FINANCIAL STRESS IN AN ADAPTIVE SYSTEM: FROM EMPIRICAL VALIDITY TO THEORETICAL FOUNDATIONS
by
MIKHAIL V. OET
Submitted in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
Weatherhead School of Management
Designing Sustainable Systems
CASE WESTERN RESERVE UNIVERSITY
May, 2016 CASE WESTERN RESERVE UNIVERSITY
SCHOOL OF GRADUATE STUDIES
We hereby approve the thesis/dissertation of
Mikhail V. Oet
Candidate for the degree of Doctor of Philosophy. *
Committee Chair
Kalle Lyytinen, Ph.D., Case Western Reserve University
Committee Member
Lucia Alessi, Ph.D., European Central Bank
Committee Member
Agostino Capponi, Ph.D., Columbia University
Committee Member
Myong-Hun Chang, Ph.D., Cleveland State University
Committee Member
Corinne Coen, Ph.D., Case Western Reserve University
Date of Defense
March 5, 2016
*We also certify that written approval has been obtained for any proprietary material contained therein.
Copyright © by Mikhail V. Oet All rights reserved Dedication
To my family.
In every systematic inquiry (methodos) where there are first principles, or causes, or elements, knowledge and science result from acquiring knowledge of these; for we think we know something just in case we acquire knowledge of the primary causes, the primary first principles, all the way to the elements. It is clear, then, that in the science of nature as elsewhere, we should try first to determine questions about the first principles. The naturally proper direction of our road is from things better known and clearer to us, to things that are clearer and better known by nature; for the things known to us are not the same as the things known unconditionally (haplôs). Hence it is necessary for us to progress, following this procedure, from the things that are less clear by nature, but clearer to us, towards things that are clearer and better known by nature. ——Aristotle, Phys. 184a10–21
i
Table of Contents List of Tables ...... v List of Figures ...... vii Abstract ...... x Executive Summary ...... 1 Problem of Practice ...... 1 Research Motivation and Goals ...... 5 Research Design ...... 8 Conceptual map ...... 8 What methods are appropriate? ...... 9 Research plan ...... 11 Chapter Outlines ...... 14 Chapter 1: The Problem of Financial Stress in Adaptive System ...... 25 1.1. Theoretical Framing ...... 25 1.1.1. Issues framed by research in financial system complexity ...... 25 1.1.2. Issues framed by research in financial system stability ...... 30 1.2. Research Precedents ...... 31 1.2.1. Financial system stress construction ...... 31 1.2.2. Stress factor decomposition ...... 33 Chapter 2: Does Financial Stability Matter to the Fed in Setting the US Monetary Policy? ...... 37 2.1. Introduction ...... 38 2.2. Conceptual framework ...... 41 2.3. Data and methodology ...... 45 2.3.1. Content analysis: FOMC discussions of monetary policy ...... 46 2.3.2. Taylor guide to monetary policy ...... 59 2.4. Thematic and Tri-mandate Monetary Policy Models ...... 66 2.4.1. Main results ...... 66 2.4.2. Sign expectations ...... 69 2.4.3. Significance ...... 74 2.5. Discussion ...... 82 2.5.1. Counterarguments ...... 82 2.5.2. Methodological limitations ...... 85 2.5.3. Implications ...... 86 Chapter 3: How to Evaluate Measures of Adverse Financial Conditions? ...... 89 3.1. Introduction ...... 89 3.2. Literature review ...... 92 3.3. Methodology ...... 97 3.3.1. Classification problem ...... 98 3.3.2. Multi-dimensional signaling ...... 99 3.3.3. Comparison of identification properties ...... 100 3.3.4. Comparison of early warning properties ...... 104 3.4. Case Study: Measures of US Systemic Conditions (1976-2014) ...... 105 3.4.1. Data and sampling ...... 105 ii
3.4.2. Results ...... 109 3.5. Conclusion: Implications and limitations ...... 122 Chapter 4: Stress in Heterogeneous Financial Agents: Validity and Dynamics ...... 126 4.1. Introduction ...... 126 4.2. Theoretical Foundation ...... 128 4.2.1. Motivation ...... 128 4.2.2. Hypotheses ...... 129 4.2.3. Micro-level stress in agents and instruments:A conjecture ...... 134 4.3. Empirical Comparison ...... 142 4.3.1. Empirical identification of micro-level stress ...... 146 4.3.2. Empirical macro-level stress in a set of representative markets ...... 146 4.3.3. Exploratory factor analysis ...... 148 4.3.4. Dynamic factor analysis ...... 151 4.4. Discussion ...... 176 Chapter 5: Connecting the Micro and Macro Levels of Financial Stress ...... 180 5.1. Agent Choices and Transmission Dynamics ...... 180 5.1.1. Dynamic analysis of agent stress ...... 181 5.1.2. Agent preferences ...... 182 5.1.3. The stress transmission process ...... 183 5.2. Methodology ...... 185 5.2.1. Revealed preference analysis ...... 185 5.2.2. Stress dynamics ...... 190 5.3. Results ...... 192 5.4. Integrated Research Findings ...... 199 5.5. Conclusion ...... 200 Appendix 1: Chapter 2 Regime Sampling ...... 204 A1.1 Regime Sampling ...... 204 A1.2 Descriptive Statistics ...... 206 A1.3 Robustness ...... 206 Appendix 2: Chapter 2 Content Analysis Methodology ...... 208 A2.1 Target of Content Analysis and Data ...... 208 A2.2 Unitizing and Coding ...... 209 Appendix 3: Chapter 2 Content Analysis Validity ...... 212 A3.1 Face Validity ...... 212 A3.2 Social Validity ...... 213 A3.3 Empirical Validity in Content Sampling and Semantics ...... 213 A3.4 Empirical Validity in Structure and Function ...... 214 A3.5 Empirical Validity in Relations to Other Variables ...... 214 Appendix 4: Chapter 3 Robustness Testing ...... 216 Appendix 5: Chapter 4 MIMIC Factor Model Specification ...... 220 A5.1. MIMIC Identification ...... 220 A5.2. MIMIC Estimation ...... 224 Appendix 6: Chapter 4 Longitudinal Factor Analysis ...... 229 Appendix 7: Chapter 4 Dynamic Factor Analysis ...... 238 A7.1. Motivations for dynamic factor analysis ...... 238
iii
A7.2. Insidious problems and remedies for longitudinal analysis ...... 240 A7.3. Two perspectives of dynamic processes ...... 244 A7.4. Model estimation ...... 251 A7.5. Empirical algorithm ...... 254 References ...... 256
iv List of Tables
Table 1 Outline of Theory Building, Methods, and Data by Research Questions ...... 13 Table 2 Financial Stress Index Construction ...... 32 Table 3 Descriptive Statistics of Themes Discussed at FOMC Meetings (1990–2012) ....48 Table 4 Granger Causality of FOMC Discussions and Monetary Policy (1990M2– 2012M6) ...... 54 Table 5 Results ...... 65 Table 6 Model Horse-Race over Regime Samples ...... 68 Table 7 Summary Statistics for the Stress Series and Benchmark Volatility Series Calculated for Monthly Data between June 2000 and February of 2014 ...... 109 Table 8 Comparison of Interventions and Multi-Dimensional Market Signals under Level Perspective ...... 112 Table 9 Comparison of Interventions and Multi-Dimensional Market Signals under Difference Perspective ...... 112 Table 10 Comparison of Coincident Measures’ Ability to Signal Stress when Selecting , Based on IV ...... 115 Table 11 Comparison of Coincident Measures’ Ability to Signal Stress when Selecting , Based on IV ...... 116 Table 12 Comparison of Coincident Measures’ Ability to Signal Stress when Selecting , Based on ...... 117 Table 13 Comparison of Coincident Measures’ Ability to Signal Stress when Selecting , Based on ...... 118 Table 14 Accuracy of Forecasts ...... 121 Table 15 Usefulness of In-Sample Data Compared to Out-of-Sample Forecasts ...... 121 Table 16 Hypothesis Testing ...... 133 Table 17 Rotated Factor Pattern (Standardized Regression Coefficients)...... 150 Table 18 Reliability of Five-Factor Stress Measure ...... 150 Table 19 Unit Root Tests of Quarterly Financial Stress ...... 156 Table 20 Cholesky Decomposition ...... 162 Table 21 Granger Causality of FX and Equity Observations with Financial Stress ...... 164 Table 22 Dynamic Parallel Factor Analysis for Quarterly Data ...... 167 Table 23 Factorability Analysis for Quarterly Data ...... 168 Table 24 Dynamic Four-Factor Extraction ...... 169 Table 25 Four-Factor Correlation Matrices (Lagged 0 through 4) ...... 170 Table 26 Factor Reliability ...... 170 Table 27 Goodness of Fit Summary ...... 173 Table 28 Analysis of Violations Revealed with Various Price Functions: Observed Price, Expected Return, Risk Premium, and Adjusted Risk Premium ...... 196 Table 29 Granger Causality among Violations Based on Adjusted Risk Premium with 12-month Memory ...... 198 Table 30 Bai-Perron Structural Break Test Results ...... 205 Table 31 Descriptive Statistics of Manifest Variables (1990M2–2012M6) ...... 206 Table 32 Two-Tailed Test of the Mean for Statistical Power (95% Confidence) ...... 207
v Table 33 Comparison of Coincident Measures’ Ability to Signal Stress when Selecting , Based on IV ...... 216 Table 34 Comparison of Coincident Measures’ Ability to Signal Stress when Selecting , Based on IV ...... 217 Table 35 Comparison of Coincident Measures’ Ability to Signal Stress when Selecting , Based on ...... 218 Table 36 Comparison of Coincident Measures’ Ability to Signal Stress when Selecting , Based on ...... 219 Table 37 Serial Correlation Testing of the Weighted Components of CFSI, the Differenced Spreads, and the Differenced Weighted Components ...... 230 Table 38 Normality Testing ...... 232 Table 39 ANOVA Deviation from Linearity F-Test ...... 234 Table 40 Correlation Matrix ...... 236 Table 41 Communality ...... 241 Table 42 KMO and Bartlett’s Test...... 241 Table 43 Anti-Image Correlation Matrix ...... 241 Table 44 Summary Nomological Comparison of Process and Shock Dynamic Factor Models ...... 251
vi List of Figures
Figure 1 Research Plan by Strand ...... 9 Figure 2 Conceptual Model: Early-Warning Policy Use in Adaptive Financial System ...... 27 Figure 3 Conceptual Model of Financial System’s Stress ...... 33 Figure 4 CFSI Components ...... 34 Figure 5 Decomposition of Stress: Components of the Markets ...... 36 Figure 6 Situational Map of Monetary Policy Themes ...... 47 Figure 7 Relative Theme Importance over Time ...... 49 Figure 8 Cross-correlograms of intentions and impact functions for FOMC discussions and monetary policy (1990M2–2008M1) ...... 53 Figure 9 Financial Stability Factors Deliberated in Setting Monetary Policy (Exploratory Analysis, 1993M1–2012M6) ...... 55 Figure 10 Financial Stability Factors Deliberated in Setting Monetary Policy (Algorithmic Analysis, 1990M2–2012M6) ...... 58 Figure 11 Comparison of Standard Taylor and Taylor-type Rules (Table 5 Columns 1–4) with Fed Funds Rate (Regimes 1–4) and Average SSR (Regime 5) ...... 60 Figure 12 Comparison of Thematic (Discussion-Augmented Taylor-Type) Model (Table 5 Column 8), Tri-Mandate Model (Table 5 Column 9), and Benchmark Taylor-Type (1999) Model (Table 5 Column 4) with Fed Funds Rate (Regimes 1–4) and Average SSR (Regime 5) ...... 64 Figure 13 Percentage of Total US Financial Assets Held by Financial Intermediaries (1952‒2013) ...... 90 Figure 14 Multidimensional Signal Compared to Several Measures of Adverse Systemic Conditions ...... 107 Figure 15 Optimal Matching of Interventions and Multi-Dimensional Market Signals ...... 111 Figure 16 VEC Forecast Results ...... 120 Figure 17 Hypotheses ...... 133 Figure 18 Conceptual Diagram of the Adaptive System Solution ...... 135 Figure 19 Comparative Theoretical Stress for Heterogeneous Agents ...... 140 Figure 20 Comparative Theoretical Stress for Heterogeneous Instruments ...... 141 Figure 21 Comparison of CFSI and Comparative Theoretical Stress ...... 142 Figure 22 Comparison of Five-Factor Stress and Comparative Theoretical Stress ...... 144 Figure 23 Comparison of Process Factor Model Stress and Comparative Theoretical Stress ...... 145 Figure 24 Comparison of Shock Factor Model Stress and Comparative Theoretical Stress ...... 145 Figure 25 MIMIC Identification of Latent Micro-Level Stress ...... 146 Figure 26 Stages and Steps of Empirical DFA ...... 152 Figure 27 Quarterly Time-Series of US Financial Stress ...... 154 Figure 28 Correlogram of Quarterly Financial Stress ...... 156 Figure 29 Stylized Block-Diagonal Toeplitz Matrix ...... 157 Figure 30 Toeplitz Intensity and Symmetry Pattern for Quarterly Financial Stress ...... 158
vii Figure 31 Correlogram of Weekly Financial Stress ...... 165 Figure 32 Process Factor Model of Quarterly Financial Stress ...... 172 Figure 33 Shock Factor Model of Quarterly Financial Stress ...... 174 Figure 34 Weak Axiom of Revealed Preferences ...... 188 Figure 35 Initial Revealed Preference Testing for All Agents ...... 193 Figure 36 Detailed Revealed Preference Testing for Heterogeneous Agents ...... 194 Figure 37 Testing Price, Expected Return, Risk Premium, Adjusted Risk Premium to Explain Agent Choice ...... 197 Figure 38 Testing Agent Memory to Explain Choice ...... 197 Figure 39 Contagion Network ...... 199 Figure 40 MIMIC Factor ...... 224 Figure 41 Scatterplot Matrix of CFSI and Market Stress Variables ...... 233 Figure 42 Outlier Boxplots of Market Stress Variables...... 235 Figure 43 Process Factor Analysis Model: PFA (1,0) ...... 247 Figure 44 Process Factor Analysis Model: PFA (0,1) ...... 247 Figure 45 Shock Factor Analysis Model: SFA (1) ...... 249
viii Acknowledgements
I am indebted to my wife Inna, my family, and Stephen Ong for making this
journey possible and for their encouragement and advice while travelling with me.
