IFTA IFTA

Journal Journal A Professional Journal Published by The International Federation of Technical Analysts 2013 Edition

13

Inside this Issue 6 The Implied Projection Range (IVPR); Extending the Statistical and Visual Capability of the VIX

13 Psychological Barriers in Asian Equity Markets

20 Regime-Switching Trading Bands Using A Historical Simulation Approach

In the business world, the rearview mirror is always clearer than the windshield. —Warren Buffett The Future of has arrived.

Discover a whole new world of possibilities at www.mav7.com/ifta IFTA JOURNAL 2013 EDITION

Letter From the Editor By Rolf Wetzer, Ph.D...... 5

Articles EDITORIAL Rolf Wetzer, Ph.D. (SAMT) The Implied Volatility Projection Range (IVPR); Extending the Statistical and Visual Capability Editor, and Chair of the Editorial Committee of the VIX [email protected] By Mohamed El Saiid, CFTe, MFTA...... 6 Aurélia Gerber (SAMT) [email protected] Psychological Barriers in Asian Equity Markets Editor By Shawn Lim, CFTe, MSTA, and Bryan Lim...... 13 Ralf Böckel, CFA (VTAD) [email protected] Editor MFTA Papers Michael Samerski Regime-Switching Trading Bands Using A Historical Simulation Approach [email protected] Editor By Ka Ying Timothy Fong, CFTe, MFTA...... 20 Mark Brownlow, CFTe (ATAA) Using a Volatility Adjusted Stop Loss (VASL) to Enhance Trading Returns [email protected] Editor By Edward Rowson, CFTe, MFTA...... 28

Send your queries about advertising Indicators: An Empirical Analysis of the Concept of Divergences information and rates to [email protected] By Stephan Belser, CFTe, MFTA...... 35

Wagner Award Paper Momentum Success Factors By Gary Antonacci...... 45

Educational Heikin-Ashi: A Better Technique to Trends in Noisy Markets By Dan Valcu, CFTe...... 54

Book Review About cover photo: Abstract Waves—Photo by piccerella Mastering Market Timing, Using the Works of L.M. Lowry and R.D. Wyckoff to Identify Key Market Turning Points By Richard A. Dickson and Tracy L. Knudsen Reviewed by Regina Meani, CFTe...... 60 Author Profiles...... 61 IFTA Board of Directors...... 62

IFTA Journal is published yearly by The International Association of Technical Analysts. 9707 Key West Avenue, Suite 100, Rockville, MD 20850 USA. © 2012 The International Federation of Technical Analysts. All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying for public or private use, or by any information storage or retrieval system, without prior permission of the publisher.

IFTA.ORG PAGE 3 IFTA2013 26TH ANNUAL CONFERENCE San Francisco Check the website for updates: www.ifta.org IFTA JOURNAL 2013 EDITION

IFTA2013 Letter From the Editor TH 26 ANNUAL CONFERENCE By Rolf Wetzer, Ph.D.

Dear IFTA Colleagues and Friends:

In all our writings and teaching on technical analysis, we never forget to quote our mantra— history repeats itself. If we go back 100 years and look at the Dow Jones Industrial Index, it seems that this is a good assumption. In 1912, the first quarter started with an upsurge. In Spring, the market went sideways, spiced with good volatility. In Summer, the market rose to a new high. Does this sound familiar? In all our writings The International Federation of Technical Analysts’ (IFTA) Journal is traditionally international, both in its contributions and teaching on and techniques described. This year, the Journal has four separate sections. In the first section, two articles were submitted by IFTA colleagues from the technical analysis, Egyptian Society of Technical Analysts (ESTA) and the Society of Technical Analysts (STA). ESTA’s contribution covers volatility bands based on the VIX Index. STA offers we never forget to information on a strict testing procedure for psychological barriers in Asian stock markets; as our 2012 Annual Conference will be in Singapore, this article might offer quote our mantra— additional insight into some of the Conference topics. In the second section, there are three papers from our Master of Financial Technical history repeats itself. Analysis (MFTA) program. One of the authors, Stephan Belser, MFTA, was awarded the John Brooks Memorial Award for 2011. Congratulations! This year, aside from our own MFTA material, we are happy to publish a paper from another organisation. With the permission of the National Association of Active Investment Managers (NAAIM), we have included a paper by Gary Antonacci, winner of the NAAIM Wagner Award 2012. We hope that you find this paper informative. We conclude with contributions from two of IFTA’s current Board members. Dan Valcu, CFTe, IFTA’s Membership Director and Vice President of Europe, has written an article on a Japanese charting technique. IFTA’s former Journal Editor and Director, Regina Meani, CFTe, contributed a book review. Again, like IFTA itself, the Journal is truly international. I would like to thank the authors for their contributions. And, not only do Journal articles come from all over the globe, our editors do too. I would like to thank Aurélia Gerber, Ralf Böckel, Michael Samerski and Mark Brownlow for their help in editing this Journal. Last but not least, I would like to thank Linda Bernetich and Jon Benjamin for the layout and their effort in putting the Journal together. We are now in the fifth year of a financial crisis. In 1912, the year ended with a drawdown. Hopefully, as good market technicians, we should be prepared this time.

