) S10Y3M ISMI ISMI M2 (g) B-B Algorithm (QPS= 0.416) (QPS= 0.450) (QPS= 0.467) (QPS= 0.471) (QPS= 0.460)

Scale-invariant Model JLS Model & JLS "negative CAPE (∆) CDLI (∆) CDLI (∆) CDLI (∆) CDLI (∆) bubble" Model (integrated (QPS= 0.392) (QPS= 0.356) (QPS= 0.305) (QPS= 0.329) (QPS= 0.343) identifications)

Note: 푀푆푃퐸푎푑푗 denotes MSPE adjusted, and QPS denotes quadratic probability score. All values are significant at the 5% level. University

260

Table 5.1 illustrates the in-sample predictability results for all models for the horizon of k = 1, k = 3, k = 6, k = 9, and k = 12 while Table 5.2 shows out-of-sample predictability results for the same horizon. These models are categorised as parametric, semi- parametric, non-parametric and scale-invariant accordingly. As a recap, the in-sample results are ranked based on the highest values of goodness of fit measured with 푅2 for parametric models. The Mcfadden pseudo 푅2and the Diebold & Rudebusch pseudo 푅2 are used for semi-parametric and non-parametric models as results produced by these models are in categorical values. Likewise, the out-of-sample are ranked based on the highest of MSPE-adj for parametric models and QPS for semi-parametric and non- parametric models.

5.2.1 Markov-switching Filtered Probabilities Malaya

The results of the in-sample predictability oftest for the filtered probabilities derived from the parsimonious Markov-switching model show that the CDLI (∆) has the highest predictive power, consistently for all horizons of k = 1, k = 3, k = 6, k = 9, and k = 12.

The DVD (g) is ascertained to be the most consistent test variable with the highest predictive power across the forecasting horizons of k = 6, k = 9, and k = 12. No significant test variable is found for the k = 1, k = 3 horizons. Rapach & Wohar (2006) noted that disparity between in-sample and out-of-sample predictability test results is commonUniversity particularly for studies on stock. Between the two, the out-of-sample test is preferred in most studies (e.g. see Chen, 2009; Qi, 2001) as it mitigates the problem of data mining (Rapach & Wohar, 2006).

The results of the Markov-switching model (filtered probabilities) indicate that the Shiller’s Financial Variables originally used to examine stock market returns (see

Campbell & Shiller, 1998) and the Estrella & Mishkin’s Financial Variables first 261 proposed for the forecasting financial crises and recessions (see Estrella & Mishkin,

1998; Qi, 2001) are also efficient in predicting the heterogeneous switching of regimes for stock market. The CDLI (∆) test variable proposed by Estrella & Mishkin (1998) is grouped in the Indexes of Leading Indicators category for the research while the DVD

(g) proposed by Shiller (2005) is grouped in the Stock Market Fundamentals category.

Therefore, the results show that market fundamentals in their most basic form e.g. DVD

(g) is still reliable for forecasting compared along highly structured ones such as index of economic leading indicators.

5.2.2 Markov-switching Smoothed Probabilities The results of the in-sample predictability test for Malaya the smoothed probabilities derived from the parsimonious Markov-switching model show that the CDLI (∆) has the highest predictive power for the shorter forecastingof horizons of k = 1, k = 3, k = 6 while the S10Y3M has the highest predictive power for the longer horizons of k = 9, and k = 12.

Similar to the result of the Markov-switching model (filtered probabilities), the DVD

(g) is found to be the most consistent test variable with the highest predictive power across the forecasting horizons of k = 6, k = 9, and k = 12. While no significant test variable is determined with the former model for the k = 1 and k = 3 horizons, the

Markov-switching model (smoothed probabilities) finds TN10Y (∆) to have the highest predictiveUniversity power for the k = 3 horizon. No significant test variable is found for the forecasting horizon of k = 1.

The results of the Markov-switching model (smoothed probabilities) for both in- sample and out-of-sample are consistent with previous studies in economic forecasting which showed evidence of predictive power for longer term rates of Treasury bills (e.g.

Keilis-Borok, Soloviev, Allègre, Sobolevskii & Intriligator, 2005; Nyberg, 2011; Qi, 262

2001) and term spreads (e.g. Resnick & Shoesmith, 2002; Hartmann, Kempa &

Pierdzioch, 2008; Estrella & Trubin, 2006). At the time of writing, the research did not find similar studies that use long-term Treasury rates to predict the stock market but there is ample evidence that showed significant relationship between term spread (10- year minus 3-month) with stock returns (e.g., see Avramov & Chordia, 2006; Bernanke

& Kuttner, 2005; Flannery & Protopapadakis, 2002). Both the test variables of TN10Y

(∆) and S10Y3M proposed by Estrella & Mishkin (1998) are grouped in the Financial

Market Fundamentals category.

5.2.3 Markov-switching Dichotomised Smoothed Probabilities The research opines that the predictability analysis’ Malayatest statistic used in a related study by Chen (2009) for the Markov-switching model (i.e. MSPE-adj) is unfitting for side- by-side comparison with the QPS, a test statisticof which is exclusive for non-parametric models. The research thus introduces a modest innovation to the model by dichotomising the smoothed probabilities at the threshold of 0.5. The simple modification changes the predictability test on the Markov-switching model outputs from a parametric procedure to a semi-parametric procedure. The outputs are tested with a probit model instead of a predictive regression model and subsequently compared in parallel with the outputs from other semi-parametric and non-parametric models.

UniversityThis innovated model yields some very interesting results. The in-sample findings are almost identical with the Markov-switching model (smoothed probabilities) i.e. the most powerful predictors identified are the CDLI (∆) for the k = 1, k = 3, k = 6 horizons and the S10Y3M for the k = 9 horizon. This is an exception for the forecasting horizon of k = 12 where the test variable with the highest predictive power is found to be S5Y3M compared to the former model. On the other hand, the out-of-sample test 263 results are similar to the in-sample test results for the k = 1, k = 3, k = 6 horizons i.e.

CDLI (∆) as the best predictor and for the k = 12 horizon where the highest predictive power variable is also S5Y3M. Likewise, the test variable of S5Y3M is also found to have the highest predictive power for the k = 9 horizon for the out-of-sample test. More interestingly, the out-of-sample findings for the Markov-switching model (smoothed probabilities) are similar to that of the Naïve moving average model which is discussed in the following.

5.2.4 Naïve Moving Average

The results of the in-sample predictability test for the naïve moving average model show that CAPE (∆) is the most powerful predictor forMalaya the shorter horizons of k = 1 and k = 3 while ISMI predicts the k = 6 and k = 9 horizons better than other test variables. The S10Y3M, one of the most efficient leadingof indicators to detect economic turning points as discussed prior (with citations), is found be the most powerful predictors for the k = 12 horizon of the in-sample test. CAPE (∆) is grouped in Stock Market

Fundamentals category for the research. The result conforms to other studies that showed evidence of predictive efficacy of CAPE for stock market returns (e.g. Shiller &

Campbell, 1998; Kiemling, 2016; Siegel, 2016).

ISMI is one of the test variables classified in the Industrial Indicators category of University the research. The use of ISMI as one of the test variables in replacement of the vendor performance, slower deliveries diffusion index is exploratory for the research

(see justification in Chapter 3). Indicators of industrial inventories are considered proxies to the future production in the economy and thus lead the business cycle in general (see Qi, 2001). The result indicates that a market fundamental that reflects the

264 health of the industry can also be useful to predict the switching of regimes in the stock market.

The out-of-sample predictability test for the naïve moving average model yields a set of results that is almost identical to the Markov-switching model (dichotomised smoothed probabilities) as presented in the preceding section, except for the k = 6 horizon. CDLI (∆) is found to have the highest predictive power for the k = 1 and k = 3 horizons while S10Y3M is found to have the highest predictive power for the k = 6 horizon. The preceding model finds CDLI (∆) as the best predictor for this horizon.

Likewise, the naïve moving average model also finds the S5Y3M as the most powerful predictors for both the horizons k = 9 and k = 12. Nonetheless, the model also finds

S10Y3M as the joint most powerful predictors for the k = 9 while the aforementioned Markov-switching model only identifies S5Y3M. Malaya of 5.2.5 Naïve Moving Average Negative Return

The result of the bear and bull markets identification derived from the naïve moving average model shows a more frequent change of regimes in relative to comparable models. As explained previously, the research discovered that the basic model could be easily skewed by outliers. The augmentation proposed by the research is aimed to decrease the volatility of regimes switching and to increase the detection sensitivity of stockUniversity market declines (see Chapter 3).

The in-sample predictability test for the augmented model shows the CDLI (∆) has the highest predictive power for the k = 1 horizon; CDLI (∆) and CAPE (∆) are the joint most powerful predictors for the k = 3 horizon; CAPE (∆) is the best predictor for the k = 6 horizon; ISMI is the most significant test variable in predicting the k = 9

265 horizon; and S10Y3M best predicts the k = 12 horizon. The in-sample results are encouraging as they are almost identical to the findings of the more complex and rigid non-parametric model of the Lunde & Timmermann’s B-B algorithm, except for the k =

6 horizon (which is discussed in the following).

The out-of-sample predictability results find the CDLI (∆), CAPE (∆), ISMI,

CDLI (∆) and M2 (g) best predict the respective horizons of k = 1, k = 3, k = 6, k = 9 and k = 12. The robustness of the model is confirmed again in the out-of-sample test as the results are similar to another non-parametric model i.e. the Candelon, Piplak &

Straetman’s B-B algorithm, except for the k = 3 and k = 6 forecasting horizons. Only both of these two models find M2 (g) with the highest predictive power, specifically for the k = 12 horizon. The efficacy of M2 (g) in predicting the directional change stock market for the longer forecasting horizon is consistentMalaya with the early study by Chen (2009). of From the macroeconomic perspective, broad money expansion in the economy is a common measure implemented by the Federal Reserve to avert financial crisis or recession (Carpenter & Demiralp, 2010). The out-of-sample evidence showing the M2

(g) as a leading indicator for stock market declines of the longer forecasting horizon may indicate that the expansionary monetary mechanism of the Federal Reserve was already in operation in most precedents but fell short to deter the imminent downturns.

University

5.2.6 Lunde & Timmermann’s B-B Algorithm

At the time of writing, there are three variants of B-B algorithm for the bear markets identification, all of which are adapted from the original specification used for the dating of recessions. The earliest framework adaption was proposed by Pagan &

266

Sossounov (2003). Another two algorithms that subsequently emerged are proposed by

Lunde & Timmermann (2004) and Candelon, Piplak & Straetmans (2008). The B-B algorithm adaption by Pagan & Sossounov (2003) is not included for examination because the research reckons that both the later B-B algorithms have incorporated some fine-tunings to the original adaption. In comparison, the Lunde & Timmermann’s variant of the B-B algorithm shows a more rigid identification result compared to the later algorithm (see 4.4.7 and 4.4.8 in Chapter 4).

The in-sample predictability test for periods of stock market declines of the

Lunde & Timmermann’s B-B algorithm shows that the most significant variables corresponding to the forecasting horizons are CDLI (∆) for k = 1, CDLI (∆) and CAPE

(∆) for k = 3, ISMI for k = 6, ISMI for k = 9 and S10Y3M for k = 12. As noted previously, the results are interestingly similar to Malaya the enhanced model of the naïve moving average negative return introduced by the research. The advantage of the innovation is in the simpler algorithm formulationof which could be particularly useful for domestic traders. The out-of-sample test finds CDLI (∆) to be the best predicts the shorter k = 1 and k = 3 while ISMI is the best indicator of market declines for the longer horizons of k = 6, k = 9 and k = 12.

5.2.7 Candelon, Piplak & Straetmans’ B-B Algorithm

TheUniversity Candelon, Piplak & Straetmans (2008) variant of B-B algorithm is the least rigid variant compared to the other variants as mentioned in the above. The specification is simpler and detects more frequent regimes switching compared to the Lunde &

Timmermann’s B-B variant.

267

The in-sample predictability results show the best predictors for the corresponding forecasting horizons of k = 1, k = 3, k = 6, k = 9 and k = 12 are CAPE

(∆), CAPE (∆), ISMI, S10Y3M and S10Y3M. The in-sample results are almost the same compared to the naïve moving average model except for the k = 9 horizon where

ISMI is found to be the more efficient test variable instead. The out-of-sample predictability results show the most powerful predictor for the k = 1 horizon is CDLI

(∆); k = 3 is S10Y3M; both k = 6 and k = 9 is ISMI; and k = 12 is M2 (g).

5.2.8 JLS model & JLS "Negative Bubble" Integrated Identifications

Dates for the bubble-induced crashes and rebounds identified with the ex-post JLS model and JLS “negative bubble” model that are withinMalaya the duration of the research’s sampling period are documented in various past studies (i.e. for JLS model, see Feigenbaum, 2001; Johansen 2004; Johansenof & Sornette, 2010; Sornette & Cauwels, 2014; and Liberatore 2011a, 2011b; for the “negative bubble” JLS model, see Yan,

Woodard & Sornette 2012). The novelty introduced by the research is the simultaneous examination of the scale-invariant JLS models with other econometric models

(parametric, semi-parametric and non-parametric) for the identification stock market declines on their predictability with stock market fundamentals.

The in-sample predictability test of the JLS integrated identifications of stock marketUniversity crashes and rebounds finds that the most significant predictor for the k = 1 and k

= 3 forecasting horizons is the S5Y3M. As for the longer horizons of k = 6 and k = 9 horizons, the most powerful predictor is found to be CDLI (∆). CFNAI best predicts the longest horizon of k = 12. Similar to the CDLI, the CFNAI is a weighted average index that summarises a broad range of economic indicators for the measurement of economic activity. The in-sample results affirms the findings by Du, Denning & Zhao (2012) and 268

Koijen, Lustig & van Nieuwerburgh (2010) that showed evidence of significant relationship between the CFNAI and stock returns. The out-of-sample test on the other hand shows the best predictors for the k = 1 horizon is CAPE (∆) while the CDLI (∆) is found to have the highest predictive power for the rest of the horizons i.e. k = 3, k = 6, k

= 9 and k = 12.

5.2.9 Predictive Consistency of Test Variables

In summary, the test variable that is consistently found to be the best predictor for the in-sample ex-ante prediction of stock market declines across all ex-post models for the k

= 1, k = 3 and k = 6 forecasting horizons is CDLI (∆). Correspondingly, ISMI is the most consistent best predictor for the k = 9 horizonMalaya while S10Y3M is consistently the best for the k = 12 horizon.

Across the out-of-sample predictabilityof tests, the CDLI (∆) is again shown to be the most consistent test variable with the highest predictive power for the shorter horizons of the k = 1 and k = 3. It is also the joint best indicator for stock market declines with another two test variables, namely the DVD (g) and ISMI for the horizon of k = 6. Both the DVD (g) and ISMI moreover are also jointly the most powerful predictors on consistent basis for the forecasting horizon of k = 9 along with another test variable, i.e. S5Y3M. The predictive reliability of DVD (g) and S5Y3M for the longer horizonsUniversity vindicated once more as both the test variables along with the M2 emerged as the best predictors for two models (out of eight) respectively for the k = 12 forecasting horizon.

269

5.3 Accomplishment of Research Objectives

Results in table 5.1 and 5.2 are self-explanatory. Discussions from subsections 5.2.1 to

5.2.7 that ensue answered emphatically the research questions in Section 1.3 and the corresponding research objectives in Section 1.4. Recapping the first research objective which is to determine the most consistent market fundamentals to predict the regime of bear markets that specified based on various models of parametric, semi-parametric and non-parametric for 1, 3, 6, 9 and 12 months in the future, discussions in the aforementioned subsections have underscored the most consistent market fundamentals with highest predictive power across these pre-determined forecasting horizons (read these sections for details).

Likewise for the second research objective which is to determine the most consistent market fundamentals to predict stock marketMalaya crashes (contrasting with the stock market rebounds) that are specifiedof based on the econophysics’ scale-invariant approaches for 1, 3, 6, 9 and 12 months in the future, discussion on the results that fulfilled the articulated objective is presented in subsection 5.2.8.

At the risk of repetition, the most consistent best predictors for the in-sample tests across all models for the shorter forecasting horizons of k = 1, k = 3 and k = 6 is the

CDLI (∆). ISMI is the most consistent best predictor for the k = 9 horizon while

S10Y3M is consistently the best for the k = 12 horizon. Results of the out-of-sample testsUniversity show that CDLI (∆) is the most consistent test variable with the highest predictive power for k = 1 and k = 3; DVD (g) and ISMI for k = 6; DVD (g), ISMI and S5Y3M for k = 9; and DVD (g), S5Y3M and M2 for k = 12. The findings fulfilled the third objective of the research.

270

5.4 Implications of Research

It was highlighted in the earlier section on the tendency of the topics related to the research being overlooked in the mainstream literature of financial economics. Against the undercurrent of orthodox theories, previous studies have argued that stock market crashes should not be treated as mere outliers (Bakshi & Madan, 1998; Johansen &

Sornette, 1998b; 2002; Mandelbrot; 2004), the movement of market is not random (Lo

& Mackinlay 2001) and bear markets are not merely “ex-post categorisation of random data” (Gonzalez, Powell, Shi & Wilson, 2005), and the market fundamentals is useful as directional indicators for stock market (Chen, 2009; McCown 2007; Shiller, 2005 etc.).

The overall result of the research is in agreement with all the propositions as above. Likewise, implications of study also concur with most that were suggested in previous studies that are on the same wavelengthMalaya with the research. The more significant implications are discussed in theof following.

5.4.1 Market Predictability for Traders’ Consideration

One persistent conundrum in the financial market that is yet to be explained adequately is the continuous occurrences of bubbles and crashes in an efficient market. These gaps in the literature have motivated the resurgence of empirical studies on market predictabilityUniversity that prioritise statistical evidence over the underlying explanations to the phenomena. Econophysicists are particularly captivated by the occurrence of major declines in asset markets. In recent years, the cross-disciplinary approaches which introduced various concepts such as scale invariance, fractal, network etc. for empirical studies on stock market are gaining wide acceptance.