I am deeply grateful to Dr. Kalle Lyytinen, my dissertation supervisor, for his guidance in shaping my research from its very beginnings, and Dr. Lucia Alessi, Dr.
Agostino Capponi, Dr. Myong-Hun Chang, and Dr. Corinne Coen, for pushing me to strengthen and extend the boundaries of this research.
I would like to thank the faculty at Case Western Reserve University doctoral program in Designing Sustainable Systems, particularly Dr. Richard Boland, Dr. Jagdip
Singh, Dr. Eileen Doherty-Sil, and Dr. Lee Hoffer, for their intensity in expanding my research horizons.
I am grateful for the insightful critiques received from Dr. Joseph G. Haubrich,
Dr. Ben R. Craig, Dr. Charles T. Carlstrom, Dr. O. Emre Ergungor, Dr. James B.
Thomson, Dr. Owen F. Humpage, Dr. Edward S. Knotek II, Dr. Filippo Occhino, Dr.
Mark Schweitzer, Dr. Mark S. Sniderman, Saeed Zaman, John M. Dooley, Jack W. Liu,
Amanda Janosko, and Tim Bianco. To John, Jack, Amanda, and Tim, I am also deeply grateful for research assistance.
I would like to thank Sue Nartker and Marilyn Chorman, the Directors of the
Doctor of Management program, for making my time at Case Western Reserve
University a memory to treasure for a lifetime.
ix Financial Stress in an Adaptive System:
From Empirical Validity to Theoretical Foundations
Abstract
By
MIKHAIL V. OET
A review of financial system stress measures reveals not only the absence of theory on financial stress, but also the absence of search for theory. To remedy this gap, this study conducts a rigorous investigation of the empirical validity and dynamic properties of financial stress measurement in the context of financial system complexity. We provide and validate four contributions to literature.
First, we establish the relevance and comparative quality of macro-level stress
measurement for the financial system relative to alternative measures of system
conditions.
Second, we establish theoretical foundations for measuring financial stress across multiple units of analysis. This measure builds on the understanding of stress origins and drivers and incorporates price, quantity, and behavioral variables to explain the pattern of
apparently irrational choices of financial agents. At the macro-level, stress is supported
empirically by hypotheses of association between behavioral aspects of heterogeneous
financial agents and overall financial system stress. At the micro-level, we apply
abductive inference to the empirical results to propose a new theoretical stress measure
x for heterogeneous agents and instruments. Defining financial stress theoretically allows
continual measurement of financial stress at the level of the various heterogeneous
partitions of the financial system (e.g. agents and instruments) as these partitions evolve
through structural changes and financial innovations.
Third, we build a theory of stress transmission across micro-level of sectoral
agents to the macro-level of the financial system. This theory describes a process of stress
transmission across financial intermediaries and the process by which its agent stress escalates to the financial system.
Fourth, we examine the process by which unusual conditions in the financial
markets manifest as critical states of financial system stress.
Keywords: financial stress; heterogeneous agents; empirical validity; factor analysis;
dynamic factors; stochastic analysis; content analysis
xi Executive Summary
Problem of Practice
People view and economists study financial tsunamis as distinct and unique events. During the roaring nineteen-twenties, financial system was “lightly regulated”1
and people believed the stock market would rise forever. Uninformed people borrowed
heavily to speculate. When the stock market collapsed in the Great Crash of October of
1929, it dragged many banks to bankruptcy. Millions of depositors in these banks were
unprotected and lost everything. As purchases shrank, manufacturing withered leading to
massive layoffs and economic collapse known as the Great Depression which ravaged the
country for the next ten years.
At the opening of 1980s, the interest rates skyrocketed to about 20%. The small
savings and loan banks, which were prevented by regulations2 from paying market rates,
lost millions of deposits as savers withdrew their funding to seek higher returns from
money market mutual funds. To remedy this, financial deregulation of 19803 lifted the
rate ceilings on deposits. The savings and loans institutions rushed in to chase the
depositors, and then followed to finance the costly deposits by chasing high yields in the
real estate markets. As real estate prices collapsed, so did the banks. From 1981 to 1995,
over 1,043 institutions holding $519 billion in assets—close to half of U.S. savings and
loan banks—collapsed.4
1 Acharya et al. (2010: 13). 2 Sherman, M. (2009). 3 The 1980 Depository Institutions Deregulation and Monetary Control Act (DIDMCA). 4 Curry and Shibut (1986), Spilimbergo et al. (2009).
1 For several years in the mid-1990s, an unregulated U.S. hedge fund5 called Long-
Term Capital Management (LTCM) enjoyed tremendous reputation and success. Run by two 1997 Nobel Laureates, Myron Scholes and Robert Merton,6 the fund borrowed heavily from some of the most significant U.S. financial institutions and accumulated
$1.3 trillion of derivative exposures, or about 32% of all U.S. commercial bank assets.7
When global investors ran for the safety of U.S. Treasuries, following the 1997-1998 fiascos of the Asian and Russian currencies, they created unusual price discrepancies in the value of the dollar. In the course of four months, LTCM exposures quickly lost $4.6 billion dollars, threatening to topple the U.S. financial system.8
On June 22, 2007, “Bear Stearns Companies, the investment bank,9 pledged up to
$3.2 billion in loans … to bail out one of its hedge funds that was collapsing because of bad bets on subprime mortgages. It [was] the biggest rescue of a hedge fund since 1998 when more than a dozen lenders provided $3.6 billion to save Long-Term Capital
5 Paredes (2006). 6 Scholes and Merton won a Nobel Prize for a new method to value financial derivatives, payment contracts that depend on future values of an underlying asset. 7 Summers et al. (1999: 29): “The notional amount of LTCM’s total OTC derivatives position was $1.3 trillion at the end of 1997 and $1.5 trillion at the end of 1998. LTCM’s balance sheet leverage was 28- to-1 at the end of 1997.” For comparison, “At the end of 1998, for instance, commercial banks had $4.1 trillion in total assets; mutual funds had assets of approximately $5 trillion; private pension funds had $4.3 trillion; state and local retirement funds had $2.3 trillion; and insurance companies had assets of $3.7 trillion.” (Ibid.: 1-2). 8 LTCM strategy was based on the idea that the value of U.S. dollar 30 years from now should be very close to the value of the U.S. dollar 29.75 years from now. However, this turned out to be false in 1998. Following a wave of speculative attacks in May 1997, the Thai Bhat tumbled, sparking off a chain reaction of dropping currency and asset prices throughout Asia. Then, in August 1998, Russia defaulted on its international obligations. As global investors ran for cover, dumping Asian and Russian assets, they fled to the safety of the U.S. dollar, clamoring for the most liquid U.S. treasuries and creating unusual price discrepancies in the value of the dollar. 9 As a consolidated entity, Bear Stearns was a part of the voluntary “Consolidated Supervised Entities (CSE) program, created in 2004 as a way for global investment bank conglomerates that lack a supervisor under law” (Securities and Exchange Commission, 2008). Also see Ryback (2015: 3). Coffee and Sale (2009: 736-737) discuss that “A key attraction of the CSE Program was that it permitted its members to escape the SEC’s traditional net capital rule, which placed a maximum ceiling on their debt to equity ratios, and instead elect into a more relaxed “alternative net capital rule” that contained no similar limitation.”
2 Management.”10 Across the Atlantic, on August 9, 2007, “BNP Paribas SA, France's
biggest bank, halted withdrawals from three investment funds because it couldn't “fairly”'
value their holdings after U.S. subprime mortgage losses roiled credit markets.”11 So
innocuously began the 2007 subprime mortgage crisis, which so far has cost the U.S.
taxpayers in excess of $3 trillion including the $356.2 billion in the Trouble Assets Relief
Program, $1.5 trillion in Federal Reserve rescue efforts, and $577.8 billion in Federal government stimulus programs.12
The notion that our lives are punctuated by these exceptional financial crises is
not new. Writing about the U.S. unregulated era,13 Wesley C. Mitchell (1923) observed
that “Fifteen times within the past one hundred and ten years, American business has
passed through a “crisis”…Further, no two crises have been precisely alike and the
differences between some crises have been more conspicuous than the similarities. It is
not surprising, therefore, that business men long thought of crises as “abnormal” events
brought on by some foolish blunder made by the public or the government. On this view
each crisis has a special cause which is often summed up by the newspapers in a
picturesque phrase “the Jay Cooke panic” of 1873, “the railroad panic” of 1884, “the
Cleveland panic” of 1893, “the rich man's panic” of 1903, “the Roosevelt panic” of
1907.”14
Furthermore, modern financial system links different types of financial
intermediaries by a sophisticated network of multilateral exposures where various risky
10 Creswell and Bajaj (2007). 11 Boyd (2007). 12 Goldman (2009). 13 The regulatory era of the United States financial system is here considered to start with the creation of the Federal Reserve regulatory system in December 23, 1913 by the Federal Reserve Act. 14 Mitchell (1923: 5).
3 activities of some institutions are financed using funds borrowed from others.
Specifically, small financial intermediaries use customer deposits to make loans to large universal intermediaries that depend on wholesale short-term funds to finance a gamut of risky activities. As the value of financial assets falls, financial institutions experience increased difficulty in repaying current obligations, raising funds, and remaining solvent and liquid. Furthermore, through these linkages, failed obligations of some institutions lead to distress and losses in other institutions, markets, and economic sectors (Acharya and Yorulmazer, 2008; Iyer and Peydro, 2011; Tedeschi et al., 2012).
The problem of contagion in the financial system has been established as a source of significant levels of systemic risk (Freixas et al., 2000; Allen and Gale, 2000; Furfine,
1999) well before the Global Financial Crisis of 2007. However, the problem is compounded viciously by structural changes occurring in the system. As the system structure changes, so does the pattern of systemic risk generation (Nier et al., 2007;
Battiston et al., 2012; Lenzu and Tedeschi, 2012; Georg, 2013; and Sachs, 2014).
Consequently, policies targeted at mitigating systemic risk need to account for the evolving system structure. We refer the reader to Allen and Gale (2007), Teteryatnikova
(2009), Acharya et al. (2012), and Capponi and Chen (2013) for a more complete discussion in this regard. Hence, financial system connections constitute an important source of systemic risk (Rochet and Tirole, 1996) and need to be carefully analyzed.
4 Research Motivation and Goals
We are motivated by a particularly vicious problem of stress in the financial system. Both the experiential and theoretical evidence for this problem suggest that financial system is a complicated and dynamic phenomenon prone to significant risks and potential failure.
The present state of knowledge of financial system’s stress displays few well- understood areas where theory is validated by empirical information. At the same time, a review of extant measures of financial system conditions reveals not only the absence of theory on the measurement of system’s conditions, but also the absence of search for such theory. In current context, our knowledge of financial stress is particularly poorly across multiple units of
analysis: the macro-level of the financial system, the micro-level of the financial system’s
agents and instruments, and the meso-level of the processes by which these agents and instruments interconnect and participate in the propagation of stressful conditions across the financial system and from one level to the others. These three challenges of stress measurement at the micro-, meso-, and macro- levels of the financial system are framed within the fundamental puzzle of system complexity.
First, there is a micro-level stress problem: As stress is experienced by some financial assets, they fall in value. The stress in the particular financial assets is also experienced by the financial institutions which hold these assets directly. They experience difficulty in repaying obligations, raising funds, and remaining solvent and liquid.
Second, there is a macro-level stress problem: As some financial institutions experience direct stress, it is transmitted indirectly through the various linkages of these institutions in the financial system. These linkages include ties that are contractual and
5 strong, when failed obligations of the stressed institutions lead to disruptions in expected payments, distress and losses in other institutions, markets, and economic sectors. The linkages also include ties that are tenuous and weak, when stressed values in specific instruments and firms affect the conditions of other institutions and instruments that share some similar characteristics and are re-valuated by other financial system participants less favorable. Thus, stress can also propagate indirectly through the contagion of suspected weaknesses.
Third, there is a meso-level stress problem. Since stress can be transmitted in a number of ways, both direct and indirect, it is possible for stress “to pile-up,” accumulating redundantly. Clearly this affects all levels of stress measurement: same institution that is stressed because of failed obligations of one of its counterparties may also be additionally stressed by association because some of its assets bear resemblance to other stressed assets. At the larger units of analysis, entire markets can become tainted with stress in multiple ways. For example, as residential mortgage backed securities experience stress, stress can transfer to other types of securitized products like auto loans and to other markets like other real estate instrument in the real estate market or the consumer durable sector of the equity market. When the levels and changes in stress are moderate, it may take a number of conditions to change so that the change in the financial system stress is experienced as critical. However, under some extreme levels and changes of stress, a disaster in a single market like equity or a small cluster of institutions like high-powered banks or hedge-funds can quickly bring the entire financial system to its knees.
6 The fundamental puzzle of stress lies in the simple fact that things change. As the financial system’s structure changes, so does the pattern of stress generation. Thus, any solution to the micro problem, to the macro problem, or the meso problem that is conditioned on the historical facts is bound to eventually outgrow its usefulness.
At the macro-level, our knowledge confronts the innate vulnerability of the financial system to change like falling asset values. At the micro-level, we don’t well understand how stress affects the choices of financial agents. At the meso-level, we have difficulty in identifying the process of stress transmission and the conditions under which stress leads to critical states when policymakers must intervene. Any hope that we have to the meaningful understanding and measurement of financial stress has to confront the following six major research questions.
First, given the evidence of financial system complexity, how can the financial stress be theorized to enable the policymakers and researchers to understand it as the system undergoes change? Specifically, to the extent that stress feeds on the aspects of system’s complexity, like uncertainty, how does system complexity condition financial stress? Second, why has financial stress has become relevant to the policymakers? Third, how can we evaluate the quality of the stress measure relative to alternative measures of financial system conditions? Fourth, how can macro-level stress be defined, reliably measured, and identified in the context of an evolving financial system? Fifth, given the complexity in the pattern of stress formation at the micro-level of analysis, can agent- and instrument- stress be described functionally in a way that remains meaningful and empirically reliable as the agents undergo change? Sixth, how does stress in set of agents or instruments transfer to another set of agents or instruments?
7 The first three research questions involve advancing our knowledge of measurement of system conditions in the context of adaptive change in the system. The next three research questions involve building multi-level theory of financial stress in the adaptive context. With the last research question, we tie both areas of inquiry together by seeking to understand the causation mechanisms by which financial stress propagates across the levels of analysis. It is our intent to develop an empirically sound and theoretically insightful understanding of financial stress in order to measure it at the various units of analysis as the system undergoes change.
Research Design
Conceptual map
Because of the dual—empirical and experiential—nature of these research questions, we pursue our questions in a sequential multiple method design. We pursue the empirical questions by a number of qualitative and quantitative methods that are appropriate to the research questions posed in each chapter. For the first study of system complexity and stress, we use qualitative methods of narrative analysis. For the second study of relevance of stress, we use content analysis of policymakers’ discussions. For the third study of comparative evaluation of information quality in alternative measures of system conditions, we use a general purpose evaluation framework for that combines standard two-class classification task methods with utility-augmented time-series methods. For the fourth study of macro-level stress, we use factor analytic methods, including dynamic factor analysis. For the fifth study of micro-level stress, we propose to use the microeconomic method of revealed preferences. For the sixth study of stress transmission across multiple units of analysis we use to augment the time-series method
8 of Granger causality. Figure 1 shows the conceptual map of the research plan. As shown, we propose to structure this research sequentially, incorporating the results from one study into the next.
Figure 1 Research Plan by Strand
What methods are appropriate?
To address the multiple process study questions, the study needs to include a multi-strand sequential mixed method approach (Caracelli and Greene, 1993; Creswell &
Plano-Clark, 2010, Creswell et al., 2002; Johnson & Onwuegbuzie, 2004). This mixed- method design of each strand starts with a narrative (qual) and proceeds with empirical
(Quant) development. In addition, because, as described below, the study adapts a quota sampling strategy and confronts some ecological validity challenges, we pursue a quasi- experimental approach (Shadish, Cook & Campbell, 2003). In our implementation of this
9 approach, we will apply multiple repeated quasi-experiments with pre-test and post-test treatments for quarterly for 41 financial agents from 1970 to 2014. We will extract a hold-out sample for subsequent out-of-sample testing of financial stress for ecological validity.
Sampling Strategy. It is clearly a nearly impossible task to include the census population of over 7000 banks, and thousands of other financial system agents and instruments. We also consider that randomization approaches may introduced an unwanted bias, since the distribution of the population is highly skewed: there are very few very large institutions that hold the majority of the financial assets, an absolute majority of very small institutions that collectively have a tangible share of assets.
Importantly, there is a small important of significant and critical regional financial intermediaries that in may financial product markets play an intermediary role between to facilitate the transactions between the very small and the very large agents. Because of the skewness, a random sample may introduce large distortions into the sample data that may subsequently fail the representative ecological and predictive validity tests.
Therefore, we propose to a convenience sampling approach, improved through the quota sampling approach. In this approach our sample will be constructed to reflect the relative representative structure in the relevant agent types (e.g. a representative quota sample of the 3 banking tiers).
Validation Strategy. In order to support generalizability claims of this research, the alternative study needs to include a comprehensive approach to validity (Campbell,
1957; Bryman, 2012) and specifically recognize the questions of experimental validity
10 that are integral to supporting the quasi-experimental design validity concerns (Shadish,
Cook & Campbell, 2003).)
Specifically, the study should demonstrate face validity in building its narratives qualitatively during the abductive theory building phase. The face validity concerns will need to address extant behavioral conceptions of stress in agents and instruments. The construct validity should be demonstrated against the sample data in the empirical analysis. The convergent validity will be compared by aggregating the meso level constructs of agent and instrument stress to the macro level of system stress. This comparison can be seen as an initial attempt for greater generalization, and thus a form of external validity. Finally, the empirical analysis needs to test the developed stress measures to test ecological validity of the measures, that is, whether these measures can reproduce the patterns the patterns of agents in natural settings. To test ecological validity, the study will reserve an out-of-sample dataset not used in the analysis, and compare the supported measures of stress to this sample to verify how well the measures of stress developed in-sample can support the “natural” out-of-sample facts.
Research plan
The seven research questions in this study aim to cycle from empirical validity research to theory generation. The reason for this broad scope is the vicious problem of adaptive heterogeneous agents in a complex evolving financial system. The dual nature of the research calls for a sequential mixed method design that combines the empirical and experiential research streams in a single study. For the empirical stream, there are additional compelling reasons to combine qualitative and quantitative components in an embedded research process. This is due to the fundamental difficulty in longitudinal
11 factor research: in order for the latent factors to be identified quantitatively, the researcher must be able to support their qualitative interpretation. Specifically, since our unit of analysis includes distinct heterogeneous agents within an identifiable financial sector, we must support deductive inference of factors and agent groups over time. To pursue this, we choose multiple quantitative data sources with purposive sampling. We use maximum variation sampling to identify extreme observations and homogeneous sampling to maintain relevant agent groups. With regards to quantitative methods, we choose multiple quantitative methods (dynamic factor analysis, structural equations modeling, time-series analysis, and stochastic modeling) to support the inductive interpretation of factors and agent groups over time. In order to support the interpretation validity of the quantitative measures we choose qualitative content analysis as a principal triangulation method. In turn, we substantiate the content analysis by choosing multiple qualitative data sources including the transcripts of the Federal Reserve discussions and the text of financial news. Table 1 provides an outline of theory building, methods, and data for each of the five research questions.
12 Table 1 Outline of Theory Building, Methods, and Data by Research Questions
ID Research questions Theory building Method Sample data Chapter
What is the evidence for the financial system complexity, and how can Multiple Qual i. the financial system →Inductive sources Chapter 1 observation distress be studied by the (1976-2014) policymakers going forward?
FOMC transcripts, Is financial system stress Mixed financial →Inductive→ ii. relevant to the (embedded): news→ Chapter 2 Deductive policymakers? → Qual→Quant Multiple sources (1991-2014)
How can we evaluate the quality of the stress Multiple measure relative to iii. →Deductive Quant analysis sources Chapter 3 alternative measures of (1976-2014) financial system conditions?
What is system-level stress? How can system Multiple and agent stress and iv. →Deductive Quant analysis sources Chapter 4 financial instability be (1976-2014) defined and reliably measured?
Is there some way to Multiple Chapter 4 remedy our understanding →Abductive → v. Quant analysis sources and of the functional form of Deductive (1976-2014) Chapter 5 financial stress?