271

Methodologies applied in the modelling and predicting of the cyclical bull and bear markets on the other hand are more conventional. The underlying premises, assumptions and methodologies in this area are fundamentally extensions to those used for studies on business cycle. The use of stock market composite indices (or their derivative i.e. aggregated stock market returns) for the ex-post modelling of heterogeneous regimes for both the scale invariant JLS models and models for the bull and bear markets provides the research with a parallel mean to examine these two schools of thought on a common empirical platform.

The wealth of literature in examining the nexus between the business cycle and market fundamentals such as macroeconomic and financial variables has motivated comparable studies with the employment of market fundamentals for the investigation of stock market declines. Advancing in similar direction,Malaya the research has extended the literature with a more extensive set of variables for the ex-ante prediction of a list of ex- post models for stock market declines. Marketof fundamentals is thus classified as one of the empirical tranches of study that share a common platform with aforementioned areas that are included in the research.

The empirical results produced by the research conform with the related studies earlier by Chauvet & Potter (2000), Chang (2009), Chen (2009) and Perez-Quiros &

Timmermann (2000) that the heterogeneous regimes of stock market are predictable to variousUniversity extents with market fundamentals. The out-of-sample test on the scale-invariant JLS models yields some interesting results too where the integrated identification of the ex-post models is shown to be consistently predictable across horizons particularly with the CDLI (∆) (except for the k = 1 where the best predictor is CAPE). The evidence of market predictability has significant implications for both the portfolio management of

272 fund managers (Chauvet & Potter, 2000) and the study of systemic risk (Johansen &

Sornette, 2002).

The results of market predictability for the longer forecasting horizons are consistent with the study by McCown (2007) which showed evidence that there is a significant time lag in the adjustment of stock prices after the business cycle has reached its peak. The market fundamental of yield curve inversion between the five-year

Treasury note and the one-year Treasury bill was used as the proxy for business cycle peak. The study nonetheless conceded that there is no plausible explanation to the slow reaction after the first sign of an impending decline in the market.

Peters (1994) suggested the investment horizons of traders played a role in the discrepancies of buying and selling decisions although traders might be looking at the same indicators. The selling decision of one horizonMalaya is thus taken up by traders with possibly longer investment horizons. The suggestionof further vindicates the approach of the research of ascertaining the most consistent and best predictors for varying horizons.

Ho & Quah (2012) offered a different perspective on the slow sell-off and proposed that the institutional investors which command the greatest influence to the movement of the market might engage in the clock game among them. No one is sure if the yield curve inversion is a false signal or otherwise to a market decline. Therefore everyone plays a waiting game which thus delays the reaction of the market at the first sight of danger.

UniversityAll in all, apart from the contribution on the academic aspects, the research hopes to establish a more solid foundation for professionals in the financial market to understand the underlying movements of the economic fundamental preceding to a stock market decline and provides a clearer guideline in defining different types of stock market declines.

273

5.4.2 Theory and Knowledge Development

The research finds that the development of studies in stock market declines over the course of history follows the trajectory of literature evolution of the economics discipline in general. Over the decades, literature in this area which began from the straight-forward post-mortem analyses on specific episodes of bear markets and crashes has split into two mainstreams, i.e. the approach of rationalising the behaviour of agents in the market (microeconomic perspective) and the approach of evidence-based empirical analyses (macroeconomic perspectives). Both approaches, developed contemporaneously are now profoundly axiomatically-based and mathematically advanced.

From the microeconomic perspective, the conventional economics holds that market agents are rational and most orthodox financialMalaya economics theories are built from this conjecture. Behavioural financeof on the contrary argued that agents are not always rational when dealing with financial matters in the highly intense market.

Volumes of contradicting empirical evidence on the behaviour of the financial market over time on another note suggested that the only constant subject in the market is change. Thus it is difficult to assimilate these arguments with the generalisation of market efficiency. One illustration to underscore the argument is the oft-cited conundrum of who buys a share when someone sells it in the market. If everyone has theUniversity same rationale and shares the same information, transactions could almost never occur in a market deemed to be “efficient”.

From the macroeconomic perspective, the modelling approach adopted in financial economics is commonly criticised for being overtly reliant on the Gaussian distribution (see Chapter 2 for elaboration and citations). Evidently, most devastating occurrences in the financial market such as financial crises and stock market crashes are

274 outliers. Methodologies originated from the physics discipline are more robust in dealing with issues related to non-normal distribution. Within the context of the research, steep declines in stock market i.e. crashes which are deemed outliers are addressed with the JLS model.

The JLS model found its roots in the econophysics that espouse the theory of complex systems. This motivates the research to probe deeper into the area by tracing the historical development of the econophysics discipline and the complex system theory in the stock market. The chronicle begins from the Mandelbrot era in the early

1960s, up to the latest advancements the JLS model (and its variants) in identifying stock market crashes and rebounds. Through the review of literature, the research also clarifies the often confused distinctions between the complex systems theory and the chaos theory. Malaya The research suggests that one of theof reasons the conventional economics is less receptive to theories from the econophysics is the lack of connections between these theories with the scientific paradigm of methodological individualism. The theoretical inferences of the econophysics are typically more inductive (e.g. complex system theory) in contrast to the more deductive leaning conventional economic theories (e.g. rational expectation). For example, the complex systems’ theoretical foundation deliberates on concepts such as self-similarity, complex interaction of units and emergingUniversity pattern etc., all of which are deemed to be more suited for explaining elements within the physics context than explaining the behaviour of agents in the market.

Moreover, the econophysics is also inclined (or to a much lesser extent) not to reconcile findings from empirical studies with theoretical expectations, which economics are prone to in general.

275

In recent development nonetheless, there is evidence that studies from the econophysics discipline are making considerable attempt to assimilate their findings with the conventional economic theories and assumptions (e.g. compare the earlier econophysics literatures of Mantegna & Stanley, 1997; Mantegna, 1999; Vandewalle,

Ausloos, Boveroux, & Minguet, 1999; Potters & Bouchaud, 2003; Siokis, 2013;

Sornette & Cauwels, 2012). Reconciliation in theoretical discourse specifically in the studies of JLS model (and its variants) with the financial economics is somewhat mixed.

Most pioneering literatures on the JLS model explained the behaviour of the market exclusively from the perspective of the complex systems theory (e.g. see Johansen &

Sornette 1998b; Johansen & Sornette 1999a; Johansen & Sornette 1999c).

Discussion from the conventional economic views such as noise trader, rational expectation and rational bubbles are sparse in thisMalaya area at the inception (e.g. see Johansen, Ledoit & Sornette 1998) but are gradually gaining momentum of late (e.g. see

Fantazzini & Geraskin, 2013; Sornette, Woodard,of Yan & Zhou, 2013; Lin, Ren &

Sornette, 2014; Pele & Mazurencu-Marinescu 2012; Yan, Woodard, & Sornette, 2010;

2012). Nonetheless, JLS studies that combined theoretical discussion from the speculative bubbles perspectives (non-conventional) are quite prevalent from the onset

(e.g. Johansen & Sornette 2000; 2002; Johansen, Sornette & Ledoit, 1999; 2000;

Sornette & Johansen 1998; 2001). The advancement in the reconciliatory direction between the econophysics and the conventional financial economic theories (or otherwise)University is promising as the synthesis of the strength from both disciplines would be beneficial to the continuous evolution of the field.

The section of literature classification in Chapter 2 underscores the diverseness in the study of stock market declines. The earlier literatures in this area are mostly focused on the investigation on specific period of selected crashes in the financial

276 market, especially major crashes in the history. Focus was subsequently shifted to the study of the behaviour of agents in the market and factors that lead to the generation of bubbles. Theoretical models rooted to the general equilibrium and rational expectations proliferated at that juncture. The evolution continued with the establishment of the

EMH as the most dominant theory in financial economics, followed by the behavioural finance.

In response to the growing disputes on the conventional theories, the AMH was introduced to address the various shortcomings raised over the years, particularly by the behavioural finance school of thought. The AMH is a more comprehensive and flexible hypothesis that attempts to reconcile the differences between the EMH and the behavioural finance. The hypothesis uses concepts that originated from the biological perspective to provide plausible explanations to theMalaya behaviour of market’s agents and the complexity of movements in the financialof market (Lo, 2004a; 2005). Other schools of thought that closely relate to the subject of stock market declines continue to develop in overlapping time. One of the more prominent developments is the study on the causes of bubbles (preludes to crashes) via the methodological individualism paradigm. Studies in this direction generally centre on issues pertaining to assumptions on agents’ behaviour and the axiomatisation of such propositions. Naturally, theories in this area are interlinked with each other in one aspectUniversity or another but based on different assumptions. One of the most common points of debate is on whether the agents in the market are rational at various stages during the build-up of bubbles.

The early advancement of studies that focused solely on empirical investigation of market predictability (particularly returns of assets) were somewhat curtailed by the universal acceptance of conventional theories that the movement of asset prices is

277 random and unpredictable. Subsequent evidences of market predictability are categorised as market anomalies. Conventional theories in financial economics argued that market anomalies are expected to normalise over the course of time when the market adapts and capitalise on the abnormal returns brought about by such predictability.

Malaya of

Figure 5.1: General Categorisation of Theories on the Espousal of Rationality Assumption University Figure 5.1 shows the general categorisation of theories or area of studies (that are relevant to the research) on the espousal of the rationality assumption. In the context of bubbles and declines, the rational bubbles and the EMH schools of thought advocate that agents in the market are rational. In contrary, the behavioural finance, speculative bubbles and AMH argue that agents in the market are not rational. Studies that examine

278 stock market declines from the cyclical and the market fundamentals perspectives tend not to deliberate on the issue of agents’ behaviour and focus solely on empirical evidence. The discipline of econophysics which specifically examines stock market crashes from the angle of complex systems holds mixed views on the rationality conjecture. Econophysics in general prioritises methodological advancements and empirical results over theories of agents’ behaviour in the market.

Malaya of

Figure 5.2: General Categorisation of Theories on the Espousal of the Bubbles UniversityProposition Figure 5.2 shows the general categorisation of theories or area of studies (that are relevant to the research) on the espousal of the bubbles proposition. Due to the espousal of the rationality assumption, the EMH is considered a proponent to the concept of market bubbles. The rational bubbles school of thought which similarly advocates the rationality of market agents nonetheless supports the bubbles proposition. Behaviour of

279 agents in contributing to the inflation of bubbles in the market is justified with wide ranging hypotheses and inferences. In general, an agent is assumed to be acting rationally so long as his trading decision is profitable even if the market is at the brink of collapse (see Chapter 2 for further reading and citations). Other schools of thought namely the behavioural finance, speculative bubbles, AMH and complex systems acknowledge the existence of bubbles in the market. The cyclical and the market fundamentals schools of thought tend not to deliberate on the issue of bubbles but focus on empirical results.

5.4.3 Policy Consideration Joseph Stiglitz, the year 2001 Nobel laureate in Malaya economics argued that the deeply entrenched economic tenet of “markets function best when left alone” is a fallacy that has caused enormous harm to the economy.of As such, the monetary authorities in the financial market should exert greater control in order to prevent the repetition of financial catastrophes (Stiglitz, 1993; 2010). In similar tone, other Nobel laureates, namely Robert Shiller and Paul Krugman had also called for greater financial regulations to keep up with the increasingly chaotic financial market. Shiller (2005), in citing cases such as the Great Depression and the Japanese stock market slump in the late 1980s, proposed that the monetary authorities had failed in various extents in containingUniversity devastating market slumps in the past. Krugman (2009) likewise furnished an in-depth critical review on the Sub-prime Crisis of 2009 and pointed out that it was due to the failure of the monetary authorities in exercising control over the market.

As highlighted throughout the research, studies on impactful market upheavals over the years are often examined on the hindsight through the lens of causal inquiry such as those underscored above. Analyses via this approach are commonly delivered in 280 the form of narration, supplemented with supporting statistics. Post-mortem studies on selected episodes of stock market crashes or impactful bear markets are prevalent because there is always a pressing need for policymakers to identify and address deficiencies in the system after a devastating slump in the market. Such studies thus have been pivotal in assisting policymakers to learn from the past for the formulation of reactionary policies for the future.

The research believes that the advancement in the study of ex-post modelling for stock market declines and their nexus with various market fundamentals could be equally important for policymakers, particularly the central banks or monetary authorities in devising pre-emptive monetary measures in the face of an impending decline in the market. The predictive function of market fundamentals based on empirical evidence along with the modelling toolsMalaya developed in academics could complement the voluminous post-mortem studies on crises in the literature and better prepare the monetary authorities in the future.of

5.5 Limitations of Research

The market fundamentals tested in the research are only limited to those that are listed and described in Chapter 3. The list is considerably extensive. It encompasses proxies from multi-segments of the market and the economy which are categorised by the researchUniversity as Stock Market Fundamentals (aggregated financial ratios mostly used for fundamental analysis of stocks), Financial Market Fundamentals (macroeconomic indicators commonly used to predict recessions and financial crises), Industrial

Indicators (industry level indicators commonly used in the forecasting of business cycle), Market Sentiments (results of periodically updated surveys that gauge the

281 sentiment of producers and consumers), Indexes of Leading Indicators (include two of the most important U.S. economic indices).

The research acknowledges that there could be more market fundamentals that significantly predict the changing of regimes in the stock market but are not examined here. For example, Chang (2009) in a related study tested the regime switching of volatility in stock market with default premium and discovered mixed results. Chen

(2009) in similar study also tested among others the industrial production, unemployment rates, federal fund rates, exchange rates and public debt, all of which yielded varying degrees of significance, using the Markov-switching model. Likewise,

Chauvet & Potter (2000) employed the CRSP value-weighted index as one of the test variables and Perez-Quiros & Timmermann (2000) tested the firm size as the sole test variable in their respective studies and both variablesMalaya are found to have significant predictive power on the cyclical variation ofof the stock market. The observation of the research begins from the 1960s which is similar to Chang

(2009) and Chen (2009). Access to the more important market fundamentals mostly are only available circa 1965 onwards. The observations of the studies by Chauvet & Potter

(2000) and Perez-Quiros & Timmermann (2000) began earlier, from year 1954.

Therefore the test variables used in both studies could be limited due to the unavailability of data. Predictability testing with nested models requires the segmentationUniversity of data into two portions, i.e. one portion for parameter estimation (R) and another portion for testing (P). Therefore, the selection of different observation periods or an extreme deviation from the optimal combination between the (R) and (P) portions for the nested models (see Clark & West 2007) could yield a different or potentially spurious result. The research observation period is only limited to the beginning of April

1967 until the end of June 2014. It is thus possible to derive different findings with the

282 same set of models and test variables by calibrating the observation period and parameter estimation period.

The research is focused exclusively on the U.S. stock market specifically the

S&P 500 Index. As aforementioned, the S&P 500 Index is chosen because it adequately represents a wide range of important stocks that constitute over 75% of the total market value. Studies with other stock market indices other than the S&P 500 Index within the

U.S. such as the DJIA, NASDAQ or AMEX is outside the boundary of the research.

5.6 Suggestions for Future Study

The research suggests that market fundamentals employed in past studies (not examined in the research) that are found to have significant relationshipMalaya with stock market returns or stock market volatility may also be significant as predictors to the cyclical variation of states in the stock market. Some of theseof more recent studies are Ali, Hwang &

Trombley (2003) with the book-to-market ratio; Binsbergen & Koijen (2010) with the present-value approach; Boudoukh, Michaely, Richardson & Roberts (2007) with the payout yield, Campbell (2008) with the dividend-price ratio; Campbell & Yogo (2006) with dividend–price and smoothed earnings–price ratios; Campbell & Thompson (2008) with a set of growth-adjusted financial ratios; Campbell & Vuolteenaho (2004) with the inflation-adjusted expected dividend growth rate; and Nyberg (2011) with one-month excessUniversity return. The study by Chung, Hung & Yeh (2012) is more extensive. Among the test variables that are found to have significant predictive power on stock returns include the size of firms, the age of firms, the volatility of return, equity ratio, tangible assets and the growth potential of firms.

283

Furthermore on evidence found in past studies, market fundamentals which are reflective of the macroeconomic aspect that have significant relationship with the stock market returns or stock market volatility (but not investigated in this research) include industrial production, orders inflow and output gap shown in Dopke, Hartmann &

Pierdzioch (2008); trade balance and employment shown in Flannery & Protopapadakis

(2002); default spread shown in Hartmann, Kempa & Pierdzioch (2008); and relative rates for money market, 3-month Treasury bill and government bond yield as shown by

Rapach & Wohar (2005). These are some of the market fundamentals that can be considered as test variables for future studies with the models used in the research.

Another potential avenue for further exploration in this area is in the calibration of the nested models. Schwert (1989) showed evidence that market fundamentals are sensitive to the sampling duration. A longer period ofMalaya R (the in-sample observations) in general yields more accurate results at the risk of overfitting, i.e. misinterpreting noise as signal in the data (see Clark, 2004; Tashman,of 2000). Chang (2009) in a related study estimated the predictive model’s parameter with an arbitrary ratio of about P/R ≈ 1/4

(i.e. 34 years and 3 months observations were used for model’s parameter estimation to predict a very short prediction period of 8 years and 4 months). The research applied the stringent Clark & West (2007) method with a ratio of P/R ≈ 2 (note that the sample observations for the portion of parameter estimation and the prediction portion vary for different horizons, refer to Section 4.5). Chen (2009) in a related study also adhered to theUniversity Clark & West (2007) suggestion which is probably one of the most optimal, parsimonious and efficient methods for the in-sample and out-of-sample testing at the time of writing. Any adjustment to the ratio of P/R needs to be treated with caution.

Lastly, future studies may consider using methodologies similar to this research to explore the stock markets that are outside of the U.S. particularly those in the

284 emerging markets. Hartmann, Kempa & Pierdzioch (2008) noted that economic downturn and financial turmoil in these markets are inclined to be more devastating and regular compared to developed market. Nonetheless, one possible hindrance to the suggestion could be the availability of a rich dataset that allow for an extensive study.