How does stress in set of Multiple Mixed agents or instruments →Abductive→ sources vi (embedded): Chapter 5 transfer to another set of Deductive (1976-2014), → qual→Quant agents or instruments? Flow of Funds
We consider the first question, what is the evidence for the financial system complexity, and how can stress be studied by the policymakers going forward, in Chapter
1. Here we detail the dimensions of the problem of study and draw on the extant literature to establish the epistemological precedents for the study. We position our research critically in reference to existing literature. We study the second question, is financial system stress relevant to the policymakers, in Chapter 2, by validating the measure of financial system stress against the incidents of policymakers’ discussions of financial system conditions during their discussions in the Federal Open Market Committee
13 meetings. We study the third question, how can we evaluate the quality of the stress measure relative to alternative measures of financial system conditions, in Chapter 3, by testing the quality of alternative measures of US financial system conditions. We consider the fourth question, what is macro-level stress, in Chapter 4, by testing hypotheses of stress toward systemic stress measurement against US financial system data. We consider the fifth question, is it possible to reconcile our understanding of micro-level stress with the observations of apparent irrationality in the choices of financial agents, in Chapters 4 and 5, by testing alternative explanations of agent choices, e.g. prospect theory, memory, cognitive dissonance, to improve explanation of individual agent choices toward an adjusted functional form of stress able to improve reconciliation of individual agent and instrument stress with systemic stress. We consider the sixth question, how does stress in set of agents or instruments transfer to another set of agents or instruments, in Chapter 5, by analyzing the transmission process stochastically. We do this by examining the causal relationships between the violations of weak axiom of revealed preference under a variety of price functions and alternative memory spans in heterogeneous representative financial agents.
Chapter Outlines
The idea that ties together the various aspects of this dissertation is the search for empirically-valid theory of financial stress. Thus, a robust investigation of validity of financial stress measurement is essential for the building of financial stress theory.
Supporting measurement validity involves testing of several types of validity. In Chapter
2, we consider face validity for the measurement of financial system stress. We support system stress prima facie since this concept has relevance in the actual decision making
14 processes of policymakers and is consistently important for their assessment of financial system conditions and financial stability. In Chapter 3, we consider comparative validity of alternative measures of financial conditions seeking to study the quality of the alternative measures of financial stress. In Chapter 4, we consider internal validity of the system stress measure by examining the reliability of components in explaining the observed variance in the series of selected financial system indicators.
After establishing a valid system-level measure of stress, to enable stress measurement at other units of analysis, it is essential to define a theoretically sound measure of agent- and instrument-level stress. In Chapter 5, we consider predictive validity of financial stress measurement. Here we test alternative definitions of agent- level financial stress to minimize the variance between stress measured at multiple units of analysis. Specifically, we seek to refine a functional form of stress measured at the agent- and instrument-level, so that measurement error between the aggregated agent- level financial stress and the aggregate system level financial system is minimized. Upon empirically valid and theoretically supported agent level financial stress, the advancement of further stress theory involves theorizing about the processes by which financial stress at the agent level propagates across agents to the system level stress. Put differently, we seek to explain the processes by which stress propagates across multiple units of analysis. Accordingly, to advance financial stress theory, we investigate three types of stress propagation processes: 1) agent preference and choice, 2) transmission, and 3) conditions.
Chapter 1: System complexity and distress. In Chapter 1, we discuss the evidence for classifying financial systems as complex rather than simply complicated. We review
15 the weaknesses and strengths of different theoretical positions in explaining the processes of emergence for such outcomes. We discuss the need for multi-level dynamic explanations and their potential value in explaining financial system risks.
In this chapter, we provide an integrative theoretical review across knowledge domains and multiple-levels of analysis to review various conflicting viewpoints and discuss an integrative path forward. This path lays the generative understanding of the financial system as a complex adaptive system, where we explain the emergent dynamic phenomena through the ability to reproduce the observed structures, functions, and behaviors across its agents by establishing simple rules for their interaction. In particular, we focus on the emerging role of financial distress as a potential tool for policymakers concerned with understanding and influencing financial system conditions and outline a research agenda relevant in the context of financial system complexity.
Chapter 2: Face validity. In Chapter 2, we address face validity by examining whether financial system conditions matter to the Fed? This entails qualitative analysis of policymakers’ discussions. We find that financial conditions matter significantly to the
US policymakers and that they can be proxied by an aggregate measure of financial system stress. Quantitatively, the measure of system stress that is found to be similar to the discussion-based series on system stability and conditions is a conjectural construct composed of representative indicators and weighted dynamically by the changing relative share of broad financial sectors in the overall US economy. Thus, this conjectural aggregate measure of stress that serves as a starting point of my research has some validity, specifically face validity. This measure is also an aggregate measure of US
16 financial system stress. Of course, many assumptions went into it, and additional types of validity remain to be addressed.
Chapter 3: Comparative validity. The intent of comparative validity is to assess the relative quality of alternative empirical measures of financial system conditions, in order to choose the optimal basis for building financial stress theory. We examine the quality of the system-level stress measure, e.g. its composition with selected representative financial indicators, by comparison of this measure against alternative measures of system conditions. While it is useful to know that a measure of aggregate financial system stress provides a good proxy for the underlying conditions of the system and its stability as captured in the policymakers' discussions, we don't know what particular measure of system stress is optimal. Choosing the best of the alternative measures of financial stress enables us to narrow and support the choice for the definition of stress at the individual unit of analysis (agent and instrument).
An initial comparison of alternative measures may begin by the comparison of correlation of alternative measures of system conditions with each other and the reference series (e.g. a reference series of financial system conditions as discussed by the policymakers or a benchmark series of outcomes). However, this approach is inadequate in comparison the relative quality with which the various alternative measures capture the dynamic aspects of critical system outcomes. The correlation is not adequate in describing the dynamic pattern of system outcomes as they occur. When critical outcomes occur, does a conditions measure identify the same period? Does the measure similarly indicate the duration of these periods? Does the measure accurately capture the pattern of outcome effects across system markets? Thus, it is desirable to compare the
17 alternative measures by the extent with which they capture the severity, persistence, and pervasiveness of market conditions. This is a more insightful comparison of alternative measures than a simple correlation. The severity aspect describes the relative magnitude of system outcomes relative to historical precedent. The persistence aspect describes the extent to which an unusual high market condition persists in that market. The pervasiveness aspect describes the extent to which an unusually high set of market conditions is present across several markets. Specifically, we seek compare the alternative measures as follows: When disturbances of particular severity occur in particular markets, which of the measures are able to better describe the effect pattern in system for both persistence and pervasiveness. Taken together, the comparison is founded on the summary quality of the comparison in these three dimensions among the alternative measures.
Chapter 4: Internal validity—static and dynamic analysis. Thinking of stress as a function of prices and quantities at the agent and instruments enables a straightforward, but unvalidated, quantification of stress at the agent- and instrument-level and its aggregation across agents and instruments. This quantification can be done, revealing the time series of agent and instrument stress and allowing for the time pattern of agent and instrument stress to be "observed." However, the quality of this quantification remains unsupported without two additional forms of validity testing: internal and predictive. In the internal validity tests (static analysis, we make modest improvements in the composition and the reliability of the macro-level stress. In the ensuing predictive validity tests (dynamic analysis), we make additional improvements in understanding the
18 time pattern of stress generation process and find that recognition of memory effects is important to varying degree across financial system markets.
Static analysis: Internal validity testing helps us to establish the reliability of the aggregate financial stress measure itself. In Chapter 4, we examine internal measurement reliability of the system-level measure by factor-analytic techniques. Being rigorous about the way the aggregate variance is explained through the selected set of markets
(factors) and the particular set of indicators chosen to represent each market, allows us to improve the internal reliability of the system-level stress measure.
Dynamic analysis: We may question the absence of behavioral effects in the definition of stress. Specifically, we observe that at the system- and market-level, the pattern of stress exhibits some stickiness (persistence and pervasiveness) properties. To the extent that the definition of stress at the individual unit of analysis fails to incorporate an explanation for this memory pattern, we fail to explain how this stickiness in aggregate emanates from the individual units of analysis. Our own heuristic rules by which the system stress is signaled include sticky properties of persistence and pervasiveness. It is reasonable that these properties have both an empirically-based and theoretically supported expression in the way that individual unit stress is defined and measured. Methodologically this means that the static factor analysis of Chapter 4 should be amended by dynamic factor analysis.15 Applying dynamic factor analysis, we find
interesting and valuable adjustments to the understanding of the dynamic properties of
the aggregate financial stress. Specifically, we find that the dynamic effects differ across
15 In the same vein, static factor rests on the absence of violations of serial correlation. Given presence of serial correlation in the results, e.g. observed in sticky time results, the dynamic effects need to be investigated through dynamic factor analysis.
19 markets: presenting short memory in highly active markets (equity and foreign exchange) and long memory effects in less active markets (credit and securitization, interbank funding, and real estate). Thus, we find that the incorporation of memory is important.
However, despite the incorporation of a series of beneficial adjustments to the aggregate financial system stress, a discrepancy remains between the system level stress and the aggregated individual level stress.
Chapter 5: Predictive validity—dynamics, preferences, and agent choice.
Despite a rigorous effort to establish an optimal macro-level measure of financial stress, clearly some variables in addition to prices and quantities are needed to explain both the pattern of apparently irrational agent choices and the apparent discrepancies between the micro- and macro-level stress. In this chapter, we advance agent-level stress theory. We examine its predictive validity by examining whether the proposed functional form of agent-level stress aggregates as expected to the system-level. Specifically, does the agent stress that is measured as conjectured integrate across all system agents to the system- level stress? If our definition of stress at the micro-level (agent and instrument) is correct, then the integration of the micro-level stress should yield the macro-level (system) stress.
Our initial literature-supported abductive conjecture is that micro-level stress is defined as a function solely of prices and quantities. Testing this, we find that macro-level
(system) stress does not equal the summation of micro-level (agent) stress and observe substantial differences between the two.
The advancement of micro-level definition of stress requires us to theorize functional form supported by literature-based hypotheses of stress and the available empirical evidence. Initially, the inference is abductive. Because we observe a reliable
20 measure of macro-level stress as a function of prices and quantities of representative indicators, we infer that stress at smaller units of analysis, e.g. agents and instruments, can be measured as a function of prices and quantities of relevant instruments which fully describe the exposure of the financial agents.
Dynamic analysis for agent stress: To make further progress, we need to critique the agent level definition of stress. The remaining reason that the aggregate stress does not equal the sum of agent stress is that the agent stress is not adequately measured. We already know that memory is important in system level stress, yet, memory is absent from agent level stress definition. The initial conjecture of agent stress includes only prices and quantities. To the extent that agents act rationally, then prices and quantities should be the only variables that explain the agent choices. In this chapter, we consider the choices across all system agents and find that using an aggregate representative agent as a unit of analysis, the allocation choices can be considered rational, with very small number of rationality exceptions (2% or less). Thus, the assumption of rationality is reasonably supported across all agents, confirming that agent allocation choices are substantially explained by prices and quantities. However, when individual agents are considered, significant number of allocations appear irrational, with violations of rationality evidenced in forming the agents’ price and quantity bundles. Thus, individually, the agents appear to be making irrational allocations. In fact, a significant number of violations exist for specific agents and appears associated with the periods of high system stress. This leads us to the idea that the functional form of agent stress is incorrectly specified and requires the incorporation of variables other than prices and quantities.
Accordingly, in this chapter we seek to recognize and include the variables missing from
21 the definition of stress. We do this, by testing alternative theories of agent preference, seeking to find the set of variables that minimizes the number of violations of agent rationality in allocating the agent choices.
Agent preferences: For the choice process, we seek to explain the pattern of apparent irrationality in agent choices and to understand the process by which the pattern of choice violations becomes system stress. This analysis is accomplished by examination of agent revealed preference and the causal pattern tying the apparent violations of rationality and financial system stress. In this chapter, we also investigate alternative variables in the definition of agent-level financial stress. The first two variables, price and quantity, form the basis of stress agent-level stress as established through the agent rationality assumption, where agent allocations among instrument bundles are determined solely on the basis of rational preferences among instrument prices and quantities. When agent choices are conditioned solely by the observations of prices and quantities of financial instruments, we observe that violations from historical rationality tend to occur during particular times—the time of high distress. Naturally, we would like to examine whether there are any relationship patterns in the transmission of stress that involve these violations. Thus, we seek to extend the specification of agent-level stress with additional variables to investigate whether they can help to explain and minimize the number of apparent violations. We expand the functional specification of agent-level stress with additional variables that provide useful information to explain actual agent choices. The variables we test come from behavioral theories of agent choice, including prospect theory, cognitive dissonance, memory (animal spirits), and liquidity preference. The
22 inclusion of additional variables completes the micro-level definition of stress and minimizes its discrepancy with macro-level stress.
The stress transmission process: In Chapter 5, we also explore the empirical pattern of connections between the violations of rationality in various sectors and financial stress. We find evidence that a set of sectoral allocations influences the transmission of financial stress from sector to sector.
Using our empirical findings, we extend financial stress theory with theory of the stress transmission process. Starting with the narrative stories of contagion in the extant literature, we advance the idea that agent choices and stress propagation are conditioned by building asset bubbles. Agents make violations of apparent rationality in choosing certain price and quantity bundles. As these violations are made, there is a snowballing effect. It becomes more profitable for agents to herd, and that is a story evidenced and tested in the data. As a bubble in asset j inflates, as indicated by the presence of apparent violations in price and quantity bundles, it becomes reasonable for agents to change their allocations to profit from short-term expectations in asset j. As the instrument j inflates, it offers a higher term short-term return and attracts increasing number of agents. Through the adjustment of instrument choices, the change in allocation is manifested as the motion of stress across agents. The asset bubble is then observed by the co-alignment of violations across sectoral agents. In this co-alignment, a number of agents make
“herding” allocation choices that involve instrument j. These choices may appear irrational from the longer perspective—violations of historically rational price and quantity bundles—but are supported by the short-term expectations. Thus, the motion in stress across instruments and agents is observed through propagation of apparent
23 violations from some sectoral agents to other sectoral agents. Put simply, asset bubbles travel through financial sectors inducing agents to realign their choices and resulting in relative adjustments of agent preferences toward other available bundles of instruments.
The apparent violations in price and quantity bundles of the bubble instruments lead to the relative valuation effects in other instruments. Thus, stress experienced in a particular sector agent or instrument can propagate.
24 Chapter 1: The Problem of Financial Stress in Adaptive System
1.1. Theoretical Framing
1.1.1. Issues framed by research in financial system complexity
The context of the economy as an adaptive16 complex system was pioneered by
Holland (1975, 1988) in his work on adaptive nonlinear networks and has been significantly extended in the past four decades. 17 Following Holland (1988: 117-118), the global economy forms an adaptive system through the following seven features: 1) interaction of many interdependent agents, 2) scarcity of global controls that allow competitive, as well as coordinated yet shifting agent associations, 3) multilevel hierarchical agent associations with asymmetric interactions across levels, 4) system adaptation through a continual recombination of agent interactions as the system accumulates experience, 5) the presence of niches exploitable by particular agent adaptations, 6) continuous creation of niches through technological innovation, and 7) suboptimal performance due to the continual thriving of niche interactions.
Nicolis and Prigogine (1977: 464) show that relative instability is a continuous dimension of adaptive systems. The main reason for the onset of an adaptive shift is that an adaptive “system is necessarily undergoing instabilities, and hence is capable of
16 Some researchers use the term self-organizing interchangeably with the term adaptive to emphasize the emergence of coordination among agents in the process of adaptation. In this study adaptive is preferred, as it refers to a more general set of interactions, including coordinating interactions. 17 See Arthur (1995) and Arthur et al. (1997) for applications of adaptive network modeling to financial markets. Brock and Hommes (1997, 1998) study financial markets as adaptive belief systems. Hommes (2001) extends this approach to markets as nonlinear adaptive evolutionary systems. See Aghion and Howitt (1992) and Howitt et al. (2008) for complexity-based macroeconomic models. See Farmer (2002) and Farmer et al. (2005) for complexity-based modeling of financial markets. See Beyeler et al. (2007), Bech and Atalay (2010), and Soramäki et al. (2007) for studies on topology and contagion in specific financial markets. See Farmer (1990) and Brock and Durlauf (2001) for critical methodologies.
25 amplifying certain disturbances including some of its own fluctuations.”18 Put differently,
a continuous state of relative financial instability is an integral aspect and an integral
problem of an adaptive financial system.
Figure 2 (Source: Oet et al., 2013) shows the conceptual model guiding this
research, supported in literature. The model suggests that macroprudential policy in
adaptive financial systems is necessarily a continuously changing process due to
perpetual financial system transformation. In this process, macroprudential policy is
repeatedly adjusted to suit its objectives through reconsideration of its functions, through
redesign of its forms, and through its methodological evaluation.
18 Nicolis and Prigogine (1977: 465). 26 Figure 2 Conceptual Model: Early_warning Policy Use in Adaptive Financial System
27
The conceptual model reflects the evolving relationship between the financial system and its risk policy objectives, functions, forms, and evaluation. Financial stability can be viewed as the ability to control one’s choices in adaptive systems in order to regulate preferential outcomes,
19 effectively placing financial stability within the risk management and prudential
purview. Borio (2003) and Nier (2011) consider the co-existence of microprudential
(aggregate risk is exogenous, independent of individual institution behavior) and
macroprudential (aggregate risk is endogenous, dependent on financial system behavior)
perspectives. This leads us to consider policy objectives in terms of the process through
which risk aggregates in the system over time and across the system participants.20 Policy
in the time dimension is concerned with aggregate risk evolution over time and the
adverse amplification between the financial system and the real economy
(procyclicality).21 Furthermore, the time dimension objectives form a dual set of long-
term and short-term goals. In the long run the objective is “to avoid macroeconomic costs
linked to financial instability,” while in the short run, the objective is to “limit financial
system-wide distress.”22 Similarly, policy in the cross-sectional dimension is concerned with a complementary dual set of issues: common exposures and interconnections among institutions. Accordingly, the cross-sectional macroprudential objectives include the common exposure imbalance-based goal to limit severity of failure, common exposure
19 Schinasi (2004: 8) proposes a related definition of financial stability as a continuous range where “A financial system … is capable of facilitating (rather than impeding) the performance of an economy, and of dissipating financial imbalances that arise endogenously or as a result of significant adverse and unanticipated events.” 20 IMF (2011: 8). 21 Borio (2003: 11). 22 Borio (2003: 2).
28 imbalance-based goal to limit probability of failure, and interconnectedness-based goal to strengthen infrastructure resilience.23
Macroprudential policy functions follow the consensus view expressed by Borio
(2006: 3413) that macroprudential policy contains two strategic dimensions: “first,
improving measurement of systemic conditions, and second focusing on limiting build-up
of imbalances.” The intrinsic functions involve identification of systemic conditions,
forward-looking and forecasting capacities, identification of systemic imbalances,
differentiation of excessive exposures, and sensitivity to systemic risk posed.
Policy forms follow the findings of Lim et al. (2011a, 2011b) and Galati and
Moessner (2013). Using established forms of macroprudential tools,24 Lim et al. (2011b)
find evidence that most tools are capable of reducing procyclicality, although their
usefulness “is sensitive to the type of shock facing the financial sector.” Specifically, they
propose that macroprudential efficacy is increased when implementation includes 1)
multiple tools, 2) targeted tools with higher ability to differentiate among exposures, 3)
time-varying tools that can be adjusted through the range of financial conditions, 4)
dynamic tools accompanied by clear rule-based communication, 5) tools that coordinate
well and reinforce associated policy initiatives.
Classic literature on policy evaluation suggests that in the context of the adaptive
financial system, the uncertainty in macroprudential policy can be addressed adaptively
(Lucas, 1976), incorporating the behavior of economic agents (Lucas, 1976; Sabatier,
1991), as a continuous dynamic process (Sabatier, 1991), and considering the
23 Cross institution differentiation between common exposures and interconnectedness is based on Bijlsma et al., (2010) who distinguish between direct and indirect interconnectivity mechanisms. 24 Including early warning systems, asset price models, stress testing, and microprudential feeds. 29 policymakers’ loss function (Brock et al., 2003) with the corresponding and ongoing
(Leeper and Sargent, 2003) robust analysis of model uncertainty space. Following Brock et al. (2003), the evaluation approaches can include expected loss calculations, model uncertainty aversion, local robustness analysis, and robustness with multiple models.
1.1.2. Issues framed by research in financial system stability
This work grounds on and is framed by the six theoretical and empirical elements developed in our prior qualitative and quantitative research on financial system stability.
First, we develop a methodology to measure overall financial system stress (Oet et al.,
2015a). Second, we enable the decomposition of the overall financial stress construct into its constituent factors across multiple markets of the financial system (Oet et al., 2015b).
Third, we demonstrate the existence of a significant association between institutional imbalances, system structure, and financial market stress and to explain this association
(Oet et al., 2013). Fourth, we discern agent structure within the banking sector (Oet et al.,
2016). Fifth, we determine the factors of financial intermediation within the interbank funding market (Oet et al., 2016). Sixth, we critically survey the feedbacks theory to develop a comprehensive understanding of the interaction of financial stress, financial instability, and the banking system factors (Oet and Pavlov, 2014).
Accordingly, there are three important epistemological aspects of the research problem that remain unknown and require discovery. First, we have not considered the implications of the fundamental problem of financial system complexity to financial stress measurement. Second, we lack an empirically-valid theory of stress measurement across multiple units of analysis in the context of an evolving financial system. Third, we do not understand the process by which to financial stress propagates across the financial
30 system and leads to critical states. These gaps in knowledge lead us directly to the formation of the specific research questions pursued in this study.
1.2. Research Precedents
1.2.1. Financial system stress construction
Building on the research precedent of Illing and Liu (2003, 2006), Oet and Eiben
(2009), and Oet et al. (2011), in a precedent-framing paper, Oet et al. (2015a) define systemic risk as a condition in which the observed movements of financial market components reach certain thresholds and persist. They develop the financial stress index in the US (CFSI) as a contemporaneous and continuous measure. The CFSI is created utilizing daily publicly observable data from the following components covering a wide spectrum of financial sectors (Table 2):(1) financial beta, (2) bank bond spread, (3) interbank liquidity spread, (4) interbank cost of borrowing, (5) corporate bond spread, (6) commercial paper–T- bill spread, (7) liquidity spread, (8) treasury yield curve spread, and
(9) stock market crashes, (10) commercial real estate spread, (11) residential real estate spread, (12) asset-based securitization spread, (13) commercial mortgage-backed securitization spread, (14) residential mortgage-backed securitization spread, (15) currency crashes, (16) covered interest spread. There are many weighting techniques available and utilized in indexing financial stress, such as equal weights, variance weights, principal component, and market size weights. Such techniques were tested in turn and the approach selected to minimize false alarms is a variation of a market size weight called the “credit weights” method. This method utilizes Flow of Funds data and measures the amount of credit outstanding in the four broad financial markets that make up the 11components. This allows for a dynamic weighting methodology where weights
31 change as conditions in financial markets shift (Oet and Eiben, 2009; Oet et al., 2011).
Bianco et al. (2012: 1–2) highlight that these components, mainly spreads, provide significant coverage of the US financial system markets. While stress in any of these markets could carry over into the broader financial system, the combined information contained in the stress components gains value as “systemic stress-related events are more likely to affect spreads in multiple markets. Observing conditions in a number of markets allows for the potential identification of a common factor, that is, financial stress.” The variables for each of the six financial markets and their construction are outlined in Table 2.