5.6 Conclusion

The research investigates the stock market declines in stages. First the heterogeneous regimes of rallies and declines are determined with the employment of a proposed list of ex-post models. The outputs derived are next tested for predictability with market fundamentals. Dissimilar to the common multivariate modelling where the selection of variables is predetermined based on the review of pastMalaya studies, the findings of the most significant predictive variables in the research is a posteriori where a large set of market fundamentals are tested individually and repeatedlyof with univariate nested models (in- sample and out-of-sample) for various forecasting horizons.

The initiative in combining various empirical approaches (including the JLS models) under a common research framework is one of the contributions of the research, which may motivate more cross-referencing among the different schools of thought for studies in this area in the future. A rigorous review of literature is undertaken to cover the vast ground of the topic on stock market declines and to explain theUniversity need of the research in Chapter 2. Besides illustrating how the research filters the theoretical studies that centres on the market agents’ behaviour in the context of stock market declines and the framing of the research’s common empirical platform for the ex-post modelling and ex-ante predictability analysis, the section on theoretical discussion also extends the literature modestly by presenting the convergent and divergent theoretical findings on the general espousal of the assumption on rationality 285 and the preposition of bubbles in the market among the relevant schools of thought.

These theoretical findings are not part of the research objectives but crucial nonetheless as they could help build the foundation for a more focused theoretical deliberation for future in this area. The meticulous reviewing, synthesising and filtering of literature from the theoretical aspect also lend more credence to the research’s empirical framework and make the research more complete.

One of the novelties of the research is the modest innovation of transforming the

Markov-switching’s smoothed probabilities into dichotomised smoothed probabilities.

The conversion of the extracted probabilities from the model into binary values enables a direct and parallel comparison of the in-sample predictability test result of the

Markov-switching model with other semi-parametric and non-parametric models using the pseudo- 푅2 (Mcfadden) and pseudo- 푅2 (DieboldMalaya & Rudebusch). Likewise, the innovation also allows results derived from the Markov-switching model to be measured with QPS for the out-of-sample ofpredictability test and compared across with other models. In that, it bridges a research gap found in Chen (2009) where results measured in 푅2 were compared with other results measured in pseudo- 푅2 and results measured in MSPE-adj were compared with results in QPS. The in-sample results derived from the innovation are almost similar to the results of the Markov-switching’s smoothed probabilities while the out-of-sample results are also very close to the results yielded from the naïve moving average model. These encouraging results are signs of robustnessUniversity of the innovation. On another note, the research also resolve a probable shortcoming found in the same study (i.e. Chen, 2009) in the proportioning of the P and

R periods for the nested models. As a recap, the sample observation selected by the former study for parameter estimation were too brief and might have not included any significant period of stock market declines in the early decades.

286

Besides, the research also introduced the naïve moving average negative return model to reduce the regime switching volatility of the original naïve moving average model. The predictability results derived are interestingly similar to the results of the

Lunde & Timmermann’s B-B algorithm for the in-sample test. The results of the out-of- sample test for the naïve moving average negative return are considerably consistent with other models, particularly the Candelon, Piplak & Straetman’s B-B Algorithm.

One of the advantages the naïve moving average negative return model possesses over the B-B algorithms is its simpler model specification and the requirement of lighter programming. The simplification may ease the analysis for domestic traders who have limited access to advanced econometric tools.

The predictability test administered on the JLS model and JLS “negative bubbles” model with market fundamentals is an exploratoryMalaya initiative. Econophysics approaches are widely acknowledged to be highly efficient, sensitive and precise (Ho,

Estrada Ruiz & Yap, 2016). Thus, the integratedof identification of crashes and rebounds with the combination of the JLS model and the JLS “negative bubbles” model is designed to identify the short and sharp heterogeneous regimes for each of the pre and post turning points with a limited window of 6-month period. Results of the scale- invariant approach are compared in parallel with the parametric, semi-parametric and non-parametric approaches used to model the more prolonged type of stock market declinesUniversity i.e. the bear markets. Unlike the study by Qi (2001) which employed the complete set of the Estrella

& Mishkin’s Financial Variables to the test the predictability of economic recession, studies related to the research such as Chang (2009) and Chen (2009) only used selected variables found in the list for similar predictability studies on stock market. Most of the

Shiller’s Financial Variables are also not tested in these previous studies. The findings

287 of the CDLI (∆) and CFNAI (categorised in Indexes of Leading Indicators); CAPE (∆)

(categorised in Stock Market Fundamentals); TN10Y (∆) (categorised in Financial

Market Fundamentals) and ISMI (categorised in Industrial Indicators) as the most significant predictors across various forecasting horizons (in the in-sample and out-of- sample tests) are discoveries that extended the literature in this research area. Other best predictors found in the research are the well-established ones such as the term spreads, dividend and money supply, which conform to a number of related studies e.g. Chang

(2009) Chen (2009) and Hartmann, Kempa & Pierdzioch (2008).

In a summation, the empirical investigation completed here is also complemented with a meticulous review on the historical, etymological and epistemological development of worldviews on the movement of stocks and stock market declines. The contradictories as well as Malaya common grounds among different schools of thought are outlined and discussed thoroughly. The research reckons that the continuous confutation of theories and counterof arguments among schools of thought has so far been futile in fully nullifying any of the fundamental propositions from each schools of thought. On the contrary, reconciling the theoretical differences and bridging the methodological gap may be a positive step forward as recurring themes such as crashes, rebounds and the cyclical switching of bear and bull markets are prominent across different schools of thought that examine the area of stock market declines. University

288

REFERENCES

Abreu, D., & Brunnermeier, M. (2003). Bubbles and crashes, Econometrica, 71(1): 173-

204.

Adam, M. C., & Szafarz, A. (1992). Speculative bubbles and financial markets. Oxford

Economic Papers, 44: 626-640.

Al-Anaswah, N., & Wilfling, B. (2011). Identification of speculative bubbles using state-

space models with Markov-switching, Journal of Banking & Finance, 35: 1073- 1086. Malaya Alexander, G. J., Peterson, M. A., & Wang B. X. (2014). The puzzling behavior of short sellers around earnings announcements,of Journal of Financial Intermediation, 23(2): 255-278.

Ali, A., Hwang, L-S., & Trombley, M. A. (2003). Arbitrage risk and the book-to market

anomaly, Journal of Financial Economics, 69(2): 355-373.

Amadeo, K. (n.d.)(a). 10-Year Treasury Note and Rate. Retrieved on 19 February, 2013,

from About.com - US Economy: Universityhttp://useconomy.about.com/od/fiscalpolicydefinitions/p/10-Year-Treasury.htm Amadeo, K. (n.d.)(b). What Is the GDP Growth Rate?: Why It's Important, How to

Calculate It. Retrieved on 7 March, 2013, from About.com - US Economy:

http://useconomy.about.com/od/grossdomesticproduct/f/GDP_Growth_Rate.htm

289

Andersen, J. V., Gluzman, S., & Sornette, D. (2000). Fundamental framework for

technical analysis of market prices, The European Physical Journal B, 14: 579-601.

Anderson, K., Brooks, C., & Katsaris, A. (2010). Speculative bubbles in the S&P 500:

Was the tech bubble confined to the tech sector? Journal of Empirical Finance, 17:

345-361.

Ang, A., & Bekaert, G. (2007). Stock return predictability: Is it there? Review of Financial

Studies, 20: 651-707.

Antoniou, A., & Garrett, I. (1993). To what extent did stock index futures contribute to the

October 1987 stock market crash? The Economic Journal, 103(421): 1444-1461.

Malaya th Apostolou, B., & Apostolou, A. G. (2004). Keys To Investing in Common Stocks (4 ed.). New York: Barron’s Educational Series,of Inc. Arneodo, A., Muzy, J. F., & Sornette. D. (1998). Direct causal cascade in the stock market,

The European Physical Journal B, 2: 277-282.

Arthur, W. B. (1995). Complexity in economic and financial markets, Complexity, 1: 20-

25.

Asako, K., & Liu, Z. (2013). A statistical model of speculative bubbles, with applications Universityto the stock markets of the United States, Japan, and China, Journal of Banking & Finance, 37: 2639-2651.

290

Auerbach, A. J. (1981). The index of leading indicators: Measurement without theory,

twenty five years later, NBER Working Paper, 761. Retrieved on 27 October, 2012,

from The National Bureau of Economic Research (NBER):

http://www.nber.org/papers/w0761.pdf?new_window=1

Avery, P. (2003). Complex Systems Theory as a Paradigm for Theoretical Formalisms

and Empirical Observations of Hymenoptera. Retrieved on 11 September, 2010,

from CSU Entomology Courses at the Colorado State University Web site:

http://www.colostate.edu/Depts/Entomology/courses/en507/papers_2003/avery.pdf

Avramov, D. (2002). Stock return predictability and model uncertainty, Journal of Financial Economics, 64: 423-458. Malaya Avramov, D., & Chordia, T. (2006). Predictingof stock returns, Journal of Financial Economics, 82: 387-415.

Bachelier, L. (2006). Louis Bachelier's Theory of Speculation: The Origins of Modern

Finance. New Jersey: Princeton University Press.

Bai, J., & Ng, S. (2005). Tests for skewness, kurtosis, and normality, Journal of Business

& Economic Statistics, 23(1):49-60. Bakshi,University G., & Madan, D. (1998). What is the probability of a stock market crash?, Paper presented at the 1998 NYU Conference on Finance and Accounting. Retrieved on

11 February, 2011, from:

citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.8.8094.pdf

291

Ball, R. (1978). Anomalies in relationships between securities' yields and yield-surrogates,

Journal of Financial Economics, 6: 103-126.

Balvers, R. J., Cosimano, T. F., & McDonald, B. (1990). Predicting stock returns in an

efficient market, The Journal of Finance, 45(4): 1109-1128.

Banz, R. (1981), The relationship between return and market value of common stock,

Journal of Financial Economics, 9: 3-18.

Barbarino, A., & Jovanovis, B. (2007). Shakeouts and market crashes, International

Economic Review, 48(2): 385-420. Barberis, N. C., & Thaler, R. H. (2007). A SurveyMalaya of Behavioral Finance, in Constantinides, G. M., Harris, M., & Stulzeds, R. M. (Eds.). Handbook of the Economics of Finance: Volume 1B,of Financial Markets and Asset Pricing (Reprinted). Amsterdam: Elsevier.

Barberis, N. C., Shleifer, A., & Vishny, R. (1998). A model of investor sentiment Journal

of Financial Economics, 49: 307-343.

Barlevy, G., & Veronesi, P. (2003). Rational panics and stock market crashes, Journal of

Economic Theory, 110: 234-263. Barnhart,University S. W., & Giannetti, A. (2009). Negative earnings, positive earnings and stock return predictability: An empirical examination of market timing, Journal of

Empirical Finance, 16(1): 70-86.

292

Barone-Adesi, G., Mancini, H., & Shefrin, L. (2012). Behavioral Finance and the Pricing

Kernel Puzzle: Estimating Risk Aversion, Optimism, and Overconfidence.

Retrieved on 13 June, 2013, from:

http://www.stern.nyu.edu/cons/groups/content/documents/webasset/con_035927.p

Barro, R. J., & Ursua. J. F. (2009). Stock market crashes and depressions, NBER Working

Paper, 14760. Retrieved on 22 June, 2011, from the National Bureau of Economic

Research (NBER): http://www.nber.org/papers/w14760.pdf?new_window=1

Bartholdy, J., & Feng, T. (2013). The quality of securities firms' earnings forecasts and stock recommendations: Do informational advantages,Malaya reputation and experience matter in China? Pacific-Basin Finance Journal, 24: 66-88.

Bartolozzi, M., Drozdz, S., Leinweber, D., ofSpeth, J., & Thomas, A. (2005). Self-similar

log-periodic structures in Western stock markets from 2000, International Journal

of Modern Physics C, 16: 1347-1361.

Barunik, J., & Vosvrda, M. (2009). Can a stochastic cusp catastrophe model explain stock

market crashes, Journal of Economic Dynamics & Control, 33: 1824-1836.

Basistha. A & Kurov, A. (2008). Macroeconomic cycles and the stock market’s reaction to Universitymonetary policy, Journal of Banking & Finance, 32: 2606-2616.

Basu, S. (1977). Investment performance of common stocks in relation to their price-

earnings ratios: A test of the efficient market hypothesis, The Journal of Finance

32(3): 663-682.

293

Basu, S. (1983). The relationship between earnings' yield, market value and return for

NYSE common stocks: Further evidence, Journal of Financial Economics, 12:

129-156.

Bates, D. S. (1991). The crash of '87: Was it expected? The evidence from options markets,

The Journal of Finance, (Papers and Proceedings, Fiftieth Annual Meeting,

American Finance Association, Washington, D. C., December 28-30, 1990) 46(3):

1009-1044.

Bates, D. S. (2012). U.S. stock market crash risk, 1926-2010, Journal of Financial

Econometrics, 105: 229-259.

Bates, D. S., & Craine, R. (1999). Valuing the futuresMalaya market clearinghouse's default exposure during the 1987 crash, Journal of Money, Credit and Banking, 31(2):

248-272. of

Baumohl, B. (2008). The Secrets of Economic Indicators (2nd ed.). New Jersey: Wharton

School Publishing.

Baur, M. N., Quintero, S., & Stevens, E. (1996). The 1986-88 stock market: Investor

sentiment or fundamental? Managerial and Decision Economics, 17(3): 319-329.

Beinhocker, E. D. (2007). The Origin of Wealth. New York: Random House Business

UniversityBook.

Benning, C. J. (1997). Prediction skills of real-world market timers, Journal of Portfolio

Management, 23(2): 55-63.

294

Bergeron, C. (2013). Dividend sensitivity to economic factors, stock valuation, and long-

run risk, Finance Research Letters, 10(4): 184-195.

Bernanke, B. S. (1990). Clearing and settlement during the crash, The Review of Financial

Studies, (National Bureau of Economic Research Conference: Stock Market

Volatility and the Crash, Dorado Beach, March 16-18, 1989) 3(1): 133-151.

Bernanke, B. S., & Kuttner, K. N. (2005). What explains the stock market's reaction to

Federal Reserve policy?, Journal of Finance 60(3): 1221-1257.

Bhaduri, S., & Saraogi, R. (2010). The predictive power of the yield spread in timing the stock market, Emerging Markets Review, 11: 261-272.Malaya Binsbergen, J., & Koijen, R. (2010).Predictive regressions: A present-value approach, Journal of Finance, 65: 1439-1471. of Birz, G & Lott, J. R. Jr. (2011). The effect of macroeconomic news on stock returns: New

evidence from newspaper coverage, Journal of Banking & Finance, 35: 2791-2800.

Black, F. (1988). An equilibrium model of the crash, NBER Macroeconomics Annual, 3:

269-275.

Black, J., Hashimzade, N., & Myles, G. (2009). Oxford Dictionary of Economics (3rd ed.). UniversityNew York: Oxford University Press. Blanchard, O. J., & Watson, M. W. (1982). Bubbles, rational expectations and financial

markets, in Wachtel, P. (Ed.), Crises in the Economic and Financial Structure (pp.

295-316). Massachusetts: Lexington Books.

295

Blume, M. E., Mackinlay, C., & Terker, B. (1989). Order imbalances and stock price

movements on October 19 and 20, 1987, The Journal of Finance, 44(4): 827-848.

Bohl, M. T. (2003). Periodically collapsing bubbles in the US stock market? International

Review of Economics and Finance, 12: 85-397.

Bolton, H. (1960). The economy of the stock market... 1929 and now, Financial Analysts

Journal, 16(1):57-62.

Bonanno, G., Caldarelli, G., Lillo, F., Micciche, S., Vandewalle, N., & Mantegna, R. N.

(2004). Networks of equities in financial markets, The European Physical Journal B, 38: 363-371. Malaya Bordo, M. (2003). Stock market crashes, productivity boom busts and recessions: some historical evidence, IMF World Economicof Outlook (Background paper). Retrieved on 15 January, 2013, from:

http://www.cfr.org/content/thinktank/Depression/Bordo_2.pdf

Bothmer, H. C. G., & Meister, C. (2003). Predicting critical crashes? A new restriction for

the free variables, Physica A, 320: 539-547.

Boudoukh, J., Michaely, R., Richardson, M., & Roberts, M. (2007). On the importance of

measuring payout yield: Implications for empirical asset pricing, Journal of

UniversityFinance, 62: 877-915.

Boudoukh, J., Richardson, M., & Whitelaw, R. (2008). The myth of long-horizon

predictability, Review of Financial Studies, 21: 1577-1605.

296

Bouwman, K. E., & Jacobs, J. P. A. M. (2011). Forecasting with real-time macroeconomic

data: the ragged-edge problem and revisions, Journal of Macroeconomics, 33:

784-792.

Boyd, J. H., Hu, J., & Jagannathan, R. (2005). The stock market's reaction to

unemployment news: why bad news is usually good for stocks Journal of Finance,

60: 649-672.

Bozos, K., Nikolopoulos, K., & Ramgandhi, G. (2011). Dividend signaling under

economic adversity: Evidence from the London Stock Exchange, International

Review of Financial Analysis, 20(5): 364-374.

Branch. B. (1976). The predictive power of stock Malaya market indicator, The Journal of Financial and Quantitative Analysis, of11(2): 269-285. Brandt, M. W., & Wang, K. Q. (2003). Time-varying risk aversion and unexpected

inflation, Journal of Monetary Economics, 50: 1457-1498.

Brav, A., & Heaton J. B. (2002). Competing theories of financial anomalies, The Review

of Financial Studies, (Special Issue: Conference on Market Frictions and

Behavioral Finance: 15(2): 575-606.

Brave, S. (2009). The Chicago Fed National Activity Index and business cycles, Chicago

UniversityFed Letter, Federal Reserve Bank of Chicago, No. 268, November.

Bree, D. S., & Joseph, N. L. (2010). Fitting the log periodic power law to financial crashes:

A critical analysis, arXiv:1002.1010 [q_fin.ST]: Retrieved on 26 April, 2011, from:

http://arxiv.org/pdf/1002.1010v2.pdf

297

Breen, W., Glosten, L. R., & Jagannathan, R. (1989). Economic significance of

predictable variations in stock index returns, Journal of Finance, 44: 1177-1189.

Brooks, C., & Katsaris, A. (2005). A three-regime model of speculative behaviour:

Modelling the evolution of the S&P 500 composite index, The Economic Journal,

115(505): 767-797.

Brown, D. P., & Jennings, R. H. (1989). On technical analysis, Review of Financial

Studies, 2(4): 527-551.

Brunnermeier, M. K. (2008). Bubbles. In Durlauf, S. N., & Blume, L. E. (Eds.). The New

nd Palgrave Dictionary of Economics (2 ed.). London:Malaya Palgrave Macmillan. Brunnermeter, M. K., & Morgan, J. (2010). Clock games: theory and experiments, Games and Economics Behavior, 68: 532-550.of Bulow, J., & Klemperer, P. (1994). Rational frenzies and crashes, Journal of Political

Economy, 102(1): 1-23.

Bureau of Economic Analysis. (2006). Why does BEA publish percent changes in

quarterly series at annual rates? Retrieved on 7 March, 2013, from:

http://www.bea.gov/faq/index.cfm?faq_id=122#sthash.Kpg3P6lz.dpuf BureauUniversity of Labor Statistics. (2009). BLS Handbook of Methods. U.S. Department of Labor. Retrieved on 17 January, 2011, from the Bureau of Labor Statistics:

http://www.bls.gov/opub/hom/pdf/homch17.pdf

Burns, A. F., and Wesley C. Mitchell. (1946). Measuring Business Cycles. New York:

National Bureau of Economic Research, Inc.

298

Cai, Y., Chou, R. Y., & Li, D. (2009). Explaining international stock correlations with CPI

fluctuations and market volatility, Journal of Banking & Finance, 33: 2026-2035.

Cajueiro, D. O., Tabak, D. M., & Werneck, F. K. (2009). Can we predict crashes: The

case of the Brazilian stock market, Physica A, 388: 1603-1609.

Camacho, M. & Perez-Quiros, G. (2002). This is what the leading indicators lead, Journal

of Applied Econometrics, 17(1): 61-80.

Camerer, C. F. (1987). Do biases in probability judgment matter in markets? Experimental

evidence, The American Economic Review, 77(5): 981-997. Camerer, C. F. (1989). Bubbles and fads in asset prices,Malaya Journals of Economic Surveys, 3(1): 3-41.

Campbell, J. Y. (1987). Stock returns and termof structure, Journal of Financial Economics, 18: 373-399.

Campbell, J. Y. (2008). Estimating the equity premium, Canadian Economic Review, 41:

1-21.

Campbell, J. Y., & Shiller R. J. (1998). Valuation ratios and the long-run stock market

outlook, Journal of Portfolio Management, 24: 11-26.

Campbell,University J. Y., & Shiller, R. J. (1988). Stock prices, earnings, and expected dividends, Journal of Finance, 43: 661-676.

Campbell, J. Y., & Vuolteenaho, T. (2004). Inflation illusion and stock prices, American

Economic Review, 94: 19-23.

299

Campbell, J. Y., & Yogo, M. (2006). Efficient tests of stock return predictability, Journal

of Financial Economics, 81: 27-60.

Campbell. J. Y., & Thompson, B. S. (2008). Predicting excess stock returns out of sample:

Can anything beat the historical average? The Review of Financial Studies, 21(4):

1509-1531.

Candelon, B., Piplak, J., & Straetmans, S. (2008). On measuring synchronization of bulls

and bears: The case of East Asia, Journal of Banking & Finance 32 (6): 1022-1035.

Caplin, A. and Leahy, J. (1994). Business as usual, market crashes and wisdom after the fact, American Economic Review, 84: 548-565.Malaya Carpenter S. B. & Demiralp, S. (2010). Money, reserves, and the transmission of monetary policy: does the money multiplierof exist?, Finance and Economics Discussion Series 2010-41. Retrieved on 9 October, 2016, from the Board of

Governors of the Federal Reserve System (U.S.):

https://www.federalreserve.gov/pubs/feds/2010/201041/201041pap.pdf

Carroll, C.D. 2001. The Epidemiology of Macroeconomic Expectations, NBER Working

Paper, W8695. Retrieved on 26 April, 2010, from the National Bureau of

Economic Research website: http://www.nber.org/papers/w8695

Casarin,University R., & Trecroci, C. (2007). Business cycle and stock market volatility: Are they

related? Retrieved on 19 February, 2011, from the Social Science Electronic

Publishing, Inc (SSRN):

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=888524

300

Case. K. E., & Shiller, R. J. (2003). Is there a bubble in the housing market? Brookings

Papers on Economic Activity, 34(2): 299-362.

Cencini, M., Cecconi, F., & Vulpiani, A. (2010). Chaos: from simple models to complex

systems, Singapore: World Science Publishing.

Chainais, P., Reidi, R. & Abry, P. (2005). On non-scale-invariant infinitely divisible

cascades, IEEE Transactions on Information Theory, 51(3): 1063-1083.

Chan, K., Ikenberry, D. L., & Lee, I. (2007). Do managers time the market? Evidence

from open-market share repurchases, Journal of Banking & Finance, 31: 2673- 2694. Malaya Chang, C., Jiang, J., & Kim, K. A. (2009). A test of the representativeness bias effect on stock prices: A study of Super Bowlof commercial likeability, Economics Letters, 103: 49-51.

Chang, G., & Feigenbaum, J. (2006). A Bayesian analysis of log-periodic precursors to

financial crashes, Quantitative Finance, 6: 15-36.

Chang, G., & Feigenbaum, J. (2008). Detecting log-periodicity in a regime-switching

model of stock returns, Quantitative Finance, 8: 723-738. Chang,University K. L. (2009). Do macroeconomic variables have regime-dependent effects on stock return dynamics? Evidence from the Markov regime switching model,

Economic Modelling, 26: 1283-1299.

Chauvet, M. (1999). Stock market fluctuations and the business cycle, Journal of

Economic and Social Measurement, 25: 235-258.

301

Chauvet, M., & Potter, S. (2000). Coincident and leading indicators of stock market,

Journal of Empirical Finance, 7: 87-111.

Chavarnakul, T., & Enke, D. (2009). A hybrid stock trading system for intelligent

technical analysis-based equivolume charting, Neurocomputing, 72: 3517-3528.

Chen, N-F. (1991). Financial investment opportunities and the macroeconomy, Journal of

Finance, 46: 529-554.

Chen, N-F., Roll, R., & Ross, S. (1986). Economic forces and the stock market, Journal of

Business 59: 383-403. Chen, S. (2012). The predictability of aggregate JapaneseMalaya stock returns: Implications of dividend yield, International Review of Economics & Finance, 22(1): 284-304.

Chen, S-S. (2009). Predicting the bear stockof market: Macroeconomic variables as leading

indicators, Journal of Banking & Finance, 33: 211-223.

Chen, Z., & Duan, Q. (2011). New models of trader beliefs and their application for

explaining financial bubbles, Economic Modelling, 28: 2215-2227.

Chin, T. C., Mital, D. P., & Chua, H. C. (2000). Interaction of mean and variance of time

series: A Markov switching analysis, Knowledge-Based Intelligent Information UniversityEngineering Systems & Allied Technologies, KES 2000: 423-427. Cho, D. I., & Ogwang, T. (2006). Conceptual perspectives on selecting the principal

variables in the Purchasing Managers' Index, Journal of Supply Chain

Management, 42(2): 44-52.

302

Choi, H., & Jayaraman, N. (2009). Is reversal of large stock-price declines caused by

overreaction or information asymmetry: Evidence from stock and option markets,

The Journal of Futures Markets, 4: 348-376.

Chordia, T., & Shivakumar, L. (2001). Earnings, Business Cycle and Stock Returns.

Retrieved on 9 November, 2010, from the Social Science Electronic Publishing,

Inc (SSRN): http://papers.ssrn.com/sol3/papers.cfm?abstract_id=281491

Chordia, T., & Shivakumar, L. (2002). Momentum business cycle and time-varying

expected returns, The Journal of Finance, 57(2): 985-1019.

Chortareas, G., & Noikokyris E. (2014). Oil shocks, stock market prices, and the U.S. dividend yield decomposition, International ReviewMalaya of Economics & Finance, 29: 639-649. of Choudhry, C. (1996). Stock market volatility and the crash of 1987: evidence from six

emerging market, Journal of International Money and Finance, 15: 969-981.

Chung, S-L., Hung, C-H., & Yeh, C-Y. (2012). When does investor sentiment predict

stock returns? Journal of Empirical Finance, 19: 217-240.

Clapham, C., & Nicholson, J. (2009). The Concise Oxford Dictionary of Mathematics (4th Universityed.). New York: Oxford University Press. Clark, T. E., & West, K. D. (2007). Approximately normal tests for equal predictive

accuracy in nested models, Journal of Econometrics, 138(1): 291-311.

Clarke, T. E. (2004). Can out-of-sample forecast comparisons help prevent overfitting?,

Journal of Forecasting, 23(2): 115-139.

303

Clement, M. B., Hales, J., & Xue, Y. (2011). Understanding analysts' use of stock returns

and other analysts' revisions when forecasting earnings, Journal of Accounting and

Economics, 51(3): 279-299.

Coakley, J., & Fuertes, A. M. (2006). Valuation ratios and price deviations from

fundamentals, Journal of Banking and Finance, 30(8): 2325-2346.

Cochrane, J. H. (1991). Production-based asset pricing and the link between stock returns

and economic fluctuations, The Journal of Finance, 46(1): 209-237.

Cochrane, J. H. (2008). Financial markets and the real economy, in Mehra, R. (Ed.), Handbook of the Equity Risk Premium: 237-325.Malaya San Diego: Elsevier. Cohen, R., Polk, C., & Vuolteenaho, T. (2005). Money illusion in the stock market: The Modigliani-Cohn hypothesis, Quarterlyof Journal of Economics, 120: 639-668. Colander, D. 2000. A Thumbnail sketch of the history of thought from a complexity

perspective. In Colander, D. (Ed.), Complexity and the History of Economics

Thought (pp. 31-44). London: Routledge.

Colby, R. W. (2002). The Encyclopedia of Technical Market Indicators (2nd ed.). New

York: McGraw-Hill. Conlon.University J. R. (2004). Simple finite horizon bubbles robust to higher order knowledge, Econometrica, 72(3): 927-936.

Cooper, G. (2008). The Origin of Financial Crisis. New York: Vintage Books.

304

Cootner, P. (Ed.) (1964). The Random Character of Stock Market Prices. Cambridge:

M.I.T.

Cormier, D., & Martinez, I. (2006). The association between management earnings

forecasts, earnings management, and stock market valuation: Evidence from

French IPOs, The International Journal of Accounting, 41(3): 209-236.

Cross, R., Grinfeld, M., Lamba, H., & Seaman, T. (2005). A threshold model of investor

psychology, Physica A, 354: 463-478.

Cunado, J., Gil-Alana, L. A., & de Gracia, F. P. (2010). Mean reversion in stock market

prices: New evidence based on bull and bear markets, Research in International

Business and Finance, 24: 113-122. Malaya

Currie, L. (1934). The failure of monetary ofpolicy to prevent the depression of 1929-32, Journal of Political Economy, 42(2): 145-177.

Daniel, K., & Titman, S. (1997). Evidence on the characteristics of cross-sectional

variation in stock returns, Journal of Finance, 52: 1-33.

Daniel, K., Hirshleifer, D., & Subrahmanyam, A. (1998). Investor psychology and

security market under- and overreactions, Journal of Finance, 53(6): 1839-1885.

Daniel,University K., Hirshleifer, D., & Teoh, S. H. (2002). Investor psychology in capital markets: evidence and policy implications, Journal of Monetary Economics, 49(1): 139-209.

305

Dargham, N. A. S. (n.d.). The implications of Behavioral Finance. (Chargée

d’enseignement à FGM). Retrieved on 13 June, 2012, from:

http://www.fgm.usj.edu.lb/files/a102009.pdf

Dasilas, A., & Leventis, S. (2011). Stock market reaction to dividend announcements:

Evidence from the Greek stock market, International Review of Economics &

Finance, 20(2): 302-311.

De Bondt, W. (1995). Investor psychology and the dynamics of security prices, in Wood,

A. S. (Ed.). Behavioral Finance and Decision Theory in Investment Management

(pp. 7-13). Virginia: AIMR.

De Bondt, W. F. M. (1991). What do economists knowMalaya about stock market? Journal of Portfolio Management, 17: 84-91. of De Bondt, W. F. M., & Thaler, R. H. (1985). Does the stock market overreact? Journal of

Finance, (40)3: 793-808.

De Chassart, M. D., & Firer, C. (2004). Risks associated with market timing under

different market conditions, Omega, 32: 201-211.

De Grauwe, P., & Grimaldi, M. (2004). Bubbles and crashes in behavioural finance model,

Sveriges Riksbank Working Paper Series, 164. . Retrieved on 30 October, 2013,

Universityfrom Econstor: http://www.econstor.eu/bitstream/10419/82489/1/wp_164.pdf

De Long, J. B., & Shleifer, A. (1991). The stock market bubble of 1929: Evidence from

closed-end mutual funds, Journal of Economic History, 51(3): 675-700.

306

De Long, J. B., Shleifer, A. M., Summer, L. H., & Waldmann, R. J. (1990). Noise trader

risk in financial markets, Journal of Political Economy, 98: 703-38.

De Long, J. B., Shleifer, A. M., Summer, L. H., & Waldmann, R. J. (1991). The survival

of noise traders in financial markets, Journal of Business, 64(1): 1-19.

Dezhbakhsh, H., & Demirgue-Kunt, A. (1990). On the presence of speculative bubbles in

stock prices, The Journal of Financial and Quantitative Analysis, 25(1): 101-112.

Diavatopoulos, D., Doran, J. S., Fodor, A., & Peterson, D. R. (2012). The information

content of implied skewness and kurtosis changes prior to earnings announcements for stock and option returns, Journal of BankingMalaya & Finance, 36(3): 786-802. Diba, B. T., & Grossman, H. I. (1987). On the inception of rational bubbles. Quarterly Journal of Economics, 102: 697-700. of Diba, B. T., & Grossman, H. I. (1988a). The theory of rational bubbles in stock prices, The

Economic Journal, 98(392): 746-754.

Diba, B. T., & Grossman, H. I. (1988b). Explosive rational bubbles in stock prices, The

American Economic Review, 78(3): 520-530.

Diebold, F. X., & Rudebusch, G. D. (1989). Scoring the leading indicators, Journal of UniversityBusiness, 62(3): 369-391. Domian, D. L., & Louton, D. A. (1995). Business cycle asymmetry and the stock market,

The Quarterly Review of Economics and Finance, 35(4): 451-466.

307

Donaldson, R. G., & Kamstra, M. (1996). A new dividend forecasting procedure that

rejects bubbles in asset price: The case of 1929's stock crash, The Review of

Financial Studies, 9(2): 333-383.

Dopke, J., Hartmann, D & Pierdzioch, C. (2008). Real-time macroeconomic data and ex

ante stock return predictability, International Review of Financial Analysis, 17:

274-290.

Downes, J., & Goodman, J. E. (2010). Finance & Investment Handbook (8th ed). New

York: Barron’s Educational Series Inc.

Drobetz, W., Kaiser, D., & Zimbehl, J. (2012). Efficient hedge fund style allocations: A rules-based model, Journal of Derivatives & HedgeMalaya Funds, 18: 274-300. Drozdz, S., Kwapien, J., Oswiecimka, P., & Speth,of J. (2008). Current log-periodic view on future world market development, Acta Physica Polonica A: 114: 539-546.

Du, D., Denning, K. & Zhao, X. (2012). Real aggregate activity and stock returns, Journal

of Economics and Business, 64: 323-337.

Duffy, J., & Unver, M. U. (2006). Bubbles and crashes with near-zero-intelligence traders,

Economic Theory, 27(3): 537-563. Dumas,University B. (1994). A test of the international CAPM using business cycles indicators as instrumental variables, in Frankel, J. A. (Ed.). The Internationalization of Equity

Markets (pp. 23-58). Massachusetts: NBER Books.

Dupernex, S. (2007). Why might share prices follow a random walk? Student Economic

Review, 21: 167-179.

308

Eagle, D. (1994). The equivalence of the cascading scenario and the backward-bending

demand curve theory of the 1987 stock market crash, Quarterly Journal of

Business and Economics, 33(4): 60-73.

Edie, L. D. (1930). The bank and the stock market crisis of 1929, The Journal of Business

of the University of Chicago, 3(1): 16-21.

Edwards, R. D., Magee, J., & Bassetti, W. H. C. (2007). Technical Analysis of Stock

Trends (9th ed.). New York: AMACOM .

Edwards, S., Biscarri, J. G., & de Gracia, F. P. (2003). Stock market cycles, financial

liberalization and volatility, Journal of International Money and Finance, 22: 925- 955. Malaya Ehrmann, M., & Fratzscher, M. (2004). Takingof stock: Monetary policy transmission to equity markets, Journal of Money, Credit and Banking, 36(4): 719-737.

Eilifsen, A., Knivsflå, K. H., & Sættem, F. (2001) Earnings announcements and the

variability of stock returns, Scandinavian Journal of Management, 17(2): 187-200.

Elton, E. J., Gruber, M. J., Brown, S. J., & Goetzmann, W. N. (2007). Modern Portfolio

and Investment Analysis. New York: John Wiley & Sons, Inc.

Engel,University C., & Hamilton, J. D. (1990). Long swings in the dollar: Are they in the data and do markets know it?, American Economic Review, 80(4): 689-713.