Table 2 Financial Stress Index Construction Financial Financial Calculation Notes Market Product Financial Beta | , | r is banking sector share prices (S&P 500 Financials), m is overall stock market (FB) | share prices (S&P 500), (t, t-1) are observations from time t to one year prior Bank Bond 10A refers to 10-year A-rated bank bond yields (Bloomberg C07010Y Index) and 10 10 Spread (BBS) 10TB refers to 10-year treasury yields Funding Interbank 3mo L is 3 month LIBOR rate and 3mo TB is 90-day treasury bill secondary Market Liquidity Spread 3 3 market rate (ILS) Interbank Cost of Borrowing 3 3mo L is 3-month LIBOR and FFR is the federal funds target rate Spread (ICOB) Corporate Bond 10CB is the 10-year Moody’s Aaa rated corporate bond yield and 10TB is the 10 10 Spread (CBS) 10-year treasury yield Commercial 90day CP 90-day is financial commercial paper (CP) rate and 3mo TB is 90-day Paper T-Bill 90 3 treasury bill secondary market rate Spread (CPTBS) Credit Market Liquidity Spread 1 moving average is calculated over the previous thirty trading days (LS) 30 2 Treasury Yield 1 thirty-day moving average of the difference between three-month t-bill yields Curve Spread 10 3 30 (3mo) on a bond equivalent basis with 10-year constant maturity yields (10yr) (TYCS) This is calculated for each of the nine subsectors of the S&P500 including Stock Market Equity Market consumer staples, consumer durables, energy, financials, health care, Crashes (SMC) ∈ | 0,1, … ,364 industrials, information technology, materials, and utilities. Commercial Real CMBS is the yield on commercial mortgage-backed securities and 7TB is the Estate Spread 7 yield on a 7-year treasury note Real Estate (CRE) Market Residential Real RMBS is the yield on agency residential mortgage-backed securities and 7TB is Estate Spread 7 the price of a 7-year treasury bond (RRE) ABS Spread ABS is the asset-backed bond yield (SYCAAB@USECON) and 5TB is the yield 5 (ABS) on a 5-year treasury note Securitization Commercial MBS CMBS is the yield on commercial mortgage-backed securities and 7TB is the 7 Market Spread (CMBS) yield on a 7-year treasury note Residential MBS RMBS is the yield on agency residential mortgage-backed securities and 7TB is 7 Spread (RMBS) the price of a 7-year treasury bond Where x is the exchange rate between the US dollar and a foreign currency. The Currency exchange rate is quoted such that it measures the price of one foreign currency Crashes (CC) ∈ | 0,1, … ,365 in USD. This calculation is performed for: AUD, CAD, EUR, GBP, JPN, MXN & ZAR. FX Market F is the 90-day forward USD-foreign currency exchange rate S* is the spot USD-foreign currency exchange rate Covered Interest ∗ r is the 90-day U.S. treasury bill rate 1 ∗ 1 Spreads (CIS) r* is the 90-day foreign treasury bill rate This calculation is performed for: Australia, Canada, Eurozone, Great Britain, Japan, Mexico and South Africa.
32 1.2.2. Stress factor decomposition
Oet et al., (2015b) extend recent research contributions to financial stress measurement (Hakkio and Keeton, 2009; Hatzius et al., 2010; Kliesen and Smith, 2010;
Brave and Butters, 2011; Oet et al., 2011; Holló, Kremer, and Lo Duca, 2012; Carlson,
Lewis, and Nelson, 2012; and Lo Duca and Peltonen, 2013) to allow a reliable decomposition of overall financial stress construct into component factors. They show that the deconstructed stress measure, adequately reflects the situation in core segments of the financial markets. Methodologically, in the formulation of financial stress the necessary elements include determining the financial system’s markets, the variables that describe market activity, and the transformation and aggregation of these variables.
Figure 3 displays the conceptual design of the stress construct.
Figure 3 Conceptual Model of Financial System’s Stress
Financial System Financial Markets Financial Products
Currency crashes Covered interest spread FX Market Financial beta Interbank liquidity spread Interbank cost of borrowing spread Funding Market Bank bond spread
Corporate bond spread Credit Market Liquidity spread Financial Treasury yield curve spread Stress CP T-Bill spread Real Estate Market Residential real estate spread Commercial real estate spread Equity Market Stock market crashes
Securitization RMBS spread Market ABS spread CMBS spread
The factor-based stress measure is constructed as a continuous stress variable,
using relative difference (spreads and spread-like) measures instead of volatility
33 measures. It is assumed that spreads will be specifically affected by increased uncertainty in the market. Conceptually, increased systemic stress should affect spreads in all markets. This means that a measure of underlying systemic financial stress has to consider spreads from a variety of different markets. By definition, we expect little correlation between the widening of spreads in separate markets if stress is non- systematic, whereas events due to systematic stress ought to affect spreads across multiple market sectors. Since spreads in each market carry some amount of market- specific idiosyncratic noise, considering aggregate spreads across different markets and over time would tend to reduce idiosyncratic noise. Put another way, considering multiple spreads in each market together reduces the likelihood of idiosyncratic causes spuriously commoving and increases the likelihood that spreads move due to a common factor, which can be interpreted as systemic financial stress.
Figure 4 CFSI Components Units of CFSI 90
75
60
45
30
15
0 1992 1995 1998 2001 2004 2007 2010 2013
Real Estate Market Funding Market Foreign Exchange Market Credit Market Equity Market Securitization Market
Figure 4 shows the movements of specific markets within the weekly CFSI, providing insight into the amount of stress that the six distinct markets contributed to the overall stress series. Measures from all markets tend to contribute significantly to the 34 overall financial stress. Their contributions in periods of financial stress tend to rise and fall together, amplifying overall changes on the financial stress.
Observations from individual components of the financial stress index offer insight into stress generation. Each panel of Figure 5 decomposes stress in a different market of CFSI. The first two panels of the figure document the components of the funding and credit markets. The funding market contributed the most to overall financial stress during the recent financial crisis. The contributions of the credit market to CFSI remain relatively constant with time. Panels C and D of Figure 5 decompose overall financial stress in the equity and foreign exchange markets. The equity market contributed most significantly to stress during the Dot-Com Bubble. Contributions from the foreign exchange market were largest more recently in conjunction with the European debt crisis. The final two panels of Figure 5 decompose stress in the securitization and real estate markets. Securitization markets contributed to stress most significantly during the financial crisis but have since been reduced.
35 Figure 5 Decomposition of Stress: Components of the Markets Panel A: Funding market Panel B: Credit market Panel C: Equity market
Units of CFSI Units of CFSI Units of CFSI 12 18 33
8 12 22
4 6 11
0 0 0 1992 1995 1998 2001 2004 2007 2010 2013 1992 1995 1998 2001 2004 2007 2010 2013 1992 1995 1998 2001 2004 2007 2010 2013 Corporate Bond Spread Liquidity Spread Consumer Durables Consumer Staples Financial Beta Interbank Cost of Borrowing Energy Financials Health Care Industrials CP - T-Bill Spread Treasury Yield Curve Spread Interbank Liquidity Spread Bank Bond Spread Information Technology Materials Utilities
Panel D: Foreign exhange market Panel E: Securitization market Panel F: Real estate market
Units of CFSI Units of CFSI Units of CFSI 15 27 15
10 18 10
5 9 5
0 0 0 1992 1995 1998 2001 2004 2007 2010 2013 1992 1995 1998 2001 2004 2007 2010 2013 1992 1995 1998 2001 2004 2007 2010 2013 Australia Canada Europe Great Britain Japan Mexico South Africa RMBS Spread ABS Spread CMBS Spread Commercial Real Estate Residential Real Estate
36 Chapter 2: Does Financial Stability Matter to the Fed in Setting the US Monetary
25 Policy?
The Taylor rule presents a traditional approach to guiding and evaluating contemporary monetary policy as a function of inflation and economic slack. While the responsibilities of the Federal Reserve (Fed) include price stability and long run growth, its mission has grown to include financial stability. Surprisingly, the question whether financial stability ought to be considered as part of monetary policy is hotly contested.
This study aims to determine whether policymakers’ discussions of financial stability and other factors systematically explain deviations of observed policy rates from the Taylor- rule-implied rates. To this end, we conduct content analysis of the Fed's monetary policy discussions to discover actual topics that enter into policy. There are two main findings: first, we find that discussion themes extracted from released Federal Open Market
Committee (FOMC) meeting minutes provide explanatory power beyond standard Taylor rule variables. Second, additional explanatory power is provided by a tri-mandate policy rule that accounts for changes in the economic and financial system as moderated by the evolving preferences of the policymakers. We show that a discussion-based thematic model with financial stability dominates Taylor-type rules during normal times.
Moreover, the tri-mandate policy model with financial stability dominates Taylor-type rules in zero lower bound conditions. Taken together, these findings reveal that financial stability has mattered to the Fed continuously and remains critical in setting monetary policy in zero lower bound.
25 An updated version of this chapter is forthcoming as Oet et al., (2016a). 37 2.1. Introduction
When the Fed was founded by Congress in 1913 as the central bank of the United
States, it was “to provide the nation with a safer, more flexible, and more stable monetary and financial system” (Board of Governors, 2013). The goals of US monetary policy are generally linked to the Fed’s responsibilities as defined by Congress in The Federal
Reserve Reform Act (1977: 1), where “The Board of Governors of the Federal Reserve
System and the Federal Open Market Committee shall maintain long run growth of the monetary and credit aggregates commensurate with the economy's long run potential to increase production, so as to promote effectively the goals of maximum employment, stable prices, and moderate long term interest rates.” Policymakers sometimes refer to the first two of these three areas, employment and prices, as Fed’s dual mandate. Throughout the Fed’s one-hundred-year history, some have viewed financial stability as a part of the
Fed’s task to promote stable prices and maintain long run growth. Others have taken a narrower view that price stability is distinct from financial stability.
It was not until The Wall Street Reform and Consumer Protection Act of 2010
(Dodd-Frank, 2010: Title VIII) that the Fed’s mission was formally expanded to include financial stability responsibilities. Bernanke (2011) affirmed that “The crisis has forcefully reminded us that the responsibility of central banks to protect financial stability is at least as important as the responsibility to use monetary policy effectively in the pursuit of macroeconomic objectives.” Most people agree that the mission of the Fed has clearly evolved with the transformation of the US financial system and currently includes
“maintaining the stability of the financial system and containing systemic risk that may arise in financial markets” (Board of Governors, 2013).
38 Paradoxically, despite the massive distress brought on by occasional financial shocks, the questions whether financial stability should be reflected, and how important it might be in a good monetary policy rule are hotly contested. Despite affirming the importance of financial stability, Bernanke (2011, 2012) still segments financial stability from monetary policy and provides no analysis about their interaction. For Bernanke, monetary policy tools are still firmly connected to macroeconomic stability while financial stability remains distinct and involves using non-monetary policy tools including liquidity provision, financial regulation, and supervision.26 In summary, financial stability is a stated mission and recognized responsibility of the Fed which
affects and is in turn dependent upon monetary policy; yet, monetary policy is often
considered too blunt an instrument to efficiently promote financial stability. Thus, the
question remains whether financial stability enters into the Fed’s deliberations and
influences its setting of monetary policy.
In this article, we pursue three positive questions. To what extent are changes in
monetary policy explained by the material considered and discussed during the Federal
Open Market Committee (FOMC) meetings? Do financial stability considerations matter to the Fed in setting the US monetary policy? Do discussion-based models with financial stability provide superior explanation of Fed’s monetary policy over time?