309

Engle, R. F. (1990). Stock volatility and the crash of '87: Discussion, The Review of

Financial Studies, (National Bureau of Economic Research Conference: Stock

Market Volatility and the Crash, Dorado Beach, March 16-18, 1989) 3(1): 103-106.

Estrella, A. (1998). A new measure of fit for equations with dichotomous dependent

variables, Journal of Business and Economic Statistics, 16(2): 198-205.

Estrella, A., & Mishkin, F. S. (1996). The yield curve as a predictor of U.S. recessions,

Current Issues in Economics and Finance, 2(7): 1-6.

Estrella, A., & Mishkin, F. S. (1998). Predicting U.S. recessions: Financial variables as leading indicators. Review of Economics and StatisticsMalaya, 80(1): 45-61. Estrella, A., & Trubin, M. R. (2006). The yield curve as a leading indicator: Some practical issues, Current Issues in Economicsof and Finance (12)5: 1-7. Evans, C. L., Liu, C. T., & Pham-Kanter, G. (2002). The 2001 recession and the Chicago

Fed National Activity Index: Identifying business cycle turning points, Economic

Perspectives, 26(3): 26-43.

Evans, G. W. (1991). Pitfalls in testing for explosive bubbles in asset prices. American

Economic Review, 81: 922-30. Faber,University M. T. (2012). Global value: Building trading models with the 10 year CAPE, Cambria Quantitative Research, 5: 1-15.

Fama, E. F. (1991). Efficient Capital Markets II, Journal of Finance, 46: 1575-1617

310

Fama, E. F. (1965). The behavior of stock market prices, The Journal of Business, 38(1):

34-105.

Fama, E. F. (1970). Efficient Capital Markets: A review of theory and empirical work,

Journal of Finance, 25(2): 383-417.

Fama, E. F. (1991). Efficient Capital Markets II, Journal of Finance, 46(5): 1575-1617.

Fama, E. F. (1998). Market efficiency, long-term returns, and behavioral finance, Journal

of Financial Economics, 49: 283-306.

Fama, E. F., & French, K. R. (1988). Dividend yields and expected stock returns Journal of Financial Economics, 22(1): 3-25. Malaya Fama, E. F., & French, K. R. (1989). Business conditions and expected returns on stock

and bonds, Journal of Financial Economics,of 25: 23-49.

Fama, E. F., & French, K. R. (1992). The cross-section of expected stock returns, The

Journal of Finance, 47(2): 427-465.

Fama, E. F., & French, K. R. (1995). Size and book-to-market factors in earnings and

returns, Journal of Finance, 50, 131-155.

Fama, E. F., & Schwert, G. W. (1977). Asset returns and inflation, Journal of Financial UniversityEconomics, 5: 115-146.

Fantazzini, D., & Geraskin, P. (2013). Everything you always wanted to know about log

periodic power laws for bubble modelling but were afraid to ask, European

Journal of Finance, 19(5): 366-391.

311

Federal Reserve Bank of Chicago (2013). Background on the Chicago Fed National

Activity Index. Retrieved on 17 April, 2014, from:

http://www.chicagofed.org/digital_assets/publications/cfnai/background/cfnai_bac

kground.pdf

Federal Reserve Bank of New York (n.d.). Estimating Yields on Treasury Securities.

Retrieved on 17 February, 2014, from:

http://www.newyorkfed.org/aboutthefed/fedpoint/fed28.html

Feigenbaum, J.A., & Freund, P. G. O. (1996). Discrete scale invariance in stock markets

before crashes, Int. J. Mod. Phys. B, 10: 3737-3745.

Feigenbaum, J.A., & Freund, P. G. O. (1998). DiscreteMalaya scale invariance and the second Black Monday, Modern Physics Lettersof B, 12: 57-60. Feigenbaum. J. (2001). A statistical analysis of log-periodic precursors to financial crashes.

Quantitative Finance, 1: 346-360.

Fernandes, M. (2006). Financial crashes as endogenous jumps: Estimation, testing and

forecasting, Journal of Economic Dynamics & Control, 30: 111-141.

Ferreira, M. A., & Santa-Clara, P. (2011). Forecasting stock market returns: The sum of Universitythe parts is more than the whole, Journal of Financial Economics, 100: 514-537. Finucane, T. J. (1999). A new measure of the direction and timing of information flow

between markets, Journal of Financial Markets, 2: 135-151.

312

Fisher (1930). The stock market panic in 1929, Journal of the American Statistical

Association, (Supplement: Proceedings of the American Statistical Association)

25(169): 93-96.

Fisher, J. D. M. (2000). Forecasting inflation with a lot of data, Chicago Fed Letter,

Federal Reserve Bank of Chicago, No. 151, March.

Flannery, M. J., & Protopapadakis, A. A. (2002). Macroeconomic factors do influence

aggregate stock returns, Review of Financial Studies, 15: 751-782.

Flood, R. P., & Garber, P. M. (1980). Market fundamentals versus price level bubbles: First test, Journal of Political Economy, 88: 745-770.Malaya Flood, R. P., & Hodrick, R. J. (1990). On testing for speculative bubbles, The Journal of Economic Perspectives, 4(2): 85-101.of Flood, R. P., & Garber, P. M. (1982) Bubbles, runs, and gold monetization, in P. Wachtel,

ed. Lexington, M. A. Crises in Economics and Financial Structure: 275-293.

Massachusetts: Lexington Books.

Flood, R. P., Hodrick, R. J., & Kaplan, P. (1986). An evaluation of recent evidence on

stock market bubbles. NBER Working Paper, 1971. Retrieved on 14 May, 2012,

from The National Bureau of Economic Research (NBER):

Universityhttp://www.nber.org/papers/w1971

Focardi, S., Cincotti, S., & Marchesi, M. (2002). Self organization and market crashes,

Journal of Economic Behavior & Organization, 49: 241-267.

313

French, K. R. (1988). Crash-testing the efficient market hypothesis, NBER

Macroeconomics Annual, 3: 277-285.

Friedman, D., & Abraham, R. (2009). Bubbles and crashes gradient dynamics in financial

markets, Journal of Economic Dynamics & Control, 33: 922-937.

Froot, K. A., & Obstfeld, M. (1991). Intrinsic bubbles: The case of stock prices, The

American Economic Review, 81(5): 1189-1214.

Fry, J. M. (2012). Exogenous and endogenous market crashes as phase transitions in

complex financial systems, The European Physical Journal B, 85(12): 1-6. Furbush, D. (1989). Program trading and price movement:Malaya Evidence from the October 1987 market crash, Financial Management, 18(3): 68-83.

Galbraith. J.K. (2009). The Great Crash 1929of (Paperback ed.). London: Penguin Books.

Gammill, J. F., & Marsh, T. A. (1988). Trading activity and price behavior in the stock

and stock index futures markets in October 1987, The Journal of Economic

Perspectives, 2(3): 25-44.

Garber, P. M. (1992). Crashes in Newman, P., Milgate, M., & Eatwell, J. (Eds.). The New

Palgrave Dictionary of Money and Finance, Volume 1 (pp. 511-513). London: UniversityMacmillan Press Limited. Garcia, E. (2005). Predator-prey: An alternative model of stock market bubbles and the

business cycle, European Journal of Economics and Economic Policies:

Intervention, 2(2): 77-105.

314

Gencay, R. (1998). Optimization of technical trading trategies and the profitability in

security markets, Economics Letters, 59: 249-254.

Gencay, R. (1998). The predictability of security returns with simple technical trading

rules, Journal of Empirical Finance, 5: 347-359.

Gennotte, G., & Leland, H. (1990). Market liquidity, hedging, and crashes, The American

Economic Review, 80(5): 999-1021.

Glassman, D. A., & Riddick, L. A. (2006). Market timing by global fund managers,

Journal of International Money and Finance, 25: 1029-1050. Goetzmann, W. N., & Jorion, P. (1993). Testing the Malayapredictive power of dividend yield, Journal of Finance, 48(2): 663-679.

Gonzalez, L., Powell, J. G., Shi, J., & Wilson,of A. (2005). Two centuries of bull and bear

market cycles, International Review of Economics and Finance 14: 469-486.

Gordon, S., & St-Amour, P. (2000). A Preference regime model of bull and bear markets,

The American Economic Review, 90(4): 1019-1033.

Gorton, G. (1988). Banking Panics and Business Cycles. Oxford Economic Papers, 40(2):

751-781.

Graham,University B., & Dodd, D. (2008). Security Analysis (6th ed.). New York: McGraw-Hill.

Greenspan, A. (1987). The great crash: Past and present: The great malaise, Challenge,

30(6): 11-14.

315

Greenwood, R., & Nagel, N. (2009). Inexperienced investors and bubbles, Journal of

Financial Economics, 93: 239-258.

Grinblatt, M., & Titman, S., Wermers, R. (1995). Momentum investment strategies

portfolio performance and herding: A study of mutual fund behavior, American

Economic Review, 85: 1088-1105.

Gros, C. 2008. Complex and Adaptive Systems: A Primer. Heidelgerg: Spinger

Grossman, S. J., & Shiller. R. J. (1981). The determinants of the variability of stock prices,

American Economic Review, 71: 222-227. Grossman, S. J., & Stiglitz, J. E. (1980). On the impossibilityMalaya of informationally efficient markets, The American Economic Review, 70(3): 393-408.

Guerrero, F. L., Stone, G. R., & Sundali, J. A.of (2012). Fear in asset allocation during and after stock market crashes. An experiment in behavioral finance, The Journal of

Behavioral Finance & Economics, 2(1): 50-81.

Guidolin, M., & Timmermann, A. (2005). Economic implications of bull and bear regimes

in UK stock and bond returns, The Economic Journal, 115(500): 111-143.

Gupta, R., & Modise, M. P. (2012). South African stock return predictability in the Universitycontext data mining: The role of financial variables and international stock returns, Economic Modelling, 29: 908-916.

Hamilton, J. D. (1986). On testing for self-fulfilling speculative price bubbles,

International Economic Review, 27: 545-552.

316

Hamilton, J. D. (1989) A new approach to the economic analysis of non-stationary time

series and the business cycle, Econometrica, 57: 357-384.

Hamilton, J. D., & Gang, L. (1996). Stock Market Volatility and the Business Cycle,

Journal of Applied Econometrics, 11(5): 573-593.

Hamilton, J. D., & Whiteman C. H. (1985). The observable implications of self-fulfilling

speculative price bubbles. Journal of Monetary Economics, 16: 353-373.

Hamilton, W. P. (1998). The Stock Market Barometer. New York: John Wiley & Sons, Inc.

Hanker, T, & Wang J-X. (2006). On the importance of timing specifications in market microstructure research, Journal of Financial MarketsMalaya, 9: 162-179. Harmon, D., De Aguiar, M. A. M., Chinellato, D. D., Braha, D., Epstein, I. R., & Bar- Yam, Y. (2011). Predicting economicof market crises using measures of collective panic, arXiv:1102.2620v1 [q_fin.ST]: Retrieved on 17 October, 2012, from:

http://arxiv.org/pdf/1102.2620.pdf

Harris, L. (1989). The October 1987 S&P 500 stock-futures basis, The Journal of Finance,

44(1): 77-99.

Harrison, M., & Kreps, D. (1978), Speculative investor behavior in a stock market with Universityheterogenous expectations, Quarterly Journal of Economics, 92: 323-336. Hartmann, D., Kempa, B., & Pierdzioch, C. (2008). Economic and financial crises and the

predictability of U.S. stock returns, Journal of Empirical Finance, 15: 468-480.

317

Hausman, D. M. (2013). Philosophy of Economics. In Zalta, E. D. (Ed.), The Stanford

Encyclopedia of Philosophy. Retrieved 13 June, 2016, from the Stanford

Encyclopedia of Philosophy (SEP):

http://plato.stanford.edu/archives/win2013/entries/economics/

Hayek, F.A. (1964). The theory of complex phenomena. In Bunge, M. (Ed.), The Critical

Approach to Science and Philosophy (pp.332-349). London: Collier McMillan.

Hearth, D., & Zaima, J. (2003). Contemporary Investments: Security and Portfolio

Analysis, (4th ed.). Ohia: South Western.

Heij, C., van Dijk, D., & Groenen, P. J. F. (2008). Macroeconomic forecasting with matched principal components, International JournalMalaya of Forecasting, 24: 87-100. Heij, C., van Dijk, D., & Groenen, P. J. F. (2011ofa). Real-time macroeconomic forecasting with leading indicators: An empirical comparison, International Journal of

Forecasting, 27: 466-481.

Heij, C., van Dijk, D., & Groenen, P. J. F. (2011b). Forecasting with leading indicators by

means of the principal covariate index, OECD Journal: Journal of Business Cycle

Measurement and Analysis, 4(1): 73-92.

Heller, W. W. (1987). The great crash: past and present: Can there be another crash?

UniversityChallenge, 30(6): 6-10.

Henriksson, R. D., & Merton, R. C. (1981). On market timing and investment

performance II: Statistical procedures for evaluating forecasting skills, Journal of

Business, 54: 513-534.

318

Hess. M. K. (2003). What drives Markov regime-switching behavior of stock markets:

The Swiss case, International Review of Financial Analysis, 12: 527-543.

Ho, Y. J., Ruiz Estrada, M. A., & Yap, S. F. (2016). The evolution of complex systems

theory and the advancement of econophysics methods in the study of stock

markets crashes, Labuan Bulletin of International Business & Finance, 14: 68-83.

Ho, Y. J., Ruiz Estrada, M. A., & Yap, S. F. (2017). Examining the heterogeneous regimes

of stock market identified with two variants of B-B algorithms that differ in

rigidness of specification, Labuan Bulletin of International Business & Finance,

(forthcoming).

Ho, Y. J., & Quah, C. H. (2012). Explaining slow sell-offMalaya at stock market after yield curve inversion with clock game, Actual Problemsof of Economics, 11: 493-500. Hodrick, R. (1992). Dividend yields and expected stock returns: Alternative procedures

for inference and measurement, Review of Financial Studies, 5: 357-386.

Hogkin, L. H. (2005). A history of mathematics: from Mesopotamia to modernity. New

York: Oxford Press.

Homm, U., & Breigung, J. (2012). Testing for speculative bubbles in stock markets: A

comparison of alternative methods, Journal of Financial Econometrics, 10: 198-

University231.

Hong, H., & Stein, J. (1999). A unified theory of underreaction, momentum trading, and

overreaction in asset markets, The Journal of Finance, 54(6): 2143-2184.

319

Hong, H., Lim, T., & Stein, J. (2000). Bad news travels slowly: Size, analyst coverage,

and the profitability of momentum strategies, The Journal of Finance, 55(1): 265-

295.

Hong, H., Scheinkman, J., & Xiong, W. (2008). Advisors and asset prices: A model of the

origins of bubbles, Journal of Financial Economics, 89: 268-287.

Hong, H., Torous, W., & Valkanov, R. (2007). Do industries lead stock markets, Journal

of Financial Economics, 83(2): 367-396.

Howard, J. A. (2012). Behavioral finance: Contributions of cognitive psychology and

neuroscience to decision making, Journal of Organizational Psychology, 12(2): 52-70. Malaya Huang, R. D., & Kracaw, W. A. (1984). Stockof market returns and real activity: A note, Journal of Finance 39: 267-273.

Hugonnier, J. (2012). Rational asset pricing bubbles and portfolio constraints, Journal of

Economic Theory, 147: 2260-2302.

Ibbotson, R., & Chen, P. (2003). Long-run stock returns: Participating in the real economy,

Financial Analysts Journal, 59: 88-98. Ikeda,University S., & Shibata, A. (1992). Fundamentals-dependent bubbles in stock prices. Journal of Monetary Economics, 30: 143-168.

Information on Recessions and Recoveries, the NBER Business Cycle Dating Committee,

and related topics. (n.d.). Retrieved 18 September, 2011, from The National

Bureau of Economic Research (NBER): http://www.nber.org/cycles/main.html;

320

Issler, J. V., & Vahid, F. (2006). The missing link: Using the NBER recession indicator to

construct coincident and leading indices of economic activity, Journal of

Econometrics, 132: 281-303.

Jacklin, C. J., Kheidon, A. W., & Pfleiderer, P. (1992). Underestimation of portfolio

insurance and the Crash of October 1987, The Review of Financial Studies, 5(1):

35-63.

Jacobsson, E. (2009). How to Predict Crashes in Financial Markets with the Log-Periodic

Power Law. (Dissertation). Stockholm: Mathematical Statistics Stockholm

University. Retrieved on 10 October, 2012, from: www2.math.su.se/matstat/reports/serieb/2009/rep7/report.pdfMalaya Jagadeesh, N. (1990). Evidence of predictable behavior of security returns, Journal of

Finance, 45(3): 881-898. of

Jagannathan, R., & Korajczyk, R. (1986). Assessing the market timing performance of

managed portfolios, Journal of Business, 59: 217-236.

Jahne, W. W. (1987). The crash of '87, Financial Analysts Journal, 43(6): 6-9.

Jeery, R. H. (1984). The folly of stock market timing, Harvard Business Review, 62: 102- University110. Jensen, M. C. (1978). Some anomalous evidence regarding market efficiency, Journal of

Financial Economics, 6(2/3): 95-101.

Jiang X., & Lee, B-S. (2007). Stock returns, dividend yield, and book-to-market ratio,

Journal of Banking & Finance, 31: 455-475.

321

Jiang, G. J., Yao, T., & Yu, T. (2007). Do mutual funds time the market? Evidence from

portfolio holdings, Journal of Financial Economics, 86: 724-758.

Jiang, W. (2003). A nonparametric test of market timing, Journal of Empirical Finance,

10: 399-425.

Jiang, Z-Q., Zhou, W-Z., Sornette, D., Woodard, R., Bastiaense, K., & Cauwels, P. (2010).

Bubble diagnosis and prediction of the 2005-2007 and 2008-2009 Chinese stock

market bubbles, Journal of Economic Behavior & Organization, 74: 149-162.

Johansen, A. (1997). Discrete Scale Invariance and Other Cooperative Phenomena in

Spatially Extended Systems with Threshold Dynamics. (Ph.D thesis) Copenhagen: CATS, Neils Bohr Institute, University of Copenhagen.Malaya Johansen, A. (2003). Characterization of of large price variations in financial markets, Physica A, 234: 157-166.

Johansen, A. (2004). Origin of crashes in three US stock markets shocks and bubbles,

Physica A, 338: 135-142.

Johansen, A. Sornette, D., & Ledoit, O. (1999). Predicting financial crashes using discrete

scale invariance, Risk, 1: 5-32. Johansen,University A., & Sornette, D. (1998a). Evidence of discrete scale invariance by canonical averaging, International Journal of Modern Physics C, 9: 219-255.

Johansen, A., & Sornette, D. (1998b). Stock market crashes are outliers, European

Physical Journal B, 1: 141-143.

322

Johansen, A., & Sornette, D. (1999a). Critical crashes, Risk, 12(1): 91-94.

Johansen, A., & Sornette, D. (1999b). Financial anti-bubbles: Log-periodicity in gold and

Nikkei collapses, International Journal of Modern Physics C, 10: 563-575.

Johansen, A., & Sornette, D. (1999c). Modeling the stock market prior to large crashes,

European Physical Journal B, 9: 167-174.

Johansen, A., & Sornette, D. (2000). The Nasdaq crash of April 2000: Yet another

example of log-periodicity in a speculative bubble ending in a crash, The European

Physical Journal B, 17: 391-328. Johansen, A., & Sornette, D. (2001). Bubbles and anti-bubblesMalaya in Latin-American, Asian and Western stock markets: An empirical study, International Journal of Theoretical and Applied Finance, 4: 853-920.of Johansen, A., & Sornette, D. (2002). Large stock market price drawdowns are outliers,

Journal of Risk, 4: 69-110.

Johansen, A., & Sornette, D. (2008). Endogenous versus exogenous crashes in financial

markets, arXiv (e-print), cond-mat/0210509v1: 1-36

Johansen, A., & Sornette, D. (2010). Shocks, crashes and bubbles in financial markets, UniversityBrussels Economic Review, 53(2): 201-253. Johansen, A., Ledoit, O., & Sornette. D. (2000). Crashes as critical points, International

Journal of Theoretical and Applied Finance, 3: 219-225.

323

Jones. S. L., & Netter, J. M. (2008). Efficient capital markets, in Henderson, D. R. (Ed),

The Concise Encyclopedia of Economics (2nd ed.). Indianapolis: Liberty Fund.

Jorgensen, B., Li, J., & Sadka, G. (2012). Earnings dispersion and aggregate stock returns,

Journal of Accounting and Economics, 53(1-2): 1-20.

Jung, J., & Shiller, R. J. (2006). Samuelson's dictum and the stock market, Economic

Inquiry, 43(2): 221-228.

Kahn, M. N. (2006). Technical Analysis: Plain and Simple (2nd ed.). New Jersey: Prentice

Hall Financial Times. Kahneman, D., & Tversky, A. (1979). Prospect theory:Malaya An analysis of decision under risk, Econometrica, 47(2): 263-292.

Kaliva, K., & Koskinen, L. (2008). Stock marketof bubbles, inflation and investment-risk,

International Review of Financial Analysis, 17: 592-603.

Kaminsky, G., Lizondo, S., & Reinhart, C.M. (1998). Leading Indicators of Currency

Crises, IMF Staff Papers, 45(1): 1-48.

Kanas, A. (2005). Nonlinearity in the stock price-dividend relation, Journal of

International Money and Finance, 24(4): 583-606.

Kaneko,University K., & Tsuda, I. (2001). Complex systems: chaos and beyond. New York: Spinger.

Kapopoulos, P., & Siokis, F. (2005). Stock market crashes and dynamics of aftershocks,

Economics Letters, 89: 48-54.

324

Karpavičius, S. (2014). Dividends: Relevance, rigidity, and signaling, Journal of

Corporate Finance, 25: 289-312.

Kauffman, S. (1993). The Origins of Order: Self-Organization and Selection in Evolution.

New York: Oxford University Press.

Keilis-Borok, V. I., Soloviev, A. A., Allègre, C. B., Sobolevskii, A. N., & Intriligator, M.

D. (2005). Patterns of macroeconomic indicators preceding the unemployment rise

in Western Europe and the USA, Pattern Recognition, 38: 423-435.

Keilis-Brook, V., Soloviev, A. A., Allègre, C. B., Sobolevskii, A. N., & Intriligator, M. D.

(2000). Pre-recession pattern of six economic indicators in the USA, Journal of Forecasting, 19: 65-80. Malaya Keim, D. B. (2008). Financial market anomalies.of In Durlauf, S. N., & Blume, L. E. (Eds.). The New Palgrave Dictionary of Economics. Retrieved on 17 October, 2015, from

the New Palgrave Dictionary of Economics Online:

http://www.dictionaryofeconomics.com/article?id=pde2008_F000308

Keim, D. B., & Stambaugh, R.F. (1986). Predicting returns in the stock and bond markets,

Journal of Financial Economics, 17: 357-390.

Keimling, N. (2016). Predicting Stock Market Returns Using the Shiller CAPE - An

UniversityImprovement Towards Traditional Value Indicators? Retrieved on 4 June, 2016,

from the Social Science Electronic Publishing, Inc (SSRN): http://

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2736423

325

Kendall, M. G., & Hill, A. B. (1953). The analysis of economic time-series-Part I: Prices

Journal of the Royal Statistical Society. Series A (General), 116(1): 11-34.

Kenny, T. (n.d.) What are Treasury Bills, Treasury Notes, and Treasury Bonds?

(About.com: US Economy). Retrieved on 19 February, 2013, from:

http://bonds.about.com/od/governmentandagencybonds/a/The-Differences-

Between-Treasury-Bills-Notes-And-Bonds.htm

Kettel, B. (2001). Economics for Financial Markets. Oxford: Butterworth-Heinemann.

Keynes, J. M. (1936). The General Theory of Employment Interest and Money. Retrieved

on 13 March, 2012, from: http://cas.umkc.edu/economics/people/facultyPages/kregel/courses/econ645/WinteMalaya r2011/GeneralTheory.pdf of Kheidon, A. W. (1992). Arbitrage, nontrading, and stale prices: October 1987, The

Journal of Business, 65(4): 483-507.

Kim, C-J., & Kim, M-J. (1996). Transient fads and the crash of '87, Journal of Applied

Econometrics, 11(1): 41-58.

Kim, H., & Han, I. (2001). Extracting trading rules from stock market data using rough Universitysets, Expert Systems, 18(4): 194-202. Kim, J. H., Shamsuddin, A & Lim, K-P. (2011). Stock return predictability and the

adaptive market hypothesis: Evidence from century-long U.S. data, Journal of

Empirical Finance, 18: 868-879.

326

Kindleberger, C. P., & Aliber, R. Z. (2005). Manias, Panics, and Crashes: A History of

Financial Crises. New Jersey: John Wiley & Sons.

Kirmana, A., & Teyssiere, G. (2005). Testing for bubbles and change points, Journal of

Economic Dynamics & Control, 29: 765-799.

Kleidon & Whaley (1992). One markets? Stocks, futures and options during October 1987,

The Journal of Finance, (Papers and Proceedings of the Fifty-Second Annual

Meeting of the American Finance Association, New Orleans, Louisiana January 3-

5, 1992) 47(3): 851-877.

Klein, P. J., & Iammartino, B. R. (2009). Getting Started in Security Analysis (2nd ed.). New Jersey: John Wiley & Sons, Inc. Malaya Klien, M. (2001). The stock market crash of of1929: A review article, The Business History Review, 75(2): 325-351.

Koehler, A. B., & Murphee, E. S. (1988). A Comparison of the Akaike and Schwarz

Criteria for selecting model order, Journal of the Royal Statistical Society. Series C

(Applied Statistics), 37(2): 187-195.

Koenig, E. F. (2002). Using the Purchasing Managers’ Index to assess the economy’s

strength and the likely direction of monetary policy, Federal Reserve Bank of

UniversityDallas Economic and Financial Policy Review, 1(6): 1-14.

327

Koijen, R. S. J., Lustig, H. N. & van Nieuwerburgh, S. (2010). The cross-section and

time-series of stock and bond returns. NBER Working Paper, w15688. Retrieved

on 21 October, 2016, from The National Bureau of Economic Research (NBER):

http://www.nber.org/papers/w15688.pdf

Kothari, S. P., & Shanken, J. (1997). Book-to-market, dividend yield, and expected market

returns: A time series analysis, Journal of Financial Economics, 44: 169-203.

Kothari, S. P., Lewellen, J., & Warner, J. B. (2006). Stock returns, aggregate earnings

surprises, and behavioral finance, Journal of Financial Economics, 79(3): 537-568.

Krugman (2009). The Return of Depression Economics and the Crisis of 2008. New York: W. W. Norton & Company. Malaya Kuan, M.K., (2002). Lecture on Markov Switchingof Model . Retrieved on 29 February, 2012, from: idv.sinica.edu.tw/ckuan/pdf/lecmarkn.pdf

Kumar, M., Moorthy, U., & Perraudin, W. R. M. (2002). Predicting Emerging Market

Currency Crashes. IMF Working Paper, 02/7: 1-38.

Kuo, N-T., & Lee, C-F. (2013). Effects of dividend tax and signaling on firm valuation:

Evidence from taxable stock dividend announcements, Pacific-Basin Finance UniversityJournal, 25: 157-180. Kurz-Kim, J. R. (2012). Early warning indicator for financial crashes using the log

periodic power law, Applied Economics Letters, 19(15): 1465-1469.

Laffont, J-J, & Maskin, E. S. (1990). The efficient market hypothesis and insider trading

on the stock market, Journal of Political Economy, 98(1): 70-93.

328

Lahiri, K., & Monokroussos, G. (2013). Nowcasting US GDP: The role of ISM business

surveys, International Journal of Forecasting, 29(4): 644-658.

Lakonishok, J., Shleifer, A & Vishny, R. W. (1994). Contrarian investment, extrapolation

and risk, The Journal of Finance, 49(5): 1541-1578.

Laloux, L., Potters, M., Cont, R., Aguilar, J. -P., & Bouchaud. J. -P. (1999). Are financial

crashes predictable? Europhysics Letters, 45: 1-5.

Laopodis, N. T. (2013). Monetary policy and stock market dynamics across monetary

regimes, Journal of International Money and Finance, 33: 381-406. Layton, A. P. & Daniel, R. S. (2007). Business cycle dynamicsMalaya with duration dependence and leading indicators, Journal of Macroeconomics, 29: 855-875.

Lee, B. S. (2010). Stock returns and inflationof revisited: an evaluation of the inflation illusion hypothesis, Journal of Banking and Finance, 34: 1257-1273.

Leigh, W., & Purvis, R. (2008). Implementation and validation of an opportunistic stock

market timing heuristic: One-day share volume spike as buy signal, Expert Systems

with Applications, 35: 1628-1637.

Leigh, W., Modani, N., & Hightower, R. (2004). A computational implementation of Universitystock charting: abrupt volume increase as signal for movement in New York Stock Exchange Composite Index, Decision Support Systems, 37: 515-530.

Leigh, W., Modani, N., Purvis, R., & Roberts, T. (2002). Stock market trading rule

discovery using technical charting heuristics, Expert Systems with Application, 23,

155-159.

329

Leigh, W., Paz, N., & Purvis, R. (2002). Market timing: A test of charting heuristics,

Economics Letters, 77: 55-63.

Leoni, P. L. (2009). Market crashes, speculation and learning in financial markets,

Economic Theory, 39(2): 217-229.

LeRoy, S. F. (1989). Efficient capital markets and Martingales, Journal of Economic

Literature, 27: 1583-1621.

LeRoy, S. F., & Porter. R. D. (1981). The present-value relation: Tests based on implied

variance bounds. Econometrica, 49 (May 1981), 555-574. Lettau, M., & Ludvigson, S. (2001). Consumption, aggregateMalaya wealth, and expected stock returns, Journal of Finance, 56: 815-849.

Levy, M. (2008). Stock market crashes as socialof phase transitions, Journal of Economic Dynamics & Control, 32: 137-155.

Lewellen, J. (2004). Predicting returns with financial ratios, Journal of Financial

Economics, 74: 209-235.

Liberatore, V. (2011a). Computational LPPL fit to financial bubbles, arXiv (e-print),

1003.2920v2: 1-12.

Liberatore,University V. (2011b). Financial LPPL bubbles with mean-reverting noise in the frequency domain, arXiv (e-prints), 1009.4835v2: 1-21.

330

Lillo, F., & Mantegna, R. N. (2000). Symmetry alteration of ensemble return distribution

in crash and rally days of financial markets, The European Physical Journal B, 15:

603-606.

Lillo, F., & Mantegna, R. N. (2001). Empirical properties of the variety of a financial

portfolio and the single-index model, The European Physical Journal B, 20: 503-

509.

Lim, K-P & Brookes, R. (2010). The evolution of stock market efficiency over time: A

survey of the empirical literature, Journal of Economic Surveys, 25(1): 69-108.

Lin, L., Ren, R. E., & Sornette, D. (2014). The volatility-confined LPPL model: A consistent model of ‘explosive’ financial bubblesMalaya with mean-reverting residuals, International Review of Financial Analysisof, 33: 210-225. Ljungqvist, L. (2008). Lucas critique. In Durlauf, S. N., & Blume, L. E. (Eds.). The New

Palgrave Dictionary of Economics. Retrieved on 17 October, 2015, from the New

Palgrave Dictionary of Economics Online:

www.dictionaryofeconomics.com/article?id=pde2008_L000159

Lo, A. W. (2004a). The adaptive markets hypothesis: Market efficiency from an

evolutionary perspective Journal of Portfolio Management, 30: 15-29.

Lo, University A. W. (2004b). Bubble, rubble, finance in trouble? Journal of Psychology and

Financial Markets, 3: 76-86.

Lo, A. W. (2005). Reconciling efficient markets with behavioral finance: The adaptive

markets hypothesis, Journal of Investment Consulting. 7(2): 21-44.

331

Lo, A., Repin, D., & Steenbarger, B. (2005). Fear and greed in financial markets: A

clinical study of day-traders, The American Economic Review, 95: 352-359.

Lo, A.W., & MacKinlay, A.C. (2001). A Non-Random Walk Down Wall Street. Princeton,

NJ: Princeton University Press.

Loewenstein, M., & Willard, G. (2000). Rational equilibrium asset pricing bubbles in

continuous trading models, Journal of Economic Theory, 91: 17-58.

Loewenstein, M., & Willard, G. (2006). The limits of investor behavior, Journal of

Finance, 61(1): 231-258. Loisel, O. (2009). Bubble-free policy feedback rules,Malaya Journal of Economic Theory, 144: 1521-1559.

Lucas, R. E. (1978). Asset Prices in an Exchangeof Economy , Econometrica, 46: 1429-1445.

Lunde, A., & Timmermann, A. (2004). Duration dependence in stock prices: An analysis

of bull and bear markets, Journal of Business & Economic Statistics, 22(3): 253-

273.

Lux, T. (1995). Herd behaviour, bubbles and crashes, The Economic Journal, 105(431):

881-896.

Lux,University T., & Sornette, D. (2002). On rational bubbles and fat tails, Journal of Money, Credit and Banking, 34(3): 589-610.

Madrigal, V., & Scheinkman, J. A. (1997). Price crashes, information aggregation and

market-making, Journal of Economic Theory, 73: 16-63.

332

Maheu, J. M., & McCurdy, T. H. (2000). Identifying bull and bear markets in stock

returns, Journal of Business & Economic Statistics, 18: 100-112.

Maheu, J. M., McCurdy, T. H., & Song, Y. (2009). Extracting bull and bear markets from

stock returns, University of Toronto, Department of Economics Working Papers,

369. Retrieved on 24 June, 2011, from:

http://www.rcfea.org/papers/2009/bayesian/maheu.pdf

Malik, F., Ewing, B. T., Kruse, J. B., & Lynch, G. J. (2009). Modeling the time-varying

volatility of the paper–bill spread, Journal of Economics and Business, 61: 404-

414.

Malkiel, B. G. (2003). The efficient market hypothesisMalaya and its critics, The Journal of Economic Perspectives, 17(1): 59-82.of Malkiel, B. G. (2007). A Random Walk Down Wall Street. New York: W. W. Norton &

Company, Inc.

Malonis, J. A. (Ed.). (2000). Encyclopedia of Business. (2nd ed). Michigan: Gale Group.

Mandelbrot, B. (1966). Forecast of future prices, unbiased markets, and martingale models,

Journal of Business, 39: 242-255. Mandelbrot,University B. (1968). Some aspects of the random-walk model of stock market prices: Comment, International Economic Review, 9: 258-259.

Mandelbrot, B. (1997a). Fractal and Scaling in Finance: Discontinuity, Concentration,

Risk. New York: Springer-Verlag.

333

Mandelbrot, B. (1997b). Three fractal models in finance: Discontinuity, concentration,

risk, Economic Notes (Banca Monte dei Paschi di Siena SpA), 26(2): 197-212.

Mandelbrot, B. (1999a). A fractal walk down Wall Street, Scientific American February:

70-73.

Mandelbrot, B. (1999b). Survey of multifractality in finance, Cowles Foundation Paper

No. 1238. Retrieved on 13 March, 2012, from the Cowles Foundation for Research

in Economics, Yale University: http://dido.wss.yale.edu/P/cd/d12a/d1238.pdf

Mandelbrot, B. (2000). Multifractal structure of financial prices and its implications.

Cowles Foundation Paper No. 991. Cowles Foundation for Research in Economics, Yale University. Retrieved on Malaya 13 March, 2012, from: http://dido.wss.yale.edu/P/cp/p09b/p0991.pdfof Mandelbrot, B. (2001a). Scaling in financial prices, I: Tail and dependence, Quantitative

Finance, 1: 113-123.