Our study makes two claims. First, we claim that the Fed’s monetary policy is explained significantly by the themes discerned from policymakers’ discussions after accounting for inflation and economic slack. Second, we claim that additional
26 Yellen (2011: 5) has similar reservations, emphasizing that “Monetary policy cannot be a primary instrument for [financial stability]. First, it has its own macroeconomic goals on which it must maintain sharp focus. Second, it is too blunt an instrument for dealing with [financial instability].” 39 explanatory power is provided by a tri-mandate policy model that accounts for changes in the economic and financial system as moderated by the evolving preferences of the policymakers. These claims are founded on the economic data characterizing monetary policy regimes (Appendix 1), content of policymakers’ discussions of monetary policy during the FOMC meetings from 1990 to 2012 (Appendixes 2 and Section 3), and regression results (Section 4).
To support the first claim, we connect five arguments. First, we demonstrate that our derived themes of discussion are valid (Appendix 3). Second, we show that the causal inference between FOMC discussions and monetary policy is partially supported (Section
3). Third, we show that the signs exhibited by the regressed themes meet theoretical expectations (Section 4). Fourth, we show that our results are consistent with authoritative prior investigations of individual effects (Section 4). Fifth, we show that the results are generalizable to provide insight on monetary policy even in zero lower bound conditions (Section 4). The Fed Funds rate serves as monetary policy instrument before
October 2008. Shadow short rate serves as monetary policy instrument starting in
October 2008.
We support our second claim by three arguments. First, we demonstrate that discussions of financial stability are representative of actual financial stability conditions
(Section 3). Second, we establish that the state of financial stability can be measured with
financial stress (Section 3). Finally, we confirm that discussion-based models with
financial stability dominate Taylor-type rules both during normal times and in zero lower
bound conditions (Section 4).
40 The rest of this chapter is structured as follows: Section 2.2 describes the conceptual framework for monetary policy. Section 2.3 examines the dataset of FOMC discussion themes and Taylor rules which frame this study in the context of five distinct monetary regimes. Section 2.4 hypothesizes, tests, and interprets the statistical and economic significance of the thematic and tri-mandate models. In the concluding Section 2.5, we discuss the impact of this research, including its counterarguments and limitations.
2.2. Conceptual framework
The studies of Taylor (1993) and Henderson and McKibbin (1993) famously expressed interest-rate instrument reactions to “the gap between actual and desired values” of output growth rates and inflation rates within a monetary policy feedback rule.
Because the Taylor rule has explained historical interest rates reasonably well using just these two factors it is frequently used to judge monetary policy decisions as good, when the actual interest rates fit the rule, and bad, when they do not. Since output is mediated by employment and inflation by prices, Taylor’s rule parallels the so-called “dual mandate” interpretation of the Federal Reserve System’s mission to maintain maximum employment and stable prices.
We suggest that the themes discussed during FOMC meetings are relevant to monetary policy alongside current or recent economic observations such as output and inflation, in addition to which they serve as component channels through which monetary policy shapes desired economic outcomes.
Two perspectives can be distinguished in the relationship of monetary policy with economic conditions: one formed by observations and another framed by expectations.
The first, ex-post, perspective drives the intentions reaction function (Khoury, 1990: 27)
41 of monetary policy, defined as “linking the movements in real GDP and inflation… to the short-term interest rate (Taylor, 1995: 14).” The second, ex-ante, perspective motivates the impact reaction function (Khoury, 1990: 27), defined as “the linkage from short-term interest rates to exchange rates and long-term interest rates, and finally to real GDP and inflation” (Taylor, 1995: 14). As Khoury (1990: 27-28) notes, “Much of the disagreement about the [monetary policy] results from a failure to distinguish between these two types of reaction functions.”
Taylor (1995: 14) further emphasizes the link between these two perspectives in that “the links of the monetary transmission actually form a circle, [beginning with the intentions function] and with the circle being closed by [the impact function].” Literature refers to this circular process as a feedback rule and defines it as “the monetary transmission mechanism: the process through which monetary policy decisions are transmitted into changes in real GDP and inflation” (Taylor, 1995: 11).
Khoury (1990) reviews some 42 distinct impact function models that explain a variety of monetary policy variables27 including combinations of explanatory variables
such as the unemployment rate, inflation rate, balance of trade, interest rate, exchange
rate, deficit, debt, and growth. Mishkin (1995) includes financial stability implicitly
through a set of relative asset price channels. For the intentions function, many
alternative explanatory variables have been proposed using a variety of additional factors
ranging from asset and oil prices to financial imbalances and the federal debt. Clerc and
Pfister (2003: 192) classify the financial factors within the transmission channel into two
27 Khoury (1990, Appendix Table 3.1) summarizes the impact reaction functions using some twenty-five distinct dependent variables from M1 to Federal-funds rate. 42 groups: “the first includes asset prices (shares, property); the second is derived from the existence of an external funding premium and credit constraints; this category is at the origin of the credit cycle.” Borio and Zhu (2012) (BZ) discuss additional financial channels in the transmission mechanism including among others: balance sheet channel, bank lending channel, interest rate channel, bank capital channel, and risk-taking channel.
In particular, BZ (p. 237) argue that “significant aspects of the overall shape of the transmission mechanism can potentially be missed if the risk-taking channel is not incorporated in the central bank’s reaction function. The argument is that there is an endogenous interaction between the reaction function and the cumulative strength and shape of the transmission chain.”
To the extent that financial stability has been considered in the monetary reaction function, it is involved as a factor of financial intermediation by which innovations in monetary policy become real. Thus, financial stability is reflected as input to inflation.
This point of view is captured by Bernanke and Gertler (1999: 18) who conclude that
“the inflation targeting approach dictates that central banks should adjust monetary policy actively and preemptively to offset incipient inflationary and deflationary pressures.
Importantly, for present purposes, it also implies that policy should not respond to changes in asset prices, except insofar as they signal changes in expected inflation.”
Cecchetti et al. (2005: 428) emphasize the positive nature of their dissent, showing that
“central banks can improve macroeconomic performance by reacting to asset price misalignments… [and not saying] that policymakers should target asset prices.” The shift towards embracing the importance of financial stability is expressed by Bernanke (2011) and FOMC (2013) who imply the need to include financial stability in the Fed’s reaction
43 function by noting that “… the Committee should explore ways to calibrate the magnitude of the risks to financial stability so that those considerations could be more fully incorporated into deliberations on monetary policy.”
If financial stability is on par with the standard Taylor rule factors of output and inflation, does this in fact change the reaction function of the Fed? Given the increased awareness of the importance of financial stability, does this have implications for monetary policy? Building on this conceptual foundation, this study applies qualitative content analysis to examine a set of positive questions about the relevance of financial stability to the Fed in setting monetary policy.
A loosely similar textual analysis approach has led Hayford and Malliaris (2004:
387) to “find no empirical evidence that the Federal Reserve policy attempted to moderate [financial stability] during the late 1990s.”28 This result contrasts with Baxa et
al. (2013: 117), who find that in the last thirty years central banks tend to change
monetary policy rates nonlinearly in response to financial instability: “When a financial
system is stable … financial stability considerations enter monetary policy discussions
only to a limited degree”; yet, when financial instability is tangible (e.g. evidenced by financial imbalances), “central banks often change policy rates, mainly decreasing them in the face of high financial stress.” We will examine whether analysis of FOMC discussions supports the absence or presence of financial stability considerations in setting the US monetary policy.
28 Hayford and Malliaris (2004) focus their study on stock market valuations during the Greenspan era. 44 2.3. Data and methodology
According to Krippendorff (1989: 403), “content analysis is a research technique for making replicable and valid inferences from data to their context.” Stemler (2001) defines it as a systematic methodology “for compressing text into content categories based on explicit rules of coding.” This study applies Krippendorff (2012) methodological recommendations for a systematic and replicable application of six stages of content analysis: design, unitizing, sampling, coding, analysis, and validation.
From the design perspective, the importance of analyzing the text of the FOMC meetings, where the policy is being debated and formed, is self-evident. Here, the strength of textual analysis as a research method lies precisely in its ability to uncover how the Fed makes decisions and what topics enter as elements into the monetary policy considerations. Its principal challenge is the empirical support for inferential validity of its results. Its principal limitation rests on the fundamental assumption that the frequency of textual occurrence of a certain idea reflects the relative importance of this idea among others. Among two methods of textual analysis, latent semantic analysis (LSA) and content analysis, we chose the latter, because it allows a dual exploratory and algorithmic approach, enabling methodological cross-validation. By contrast, the alternative LSA method (Deerwester et al., 1990) is purely an automated dimension reduction technique.
In addition, as shown by Boukus and Rosenberg (2006), LSA reflects FOMC theme correlations through component cross-loadings, complicating interpretation of factors and reducing the possibility of parsimonious factor analysis.
Our data consists of the FOMC meeting minutes. To enable cross-validation, we chose two units of analysis: paragraph for the initial exploratory analysis of discussion
45 themes and phrase for the subsequent algorithmic analysis.29 To establish the discussion
themes, we sample the FOMC discussions in their modern form which began in 1993
(Danker and Luecke, 2005). For the algorithmic analysis, we extend the sample to span
the period from February 1990 to July 2012 to align the sample with the supported proxy measure of financial system stability (Oet et al., 2012). Appendix 3 outlines the data preparation stages. Appendix 4 details the validity stage, directed to the epistemological challenge of recognizing meaning which lies at the core of content analysis as a scientific method (Krippendorff, 2012). The dataset is summarized below.
2.3.1. Content analysis: FOMC discussions of monetary policy
Significance of discussion themes. The dataset revealed by the analysis of FOMC discussions displays surprising variability in the composition and relative importance of themes. Figure 6 presents the situational map (Clarke, 2003) for themes that emerged during the exploratory analysis of FOMC discussions at the paragraph level.
29 Appendix 3 provides additional details for these two methods known respectively as emergent coding and selective coding. 46 Figure 6 Situational Map of Monetary Policy Themes OUTPUT FINANCIAL SYSTEMS STABILITY Consumption (household consumption); Retail sales/ Durable Goods (Motor United States financial regulation system vehicles and parts, Home appliances) / Non-Durable Goods (Food, Clothing, Regulatory authorities (US Securities and Exchange Commission (SEC), Oil/gasoline, Pharmaceuticals); Financial Industry Regulatory Authority (FINRA), Commodity Futures Trading Services; Commission (CFTC), Federal Deposit Insurance Corporation (FDIC) Household saving rate; [(Liquidity Guarantee Program (TLGP)], Office of the Comptroller of the Investment / Business investment (Investment in equipment and software, Currency (OCC), National Credit Union Administration (NCUA), Office of Other business investment, Business sentiment) / Construction (Residential Thrift Supervision (OTS), Consumer Financial Protection Bureau (CFPB), construction, Housing starts, Non-residential (commercial) construction) / Federal Reserve [Term Asset-Backed Securities Loan Facility (TALF), Changes in inventories; Troubled Asset Relief Program (TARP), Term Securities Lending Facility Government Spending; (TSLF), Term Auction Facility (TAF), Term Deposit Facility (TDF), Primary Net Export (Imports, Exports, US trade deficit); Dealer Credit Facility (PDCF), Asset-Backed Commercial Paper Money Industrial production and capacity utilization (Manufacturing; Mining; Utilities; Market Mutual Fund Liquidity Facility (AMFL), Commercial Paper Funding Purchasing managers index) Facility (CPFF), Money Market Investor Funding Facility (MMIFF), Supervisory Capital Assessment Program (SCAP), Asset-Backed EMPLOYMENT Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF)]; Hiring (Nonfarm payroll employment); Unemployment (Unemployment rate, Treasury; Government National Mortgage Association (GNMA); Fed’s target unemployment rate, Unemployment insurance claims); Congressional regulatory laws [Financial Svc Modernization Act ’99, Banking Compensation/wages/salaries; Labor force participation Act ’33] Foreign central banks INFLATION European Central Bank (ECB); Bank of England; Bank of Japan; Bank of Consumer prices, Consumer Price Index (CPI); Survey of Consumer Canada; Swiss National Bank; Bank of Mexico Confidence Sentiment; Producer prices; Personal Consumption Expenditures Price Index (PCE) IDIOSYNCRATIC EVENTS FOREIGN ECONOMIC ACTIVITY Hurricanes (Hurricane Katrina, Hurricane Sandy, Unidentified hurricane); Europe (after 1999: Euro area) (Spain, Germany, Portugal, Italy, France, Drought; 2011 Tohoku earthquake and tsunami; Unusual temperatures Greece, European sovereign debt crisis, United Kingdom (UK); (Unusually warm weather, Unusually cold weather); 2001 terrorist attacks Asia (1997 Asian financial crisis, China, Japan, Collapse of the Japanese asset price bubble (1990), India, Korea, Singapore, Indonesia, Malaysia, LOGISTICS Hong Kong); Attendees; Disclosure policy / communications; Implementation of monetary Latin America (Latin American debt crisis, Brazil, Mexico); policy (Short term rates [Open market operations, System open market Canada account (SOMA), Federal reserve asset purchases, Discount rate/target federal funds rate, Interest on excess reserves (IOER), Reserve FISCAL POLICY requirements], Long term rates) Federal budget; Taxes; Fiscal stimulus; Federal spending sequester
MONEY SUPPLY M1; M2; M3
The descriptive statistics from exploratory analysis (Table 3, Panel A) provide evidence for six of the nine major themes from Figure 6. On average, the output theme is the most significant, with 40% of all themes, followed by financial stability (28%), inflation (16%), employment (10%), money supply (4%), and fiscal policy (1%). An additional theme of foreign activity arises through algorithmic analysis (Table 3, Panel
B). The table includes stationarity statistics of the significant themes for both types of content analysis. Tested in level form, only the financial stability and output themes do not exhibit a stationary process. After differencing, all the series exhibit stationary processes at the 1% level, indicating that the FOMC did not drastically change the content of their discussion from one meeting to another. Rather, the change in the importance of each theme from meeting to meeting is gradual. The differences between
47 the relative importance of the dominant themes in exploratory analysis and algorithmic analysis are statistically significant.