Mandelbrot, B. (2001b). Scaling in financial prices, II: Multifractals and the star equation,

Quantitative Finance, 1: 124-130.

Mandelbrot, B. (2001c). Scaling in financial prices, III: Cartoon Brownian motions in Universitymultifractal time, Quantitative Finance, 1: 427-440. Mandelbrot, B. (2001e). Scaling in financial prices, IV: Multifractal concentration,

Quantitative Finance, 1: 641-649.

Mandelbrot, B. (2004). Fractal and Chaos: The Mandelbrot Set and Beyond. New York:

Springer-Verlag.

334

Mandelbrot, B., & Hudson, R.L. (2008). The (Mis)Behaviour of Markets: A Fractal View

of Risk, Ruin and Reward (3rd ed.). London: Profile Books.

Mantegna, R. N. (1999). Hierarchical structure in financial markets, The European

Physical Journal B, 11: 193-197.

Mantegna, R. N., & Stanley, H. E. (1997). Econophysics: Scaling and its breakdown in

finance, Journal of Statistical Physics, 89(1/2): 469-479.

Mantegna, R. N., & Stanley, H. E. (2000). Introduction to Econophysics: Correlations &

Complexity in Finance. Cambridge: Cambridge University Press. Marcellino, M. (2006). Leading Indicators in Arrow,Malaya K., & Intriligator, M. (Eds.), Handbooks in Economics (pp. 879-960). Amsterdam: North-Holland Publishing Company. of Marsh, T. A., & Merton R. C. (1986). Dividend variability and variance bounds tests for

the rationality of stock market prices, American Economic Review, 76: 483-498.

Masulis, R. W., & Ng, V. K. (1995). Overnight and daytime stock-return dynamics on the

London Stock Exchange: The impacts of Big Bang and the 1987 stock-market

crash, Journal of Business & Economic Statistics, 13(4): 365-378. Matsushita,University R., da Silva, S., Figueiredo, A., & Gleria, I. (2006). Log-periodic crashes revisited, Physica A, 364: 331-335.

McCown, J. R. (2007). The Role of Stock Prices in Business Cycles. Retrieved on 11

September, 2010, from the Social Science Electronic Publishing, Inc (SSRN):

http://ssrn.com/abstract=961872 or http://dx.doi.org/10.2139/ssrn.961872

335

McGratten, E. R., & Prescott, E. C. (2004). The 1929 stock market: Irving Fisher was

right, International Economic Review, 45(4): 991-1009.

McGuckin, R. H., Ozyildirim, A., & Zarnowitz, V. (2007). A more timely and useful

index of leading indicators, Journal of Business & Economic Statistics, 25(1): 110-

120.

McMillan, D. G. (2001). Nonlinear predictability of stock market returns: evidence from

nonparametric and threshold models, International Review of Economics and

Finance, 10: 353-368.

McQueen, G., & Roley, V.V. (1993). Stock prices, news, and business conditions, Review of Financial Studies, 6: 683-707. Malaya McQueen, G., & Thorley, S. (1994). Bubbles,of stock returns, and duration dependence, The Journal of Financial and Quantitative Analysis, 29(3): 379-401.

Menzly, L., Santos, T., & Veronesi, P. (2004). Understanding predictability, Journal of

Political Economy, 112: 1-47.

Merriam-Webster’s Collegiate Dictionary (10th ed.). (2002). Springfield, Massachusetts:

Merriam Webster. Meyer,University B. (1999). Intertemporal Asset Pricing: Evidence from Germany. Gernamy: Physica Verlag Heidelberg

Meyers, R. A. (Ed.). (2011). Complex Systems in Finance and Econometrics. New York:

Springer.

336

Milonas, N. T., Travlos, N. G., Xiao, J. Z., & Tan, C. (2006). The ex-dividend day stock

price behavior in the Chinese stock market, Pacific-Basin Finance Journal, 14(2):

155-174.

Minsky, H. P. (1986). Stabilizing the Unstable Economy. New Haven, Connecticut: Yale

University Press.

Minsky, H. P. (1992). The Financial Instability Hypothesis, Jerome Levy Economics

Institute Working Paper, 74. Retrieved on 13 March, 2012, from:

http://www.levyinstitute.org/pubs/wp74.pdf

Munira, S., & Muradoglu, Y. G. (2008). Momentum in Credit Rated Stocks: Is Momentum Investors' Reaction to Uncertainty Across BusinessMalaya Cycle? Retrieved 6 March, 2010, from the Social Science Electronic Publishing, Inc (SSRN):

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1130792of

NAPM (National Association of Purchasing Managers). (n.d.). National Association of

Purchasing Managers (NAPM). Retrieved 17 March, 2014, from:

http://pages.stern.nyu.edu/~nroubini/bci/napm.htm

Nasseh, A., & Strauss, J. (2004). Stock prices and the dividend discount model: did their

relation break down in the 1990s? The Quarterly Review of Economics and UniversityFinance, 44(2): 191-207.

NBC News.com. (n.d.). 11 historic bear markets: From the Great Depression to the Great

Recession. Retrieved on 13 September, 2016, from:

337

http://www.nbcnews.com/id/37740147/ns/business-

stocks_and_economy/t/historic-bear-markets/#.V9gKu1R97IU

NBER (The National Bureau of Economic Research). (n.d.). Information on Recessions

and Recoveries, the NBER Business Cycle Dating Committee, and related topics.

Retrieved on 16 September, 2010, from: http://www.nber.org/cycles/main.html

NECSI (New England Complex Systems Institute). (n.d.). About Complex Systems.

Retrieved on 10 September, 2010, from: http://www.necsi.edu/guide/study.html

Nelson, C. (1976). Inflation and rates of return on common stocks, Journal of Finance, 31: 471-483. Malaya New York Stock and Commodity Exchange. (n.d.). Standards & Poor’s 500 Index History. Retrieved on 16 September, 2010,of from: http://www.nyse.tv/s-and-p-500- history.htm

Nicholas, T. (2008). Does innovation cause stock market runups? Evidence from the great

crash, The American Economic Review, 98(4): 1370-1396.

Nobleprize.org. (n.d.). Friedrich August von Hayek. Retrieved on 12 September, 2010,

from: http://nobelprize.org/nobel_prizes/economics/laureates/1974/hayek.html Nyberg,University H. (2011). Forecasting the direction of the US stock market with dynamic binary probit models, International Journal of Forecasting, 27: 561-578.

Odean, T. (1999). Do investors trade too much? American Economic Review, 89: 1279-

1298.

338

Ofek, E., & Richardson, M. (2003). Dotcom mania: The rise and fall of internet stock

prices, Journal of Finance 58: 1113-1137.

Online Data Robert Shiller. (n.d.). Retrieved on 31 December, 2014, from:

http://www.econ.yale.edu/~shiller/data.htm

Onnela, J-P. (2006). Complex Networks in the Study of Financial and Social System. (PhD

dissertation). Helsinki: Helsinki University of Technology Laboratory of

Computational Engineering Publications.

Onnela, J-P., Chakraborti, A., Kaski, K., & Kertesz, J. (2002). Dynamic asset trees and portfolio analysis, The European Physical JournalMalaya B, 30: 285-288. Onnela, J-P., Chakraborti, A., Kaski, K., & Kertesz, J. (2003). Dynamic asset trees and Black Monday, Physica A, 324: 247-252.of Onnela, J-P., Chakraborti, A., Kaski, K., J. Kertész & Kanto, A. (2003). Asset trees and

asset graphs in financial markets, Physica Scripta, 106: 48-54.

Oxman, J. (2012). Price inflation and stock returns, Economics Letters, 116: 385-388.

Pagan, A. R., & Sossounov, A. S. (2003). A simple framework for analysing bull and bear

markets, Journal of Applied Econometrics, 18: 23-46.

Park,University C. (2010). When does the dividend-price ratio predict stock returns? Journal of Empirical Finance, 17: 81-101.

Pass, C., Lowes, B., & Davies, L. (2005). Collins Dictionary of Economics (4th ed.).

Glasgow: HyperCollins Publishers.

339

Park, C-H., & Irwin, S. H. (2007). What do we know about the profitability of technical

analysis? Journal of Economics Surveys, 21(4): 786-826.

Pastor, L., & Veronesi, P. (2006). Was there a nasdaq bubble in the late 1990s? Journal of

Financial Economics, 81(1): 61-100.

Peláez, R. F. (2003a). A new index outperforms the purchasing managers' index,

Quarterly Journal of Business and Economics, 42(1/2): 41-55.

Peláez, R. F. (2003b). A reassessment of the Purchasing Managers’ Index, Business

Economics, October 2003: 35-41. Retrieved on 9 March, 2014, from: http://nabe- web.com/publib/be/0304/030435.pdf Malaya Pele, D. T. (2012). An LPPL algorithm for estimating the critical time of a stock market bubble, Journal of Social and Economicof Statistics , 2(1): 14-22. Pele, D. T., & Mazurencu-Marinescu, M. (2012). Modelling stock markets crashes: The

case of Bucharest Stock Exchange, Procedia - Social and Behavioral Sciences, 58:

533-542.

Perez-Quiros, G., & Timmermann, A. (1998). Variations in the mean and volatility of

stock returns around turning points of the business cycle, in Knight, J., & Satchell,

S. (Eds.). Forecasting Volatility in the Financial Markets (pp. 287–306). Oxford:

UniversityButterworth-Heinemann.

Perez-Quiros, G., & Timmermann, A. (2000). Firm size and cyclical variations in stock

returns, The Journal of Finance, 55(3): 1229-1262.

340

Perez-Quiros, G., & Timmermann, A. (2001). Business cycle asymmetries in stock returns:

evidence from higher order moments and conditional densities, Journal of

Econometrics, 103: 259-306.

Persaud, A. D. (2007). The Crash of 2007-08, Economic and Political Weekly, 42(34):

3444-3446.

Pesaran, M. H., & Timmermann, A. (2002). Market timing and return prediction under

model instability, Journal of Empirical Finance, 9: 495-510.

Pesaran, M. H., & Timmermann, A. (1995). Predictability of stock returns: Robustness and economic significance, The Journal of FinanceMalaya, 4: 1201-1228. Peters, E. E. (1994). Fractal Market Analysis: Applying Chaos Theory to Investment and Economics. New York: Wiley. of Phillips, P. C. B., Wu, Y., & Yu, J. (2011). Explosive behavior in the 1990s Nasdaq: when

did exuberance escalate asset values? International Economic Review, 52: 201-

226.

Pierdzioch, C., Dopke, J., & Hartmann, D. (2008). Forecasting stock market volatility with

macroeconomic variables in real time, Journal of Economics and Business, 60: University256-276. Pontiff, J., & Schall, L. (1998). Book-to-market ratios as predictors of market returns,

Journal of Financial Economics, 49: 141-160.

341

Popper, K. R. (1959). The Logic of Scientific Discovery. New York: Basic Books

Retrieved 12 September, 2010, from:

http://www.questia.com/PM.qst?a=o&d=9219121

Potter, M., & Bouchaud, J-P. (2003). More statistical properties of order books and price

impact, Physica A, 324: 133-140.

Pring, M. J. (2006). The Investor's Guide to Active Asset Allocation: Using Technical

Analysis and ETFs to Trade the Markets. New York: McGraw-Hill.

Pring, M. J. (2002). Technical Analysis Explained (4th ed.). New York: McGraw-Hill. Qi, Min. (2001). Predicting recessions with leading indicatorsMalaya via neural network models, International Journal of Forecasting, 17: 383-401.

Racine, J. (2001). On the nonlinear predictabilityof of stock returns using financial and economic variables, Journal of Business & Economic Statistics, 19(3): 380-382.

Raines, J. P., & Leathers, C. G. (1994). The new speculative stock market: Why the weak

immunizing effect of the 1987 crash? Journal of Economic Issues, 28(3): 733-753.

Rapach, D. E., & Wohar, M. (2006). In-sample vs. out-of-sample tests of stock return

predictability in the context of data mining, Journal of Empirical Finance, 13(2): University231-247. Rapach, D. E., Wohar M. E., & Rangvid, J. (2005). Macro variables and international

stock return predictability, International Journal of Forecasting, 21: 137-166.

342

Rappoport, P., & White, E. N. (1993). Was there a bubble in the 1929 stock market?, The

Journal of Economic History, 53 (3): 549-574.

Rappoport, P., & White, E. N. (1994). Was the crash of 1929 expected?, American

Economic Review, 84: 271-281.

Reilly, F. K., & Brown, K. C. (2011). Investment Analysis and Portfolio Management (10th

ed.). Ohio: South-Western Cengage Learning.

Reinganum, M. R. (1981). Misspecification of capital asset pricing: empirical anomalies

based on earnings' yields and market values, Journal of Financial Economics, 9: 19-46. Malaya Resnick, B.G., & Shoesmith, G.L. (2002). Using the yield curve to time the stock market, Financial Analysts Journal, 58(3): 82-90.of Rhea, R. (1993). The Dow Theory. Rochester: Fraser Publishing Library.

Risso, W. A. (2008). The informational efficiency and the financial crashes, Research in

International Business and Finance 22: 396-408.

Ritter, J., & Warr, R. (2002). The decline of inflation and the bull market of 1982–1999,

Journal of Financial and Quantitative Analysis, 37: 29-61.

Roberts,University H. (1959). Stock-market patterns and financial analysis: Methodological suggestions, Journal of Finance, 14(1): 1-10.

343

Roberts, H. (1967). Statistical versus Clinical Prediction of the Stock Market

(Unpublished manuscript), Center for Research in Security Prices, University of

Chicago.

Robertson, D., & Wright, S. (2006). Dividends, total cashflow to shareholders and

predictive return regressions, Review of Economics and Statistics, 88: 91-99.

Robinson, L. R. (1930). The role of investment companies in the security markets of 1929,

Journal of the American Statistical Association, (Supplement: Proceedings of the

American Statistical Association) 25(169): 83-87.

Roll, R. (1988). The international crash of October 1987, Financial Analysts Journal, 44(5): 19-35. Malaya Romer, C. D. (1990). The great crash andof the onset of the Great Depression, The Quarterly Journal of Economics, 105(3): 597-624.

Rozeff, M. (1984). Dividend yields are equity risk premium, Journal of Portfolio

Management, 11:68-75.

Rubinstein, M. (1988). Portfolio insurance and the market crash, Financial Analysts

Journal, 44(1): 38-47. Rubinstein,University M. (2001). Rational markets: Yes or no? The affirmative case, Financial Analysts Journal, 57(3): 15-29.

Samuelson. P. A. (1965). Proof that properly anticipated prices fluctuate randomly,

Industrial Management Review, 6: 41-49.

344

Samuelson. P. A. (1973a). Mathematics of speculative price, SIAM Review, 15(1): 1-42.

Samuelson. P. A. (1973b). Proof that properly discounted present values of assets vibrate

randomly, The RAND Journal of Economics, 4: 369-374.

Sandroni, A. (1998). Learning, rare events, and recurrent market crashes in frictionless

economies without intrinsic uncertainty, Journal of Economic Theory, 82(1): 1-18

Santoni, G. (1987). The great bull markets, 1924-1929 and 1982-1987: Speculative

bubbles or economic fundamentals? Federal Reserve Bank of St. Louis Review, 69:

16-29. Santoni, G., & Gerald P. D. Jr. (1990). Bubbles vs. fundamentals:Malaya New evidence from the great bull markets, in White, E. N. (Ed.). Crises and Panics: The Lessons of History (pp. 188-210). Illinios: Homewood.of Santos, M., & Woodford, M. (1997). Rational asset pricing bubbles, Econometrica, 65:

19-57.

Sarno, L., & Valente, G. (2005). Modelling and forecasting stock returns: exploiting the

futures market, regime shifts and international spillovers, Journal of Applied

Econometrics, 20(3): 345-376. Savoiu,University G. & Iorga-Siman, I. (2008). Some relevant econophysics’ moments of history, definitions, methods, models and new trends, Romanian Economic and Business

Review, 3(3): 29-41.

Scharfstein, D. S., & Stein, J. C. (1990). Herd behavior and investment, American

Economic Review, 80: 465-479.

345

Scheinkman, J., & Xiong, W. (2003). Overconfidence and speculative bubbles, Journal of

Political Economy, 111: 1183-1219.

Schwert, G. W. (1983). Size and stock returns, and other empirical regularities, Journal of

Financial Economics, 12: 3-12.

Schwert, G. W. (1989). Why does stock market volatility change over time?, Journal of

Finance, 44(5): 1115-1153.

Schwert, G. W. (1990). Stock volatility and the crash of '87, The Review of Financial

Studies, 3(1): 103-106. Schwert, G. W. (2003). Anomalies and market efficiency,Malaya in Constantinides, G., Stulz, R. M., & Harris, M. (Eds.). Handbook of the Economics of Finance: Financial Markets and Asset Pricing Volume 1Bof: 939-974. Amsterdam: Elsevier. Seguin, P. J., & Jarell, G. A. (1993). The irrelevance of margin: Evidence from the crash

of '87, The Journal of Finance, 48(4): 1457-1473.

Seip, K. L., & McNown, R. (2007). The timing and accuracy of leading and lagging

business cycle indicators: A new approach, International Journal of Forecasting,

23: 277-287. SEP.University (Stanford Encyclopedia of Philosophy). (n.d.). Karl Popper. Retrieved 11 September, 2010, from: http://www.plato.stanford.edu/archives/sum2009/entries/popper/

Sewell, M. (2010). Behavioural Finance. Retrieved on 13 March, 2012, from:

http://www.behaviouralfinance.net/behavioural-finance.pdf

346

Seyhun, H. N. (1990). Overreaction or fundamentals: Some lessons from insiders'

response to the market crash of 1987, The Journal of Finance, 45(5): 1363-1388.

SFI (Santa Fe Institute). (n.d.) History of the Santa Fe Institute. Retrieved on 15

September, 2010, from: http://www.santafe.edu/about/history/

Sharpe, W. F. (1975). Likely gains from stock market timing. Financial Analysts Journal,

31: 60-90.

Shiller, R .J. (1981), Do stock prices move too much to be justified by subsequent changes

in dividends? American Economic Review, 75: 421-436. Shiller, R. J. (1984). Stock prices and social dynamics,Malaya Brookings Papers on Economic Activity, 2: 457-498.

Shiller, R. J. (1988). Portfolio Insurance andof other investor fashions as factors in the 1987 stock market crash, NBER Macroeconomics Annual, 3: 287-297.

Shiller, R. J. (1990). Speculative prices and popular models, The Journal of Economic

Perspectives, 4(2): 55-65.

Shiller, R. J. (2000). Measuring bubble expectations and investor confidence, Journal of

Psychology and Financial Markets, 1(1): 49-60.

Shiller,University R. J. (2002). Bubbles, human judgement, and expert opinion. Financial Analysts Journal, 58(3): 18-26.

Shiller, R. J. (2003). From efficient market theory to behavioral finance, Journal of

Economic Perspectives, 17 (1): 83-104.

347

Shiller, R. J. (2005). Irrational Exuberance (2nd ed.). New York: Double Day.

Shiller, R. J. (2008). Historic turning points in real estate, Eastern Economic Journal, 34:

1-13.

Shim, J. K., & Siegel, J. G. (2001). Dictionary of International Investment Terms. New

York: Barron’s Educational Series Inc.

Shivakumar, L. (2007). Aggregate earnings, stock market returns and macroeconomic

activity: A discussion of ‘does earnings guidance affect market returns? The nature

and information content of aggregate earnings guidance’, Journal of Accounting

and Economics, 44(1-2): 64-73.

th Malaya Siegel, J. J. (2008). Stocks for the Long Run (4 ed.). New York: Mcgraw-Hill.

Siegel, J. J. (2016). The Shiller Cape ratio: Aof new look, Financial Analysts Journal, 72(3): 41-50.

Simon, H. A. (1972). Theories of bounded rationality, in McGuire, C. B., & Radner, R.

(Eds.), Decision and Organization (pp. 161-176). Amsterdam: North-Holland

Publishing Company.

Sinai, A. (1988). The crash of '87 and the economy of '88, Challenge, 31(1): 11-21.

Sinai,University Y. G. (2010). Chaos Theory yesterday, today and tomorrow, Journal of Statistical Physics, 139: 2-7.

Singal, V. (2004). Beyond the Random Walk: A Guide to Stock Market Anomalies and

Low-Risk Investing. New York : Oxford University Press.

348

Siokis, F. M. (2012). Stock market dynamics: Before and after stock market crashes,

Physica A, 391: 1315-1322.

Siokis, F. M. (2013). Multifractal analysis of stock exchange crashes, Physica A, 392:

1164-1171.

Sirkin, G. (1975). The stock market of 1929 revisited: A note, The Business History

Review, 49(2): 223-231.

Smith, P. N., Sorensen, S., & Wickens, M. R. (2006). The asymmetric effect of the

business cycle on the relation between stock market returns and their volatility,

University of York, Discussion Paper, 2006/03. Retrieved on 30 April, 2010, from: http://dx.doi.org/10.2139/ssrn.676383 Malaya Smith, V. L., Suchannek, G. L., & William,of A.W. (1988). Bubbles, crashes, and endogenous expectations in experimental spot asset markets, Econometrica, 56(5):

1119-1151.

Smith, V. L., Van Boining, M., & Wellford, C. P. (2000). Dividend timing and behavior in

laboratory asset markets, Economic Theory, (Symposium: Experimental Markets)

16(3): 567-583.

Sornette, D. (1998). Discrete scale invariance and complex dimensions, Physics Reports

University297(5): 239-270.

Sornette, D. (2003). Critical market crashes, Physics Reports, 378: 1-98.

Sornette, D. (2003). Why Stock Markets Crash: Critical Events in Complex Financial

Systems. Princeton: Princeton University Press.

349

Sornette, D., & Cauwels, P. (2014). The illusion of the perpetual money machine, Risks, 2:

103-131.

Sornette, D., & Johansen, A. (1997). Large financial crashes, Physica A, 245: 411-422.

Sornette, D., & Johansen, A. (1998). A hierarchical model of financial crashes, Physica A,

261: 581-598.

Sornette, D., & Johansen, A. (2001). Significance of log-periodic precursors to financial

crashes, Quantatitive Finance, 1(4): 452-471.

Sornette, D., & Malevergne, Y. (2001). From rational bubbles to crashes, Physica A, 299: 40-59. Malaya Sornette, D., & Zhou, W-X. (2006). Predictability of large future changes in major financial indices, International Journalof of Forecasting , 22: 153-168.

Sornette, D., Johansen, A., & Bouchaud, J. -P. (1996). Stock market crashes, precursors

and replicas, Journal de Physique I, France, 6: 167-175.

Sornette, D., Woodard, R., Yan, W., & Zhou, W-X. (2013). Clarifications to questions and

criticisms on the Johansen-Ledoit-Sornette bubble model, Physica A, 392(19):

4417-4428.

Sperandeo,University V. (1990). Principles of Professional Speculation. New York: Wiley.

Standard & Poor’s. (2009) Standard & Poor’s 500 Guide 2009. New York: The McGraw-

Hill Companies, Inc.

350

Stanley, H. E., Amaral, L. A. N., Canning, D., Gopikrishnan, P., Lee, Y. & Liu, Y. (1999).

Econophysics: Can physicists contribute to the science of economics? Physica A,

269: 156-169.

Stephen Wolfram (About Stephen Wolfram). (n.d.). About Stephen Wofram. Retrieved on

10 September, 2010, from: http://www.stephenwolfram.com

Stiglitz, J. E. (1990). Symposium on bubbles, The Journal of Economic Perspectives, 4(2):

13-18.

Stiglitz, J. E. (1993). The role of the state in financial markets. World Bank Econ Rev, 7 (suppl 1): 19-52. Malaya Stiglitz, J. E. (2010). Freefall: America, Free Markets, and the Sinking of the World Economy. New York: W. W. Norton of& Company. Stock, J. H., & Watson, M. W. (1989). New indexes of coincident and leading economic

indicators, in Blanchard, O., & Fischer, S. (Eds.), NBER Macroeconomics Annual

(pp. 351-394). Cambridge, MA: MIT Press.

Stock, J. H., & Watson, M. W. (1993). A procedure for predicting recessions with leading

indicators: Econometric issues and recent experience, in Stock, J. H., & Watson, M.

W. (Eds.), Business Cycles, Indicators, and Forecasting (pp. 95-153). Chicago:

UniversityThe University of Chicago Press.

Stock, J. H., & Watson, M. W. (1999). Forecasting inflation, Journal of Monetary

Economics, 44(2): 293-335.

351

Stock, J. H., & Watson, M. W. (2002). Macroeconomic forecasting using diffusion

indexes, Journal of Business and Economic Statistics, 20(2): 147-162.

Stock, J. H., & Watson, M. W. (n.d). Stock-Watson Experimental Coincident, Leading and

Recession Indexes. Retrieved on 6 August, 2013, from the The National Bureau of

Economic Research (NBER): http://www.nber.org/xri/xri/

Summers, L. H. (1986). Does the stock market rationally reflect fundamental values?

Journal of Finance, 91: 591-601.

Szafarz, A. (2012). Financial crises in efficient markets: How fundamentalists fuel volatility, Journal of Banking & Finance, 36: 105-111.Malaya Taleb, N.N. (2007). The Black Swan. London: Penguin Books.

Tashman, L. J. (2000). Out-of-sample testsof of forecasting accuracy: An analysis and review, International Journal of Forecasting, 16(4): 437-450.

Thaler, R. H. (1992), The Winner's Curse: Paradoxes and Anomalies of Economic Life.

Princeton, New Jersey: Princeton University Press.

Thomakos, D. D., Wang, T., & Wu, J. (2007). Market timing and cap rotation,

Mathematical and Computer Modelling, 46: 278-291.

Thorbecke,University W. (1994). Trade deficit news, systematic risk and the crash of 1987, Eastern Economic Journal, 20(1): 97-106.

Timmermann, A., & Granger, C. W. J. (2004). Efficient market hypothesis and forecasting,

International Journal of Forecasting 20: 15-27.

352

Tirole, J. (1982). On the possibility of speculation under rational expectations,

Econometrica, 50(5): 1163-1182.

Tirole, J. (1985). Asset bubbles and overlapping generations, Econometrica, 53(6): 1499-

1528.

Tobago, M. (2011). Under-/over-valuation of the stock market and cyclically adjusted

earnings, International Finance, 14(1): 135-164.

Treynor, J. (1998). Bulls, bears, and market bubbles, Financial Analysts Journal, 54(2):

69-74. Trueman, B. (1994). Analyst forecasts and herding Malaya behavior. The Review of Financial Studies, 7(1): 97-124.

Tversky, A., & Kahneman, D. (1981). The of framing of decisions and the psychology of choice, Science, New Series, 211(4481): 453-458.

United State Census Bureau. (n.d.)(a). Housing Units Authorized by Building Permits.

Retrieved on 17 April, 2014, from the United State Census Bureau site:

http://quickfacts.census.gov/qfd/meta/long_BPS030210.htm

United States Bureau of Census. (1968). BCD; Business Cycle Developments. Michigan: UniversityUniversity of Michigan. United States Census Bureau (n.d.)(b). Composite Indexes of Leading, Coincident, and

Lagging Economic Indexes: 2000 to 2010. Retrieved on 8 May, 2011, from the

U.S. Census Bureau:

http://www.census.gov/compendia/statab/2012/tables/12s0787.pdf

353

University of Michigan (n.d.). Surveys of Consumers. Retrieved on 28 August, 2013, from:

http://www.sca.isr.umich.edu/fetchdoc.php?docid=24774

Valkanov, R. (2003). Long-horizon regressions: theoretical results and applications,

Journal of Financial Economics, 68: 201-232. van Bothmer, H.-C. G., & Meister, C. (2003). Predicting critical crashes? A new

restriction for the free variables, Physica A, 320: 539-547. van Norden, S. (1996). Regime switching as a test for exchange rate bubbles, Journal of

Applied Econometrics, 11(3): 219-251. Vandewalle, N., Ausloos, M., Boveroux, Ph., & Minguet,Malaya A. (1999). Visualizing the log periodic pattern before crashes, The European Physical Journal B, 9: 355-359.

Veldkamp, L. L. (2005). Slow-boom, suddenof crash, Journal of Economic Theory, 124: 230-257.

Vines, S. (2009). Market Panic: Wild Gyrations, Risks and Opportunities in Stock

Markets (2nd ed.). Singapore: John Wiley & Sons (Asia) Pte Ltd.

Vissing-Jorgensen, A. (2004). Perspectives on behavioral finance: Does irrationality

disappear with wealth? Evidence from expectations and actions, NBER UniversityMacroeconomics Annual 2003, 18: 139-208. Wang, J., Meric, G., Liu, Z., & Meric, I. (2009). Stock market crashes, firm characteristics,

and stock returns, Journal of Banking & Finance, 33: 1563-1574.

354

Wang, P., & Wen, Y (2009). Speculative bubbles and financial crisis, Federal Reserve

Bank of St. Louis Working Paper Series, 2009-029B.

Watanabe, K., Takayasu, H., & Takayasu, M. (2007a), Extracting the exponential

behaviors in the market data, Physica A, 382: 336-339.

Watanabe, K., Takayasu, H., & Takayasu, M (2007b) A mathematical definition of the

financial bubbles and crashes, Physica A, 383: 120-124.

Welch, I. (2000). Herding among security analysts, Journal of Financial Economics, 58(3):

369-396. Wessels, W. J. (2012). Economics (5th ed.). New York:Malaya Barron's Educational Series. West, K. D. (1987). A specification test for speculative bubbles, The Quarterly Journal of Economics, 102(3): 553-580. of

West, K. D. (1988). Bubbles, fads and stock price volatility tests: A partial evaluation.

Journal of Finance, 43: 636-661.

White, E. H. (1990a). The stock market boom and crash of 1929 revisited. Journal of

Economic Perspectives, 4(2): 67-83.

White, E. N. (1990b). When the ticker ran late: The stock market boom and crash of 1929, Universityin White, E. N. (Ed.). Crises and Panics: The Lessons of History (pp. 143-87). Illinios: Homewood.

355

White, E. N. (1996). Stock market crashes and speculative manias, The International

Library of Macroeconomic and Financial History, 13. Cheltenham and Brookfield:

An Elgar Reference Collection.

White, W. R. (2008). Past financial crises, the current financial turmoil, and the need for a

new macrofinancial stability framework. Journal of Financial Stability, 4(4): 307-

312.

Wible, J. (2000). What is complexity? In Colander, D. (Ed.), Complexity and the History

of Economics Thought (pp. 15-30). London: Routledge.

Wigmore, B. A. (1998). Revisiting the October 1987 crash, Financial Analysts Journal, 54(1): 36-48. Malaya Williams, J. B. (1997). The Theory of Investmentof Value (1997 ed. reprint c1938), Cambridge: Harvard University Press.

Willis, H. P. (1930). Who caused the panic of 1929, The North American Review, 229(2):

174-183.

Wilson, J. W., Sylla, R. E., & Jones, C. P. (1990). Financial market panics and volatility in

the long run, 1830-1988, in White, E. N. (Ed.). Crises and Panics: The Lessons of UniversityHistory (pp. 85-125). Illinios: Homewood. Wolf, M. (2000). Stock returns and dividend yields revisited: a new way to look at an old

problem, Journal of Business and Economic Statistics, 18: 18-30.

Wolfram, S. (1984). Cellular automata as models of complexity, Nature, 311: 419-424.

356

Wolfram, S. (1988). Complex systems theory, in Pines, D. (Ed.), Emerging Syntheses in

Science: Proceedings of the Founding Workshops of the Santa Fe Institute: 183-

189. Massachusetts: Addison Wesley.

Wolfram, S. (2002). A New Kind of Science. Illinois: Wolfram Media Inc.

Yahoo Finance. S&P 500 Index. (n.d.). Retrieve October 20, 2010, from:

http://www.finance.yahoo.com

Yahoo! Finance. (2015). S&P 500 Index Historical Prices. Retrieved on 31 December,

2015, from: http://finance.yahoo.com/q/hp?s=%5EGSPC Yamarone, R. (2007). The Trader’s Guide to Key EconomicMalaya Indicators: Updated and Expanded Edition. New York: Bloomberg Press.

Yan, W. (2011). Identification and Forecast ofof Financial Bubbles. (Thesis). Zurich: Swiss Federal Institute of Technology. Retrieved on 28 October, 2014, from:

https://www.ethz.ch/content/dam/ethz/special-interest/mtec/chair-of-

entrepreneurial-risks-dam/documents/dissertation/PHD_YanWanfeng

_Final_thesis_911.pdf

Yan, W., Rebib, R., Woodard, R., & Sornette, D. (2011). Detection of crashes and

rebounds in major equity markets, ETH Risk Center Working Paper Series, ETH-

UniversityRC-11-001.

Yan, W., Woodard, R. & Sornette, D. (2010). Diagnosis and prediction of tipping points

in financial markets: Crashes and rebounds, Physics Procedia, 3: 1641-1657.

357

Yan, W., Woodard, R., & Sornette, D. (2012). Diagnosis and prediction of market

rebounds in financial markets, Physica A, 391: 1361-1380.

Yang, J., & David A. Bessler, D. A. (2008). Contagion around the October 1987 stock

market crash, European Journal of Operational Research, 184 (1): 291-310.

Yanis, M. E. (2002). Riding the Bull, Beating the Bear: Market Timing for Long-Term

Investor. New York: John Wiley & Sons, Inc.

Yu, Pei. (2006). Bifurcation, limit cycles and chaos of nonlinear dynamical systems. In

Sun, J. Q., & Lou, A. C. J. Bifurcation and chaos in complex systems (pp. 1-121). Oxford: Elsevier. Malaya Zarnowitz, V. (1992). Business Cycles: Theory, History, Indicators, and Forecasting. Chicago: University of Chicago Press.of Zeidan, R. (2007). Classic complexity: The theory of complex phenomena. Emergence:

Complexity & Organization, 9(1-2): 131-142.

Zeira, J. (1999). Informational overshooting, booms, and crashes, Journal of Monetary

Economics, 43: 237-257.

Zhou, W-X., & Sornette, D. (2003a). Evidence of a worldwide stock market log-periodic Universityanti bubble since mid-2000, Physica A, 330: 543-583. Zhou, W-X., & Sornette, D. (2003b) Renormalization group analysis of 2000-2002 anti-

bubble in the US S&P 500 index: Explaination of the hierarchy of five crashes and

prediction, Physica A, 330: 584-604.

358

Zhou, W-X., & Sornette, D. (2006). Fundamental factors versus herding in the 2000-2005

US stock market and prediction, Physica A, 360: 459-482.

Zivot, E., & Andrews, D. W. K. (2002). Further evidence on the great crash, the oil-price

shock, and the unit-root hypothesis, Journal of Business & Economic Statistics,

20(1): 25-44.

Malaya of

University

359

LIST OF PUBLICATIONS

Ho, Y. J. & Quah, C. H. (2012). Explaining slow sell-off at stock market after yield curve

inversion with clock game, Actual Problems of Economics, 11: 493-500.

Ho, Y. J., Ruiz Estrada, M. A., & Yap, S. F. (2016). The evolution of complex systems

theory and the advancement of econophysics methods in the study of stock markets

crashes, Labuan Bulletin of International Business & Finance, 14: 68-83.

Ho, Y. J., Ruiz Estrada, M. A., & Yap, S. F. (2017). Examining the heterogeneous regimes

of stock market identified with two variants of B-B algorithms that differ in rigidness of specification, Labuan Bulletin ofMalaya International Business & Finance, (forthcoming). of

University

390