Table 3 Descriptive Statistics of Themes Discussed at FOMC Meetings (1990–2012) Financial Money Panel A: Exploratory analysis, 1993–2012M06 Output Inflation Employment Fiscal policy stability supply
Observations 21 21 21 21 21 21 Mean 28.1% 40.2% 16.3% 9.7% 4.4% 1.3% Standard deviation 11.1% 10.5% 3.5% 2.5% 1.8% 1.5% Median 23.8% 41.6% 15.7% 9.7% 4.3% 0.8% Maximum 50.0% 55.2% 24.0% 16.3% 7.2% 5.9% Minimum 15.2% 20.8% 11.0% 5.9% 0.4% 0.0% ADF T-Statistic (Levels) 0.14 -0.79 -3.00* -3.31** -2.08 -2.26 ADF T-Statistic (1st Difference) -6.48*** -4.07*** -5.30*** -6.20*** -3.99*** -5.83*** Financial Money Foreign Panel B: Algorithmic analysis, 1990M02–2012M06 Output Inflation Employment Fiscal policy stability supply activity Observations 269 269 269 269 269 269 269 Mean 22.3% 40.3% 19.5% 11.7% 1.7% 0.6% 4.0% Standard deviation 8.2% 8.8% 5.6% 2.6% 3.4% 0.7% 1.3% Median 19.5% 41.9% 17.9% 11.5% 0.1% 0.3% 4.0% Maximum 45.6% 56.7% 38.4% 19.5% 14.2% 2.8% 8.4% Minimum 10.1% 14.7% 9.2% 4.1% 0.0% 0.0% 1.3% ADF T-Statistic (Levels) -2.48 -2.71* -2.65* -4.33*** -3.39** -3.73*** -1.65 ADF T-Statistic (1st Difference) -11.19*** -10.95*** -15.57*** -7.71*** -5.13*** -10.20*** -5.98*** Note: Results of emergent coding at paragraph unit of analysis are summarized in annual statistics, whereas results of selective coding at phrase unit of analysis are summarized in monthly statistics. The units of summary statistics are relative percentage of theme usage per observation. p < 0.01***; p < 0.05**; p < 0.10*.
Moreover, considering the relative weights of dominant FOMC discussion themes for both types of content analysis (Figure 7), we find that the portion of each meeting’s discussions ascribed to each theme changes over regime samples (Appendix 1).30
Exploratory analysis (Figure 7, Panel A) suggests an overall pattern of increased importance of financial stability associated with a decline in the importance of output.
Algorithmic analysis (Figure 7, Panel B) suggests a richer variation, supporting a dominant role for financial stability in the first regime, its relatively constant importance in regimes 2, 3, and 4, and its substantial rise in importance during regime 5 coinciding with the Great Recession. This data suggests that the themes of inflation and financial
30 As described in Appendix 1, structural break testing of the effective Fed Funds rate time series identifies four structural breaks (January 1994, January 2001, July 2004, and February 2008) that define the following regime samples: regime 1 (1990M21993M12), regime 2 (1994M12000M12), regime 3 (2001M12004M6), regime 4 (2005M72008M1), and regime 5 (2008M22012M6).
48 stability tend to move in opposite directions of importance. During the last two regimes, their relative weights have separated distinctly for the first time since the early 1990s.31
Figure 7 Relative Theme Importance over Time Panel A: Exploratory analysis (1993M1-2012M7) Panel B: Selective Coding (1990M2-2012M6) Regime 2 Regime 3Regime 4 Regime 5 Regime 1 Regime 2 Regime 3 Regime 4 Regime 5 60% 60%
50% 50%
40% 40%
30% 30%
20% 20%
10% 10%
0% 0% Feb-90 Feb-91 Feb-92 Feb-93 Feb-94 Feb-95 Feb-96 Feb-97 Feb-98 Feb-99 Feb-00 Feb-01 Feb-02 Feb-03 Feb-04 Feb-05 Feb-06 Feb-07 Feb-08 Feb-09 Feb-10 Feb-11 Feb-12 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Financial stability Output Financial stability Output Inflation Employment Inflation Employment Money supply Fiscal policy Money supply Fiscal policy
Cross-correlogram data between the FOMC discussions and the monetary policy provides additional information on the pattern of their relationships. As shown in Figure
8, Panel A, over time, the cross-correlations between discussions of financial stability and monetary policy exhibit asymmetric feedback between the leads and lags of the changes in financial stability discussions and changes in monetary policy. Consider the intentions function correlogram ⋆ between lagged discussion of financial stability and the ensuing monetary policy . A rise in financial stability (↗ , measured by a drop in financial stability discussions (↘ ), tends to be associated with subsequent tightening of monetary policy (↗ increases of Fed Funds rate) with a convex cross- correlation reaching -21% one year later. However, as shown by the impact function
31 This changing data pattern supports the idea that the key themes in the FOMC discussions shift meaning. This suggests the potential relevance of Blumer’s (1969) “symbolic interactionism”—the idea that the Fed’s policymakers continue to shape monetary policy in response to the evolving meaning from relevant economic factors, their own actions in response to this information, and through the discussions and interaction with their peers on the FOMC. 49 correlogram ⋆ between lagged monetary policy instrument and the ensuing discussion of financial stability , tightening monetary policy (↗ ) tends to be associated with subsequent financial instability (↘ ) reflected in increased
discussions ↗ ) with a concave cross-correlation reaching 10% one year later:
↗ ↘ ↗ | ⋆ ≅ 21% (1) ↗ ↘ ↗ | ⋆ ≅ 10%
Put another way: a decrease in the Fed funds rate tends to be associated with
subsequent increases in financial stability, reflected in decreasing discussions of this topic
(impact function). However, a rise in financial stability tends to lead to a tightening
policy (intentions function), which in turn tends to lead to financial instability. The data
shows that pattern is asymmetric, where the cross-correlations of increasing financial
stability discussions with the subsequent loosening of monetary policy are stronger than the cross-correlations of a tightening policy with subsequent discussions of financial
stability. Of course, correlation is not causation, and a stronger understanding of the interaction between financial stability and monetary policy is needed to obtain further insights. For considerations of brevity, we follow the notation introduced above to summarize the remaining correlograms in Figure 8, Panel A for the period from 1990M2 to 2008M1. This sample includes the first four regime samples, using effective Fed Funds rate as MPI. It excludes regime 5 that is characterized by zero lower bound and the loss of normal functionality of the Fed Funds rate as MPI (see section 3.2). The correlograms
for the FOMC discussions of inflation ( ), output ( ), employment ( ), foreign
activity ( ), fiscal policy ( ), and money supply ( ) show that the policymakers’
50 talk tends to have a tangible asymmetric and frequently nonlinear relationship with monetary policy:
↗ ↗ | ⋆ ≅ 31% (2); ↗ ↘ | ⋆ ≅ 44%
↗ ↘ | ⋆ ≅ 17% (3); ↘ ↗ | ⋆ ≅ 27%
↗ ↘ | ⋆ ≅ 11% (4); ↘ ↗ | ⋆ ≅0% ↘
↗ ↘ | ⋆ ≅ 11% (5); ↘ ↗ | ⋆ ≅ 9% ↘
↘ ↗ | ⋆ ≅ 43% (6); ↗ ↗ | ⋆ ≅ 38%
↗ ↗ | ⋆ ≅ 1% ↘ (7). ↗ ↘ | ⋆ ≅ 48%
Further support to the causal interpretation for the FOMC thematic content is
provided by the significant one-way Granger causality from the FOMC discussions of
financial stability, inflation, and employment to the monetary policy instrument (Table
4). Analysis of mutual precedence patterns of the discussions of financial stability and
monetary policy shows that, over the sample period, FOMC discussions of these topics
unidirectionally Granger-cause the subsequent monetary policy actions. Put differently,
these findings support the ideas that through the sample period from 1990M2 to 2012M6:
1) Financial stability entered the policymakers’ intentions function;
2) Discussions of inflation and slack in employment entered the policymakers’
intentions function;
51 3) Discussions of monetary aggregates entered the policymakers’ intentions
function;
4) Monetary policy had an impact on the ensuing fiscal policy, but not
conversely.
52 Figure 8 Cross-correlograms of intentions and impact functions for FOMC discussions and monetary policy (1990M2–2008M1) 0.6 52%
0.4
Financial 0.2 7% stability 0
-0.2 0 3 6 9 1215182124 0.5 31%
0
Inflation -0.5 -44%
-1 0 3 6 9 12 15 18 21 24 0
-0.1
-0.2 Output -17% -0.3 -27% -0.4 0 3 6 9 1215182124 0.3 0.2 0.1 0% Employment 0 -0.1 -0.2 -11% 0 3 6 9 1215182124 0.3 0.2 0.1 Foreign activity 0 -0.1 -9% -10% -0.2 0 3 6 9 1215182124 0
-0.2
Fiscal -0.4 policy -38% -0.6 -43%
-0.8 0 3 6 9 1215182124 0.6 48%
0.4
Money 0.2 supply 0 -1% -0.2 0 3 6 9 1215182124 Note: indicates impact function cross-correlation between lagged monetary policy instrument in Panel A (lagged manifest variable in Panel B) and ensuing discussion of a particular theme; indicates intentions function cross-correlation between lagged discussion of the corresponding theme and the ensuing monetary policy instrument in Panel A (lagged manifest variable in Panel B). The period from 1990M2 to 2008M1 includes 4 regime samples with effective Fed Funds rate used as MPI and excludes regime 5 which is characterized by zero lower bound when monetary policy instrument is better represented by average short shadow rate.
53 Table 4 Granger Causality of FOMC Discussions and Monetary Policy (1990M2–2012M6)