IFTA.ORG PAGE 5 IFTA JOURNAL 2013 EDITION

The Implied Volatility Projection Range (IVPR); Extending the Statistical and Visual Capability of the VIX By Mohamed El Saiid, CFTe, MFTA

1. Abstract SPX and the VIX movements. According to Whaley, the latter This paper proposes a new method for extending the feature is brought about by portfolio insurance. statistical and visual capability of the Chicago Board Options Exchange (CBOE) market volatility index (VIX) over the S&P 500 To support Whaley’s argument, we have performed a series (SPX). The paper subsequently provides several visual examples of linear correlation tests between the SPX and the VIX over of the proposed method and discusses its implications and uses 5,520 daily closing values (from January 1990 to December 2011). over the chart from a technical standpoint. Finally, the paper The first correlation test performed was conditional exclusively discusses implementing the same methodology on other implied on positive SPX returns. The second test was conditional volatility-based indices over their corresponding/underlying only on negative SPX returns. The third to seventh tests were equity market indices. The methodology proposed in this conditional on achieving returns greater than +/-0.50%, paper will be referred to hence forth as the “Implied Volatility +/-1.00%, +/-2.00%, +/-3.00% and +/-4.00% in the SPX. The Projection Range” or “IVPR”. outcomes of these tests were then compared to a final non- conditional correlation test between the SPX and the VIX and 2. Introduction the results are shown in table 1.

2.1 The VIX definition Table 1: Non-conditional vs. conditional correlation results between The Chicago Board Options Exchange (CBOE) market the SPX and the VIX volatility index (VIX) is a forward-looking index of the expected Correlation . return volatility of the S&P 500 index (SPX) over the next 30 Correlation test CoefficientR) ( days. This forward-looking feature is implied from the at- Non-conditional correlation -0.70 the-money SPX option prices. As such, the VIX measures the volatility that investors expect to see, rather than what has Conditional Correlation Tests been recently realized.1 Positive SPX returns correl. -0.46 The VIX estimates the expected volatility by averaging the Negative SPX returns correl. -0.60 weighted prices of the SPX puts and calls over a wide range of strike prices.2 In that respect and as opposed to equity Greater than +/-0.5% SPX returns correl. -0.76 market indices which are comprised of stocks, the VIX index Greater than +/-1% SPX returns correl. -0.79 is comprised of options, where each option price is intended to Greater than +/-2% SPX returns correl. -0.82 reflect the market’s expectation of future volatility.3 Greater than +/-3% SPX returns correl. -0.84 The VIX was initially developed by Prof. Robert E. Whaley in 1993 and is a registered trademark of the CBOE. Since then, Greater than +/-4% SPX returns correl. -0.87 several modifications were introduced in its calculations.4 From the results presented above we can make the following 2.2 The VIX implications (indications) over the SPX deductions: In his writings, Whaley discussed that while volatility implies unexpected market moves regardless of direction, the VIX is ƒƒ A negative correlation does exist between the SPX and the VIX dubbed as the “investor fear gauge”. He justified this to be on the regardless of the correlation conditions. back of the current domination of the SPX option market by the ƒƒ The negative correlation appears stronger at (R) conditional hedgers. Hedgers’ demand on puts tends to increase when there to negative SPX returns (-0.6) than at (R) conditional to is a concern for a potential decline in the stock market. Once positive SPX returns (-0.46). This reflects the asymmetry of that concern manifests, the VIX values tend to increase. Whaley movements of both indices as a result of portfolio insurance. coined this feature as “portfolio insurance”. As such, the VIX ƒƒ Negative correlation increases or becomes more significant as indicator tends to reflect the price of portfolio insurance.5 the SPX returns become more volatile. In other words, the VIX Attempting to prove his argument, Whaley’s tests values become more significant as the SPX daily returns surge established the following:6 and/or plunge.

ƒƒ Small SPX daily changes result in negligible VIX changes. 2.3 How the VIX is currently being used ƒƒ Volatility tends to follow a mean-reverting process. The VIX generally exhibits two strong characteristics. One ƒƒ There is an inverse, yet asymmetric relationship between the being a consistent negative correlation with the SPX, while the

PAGE 6 IFTA.ORG IFTA JOURNAL 2013 EDITION

other is the strong tendency for the VIX to revert to its long 3. Statistical interpretation and term mean, thus reflecting the absence of deterministic growth inferences of the VIX values in volatility (unlike stocks).7 According to these two features, the common interpretations of the VIX values (as a standalone 3.1 Statistical interpretation index) include the following: The VIX calculation produces a probability-based interpretation with respect to the estimated range of the SPX ƒƒ Abnormally high VIX readings imply a potential bottom, rates of returns over the next 30 days.8 or the occurrence of a counter-trend rally in the SPX. Example: Assume that the SPX closed at 1,300.00 and the The opposite is true at relatively low VIX readings. VIX closed at 25.00 today. Since the VIX values are annualized ƒƒ On the other hand, an up trending VIX implies a potentially values multiplied by 100, to transform the sustainable downtrend in the SPX and vice versa. value back to represent the 30 days (or 1 month) sigma, we divide 25 by 100 and then divide the outcome by the square Figure 1 visually depicts both features of the VIX; the tendency root of 12.9 to revert in an oscillatory-type motion despite experiencing In this example the result was 7.22%. According to the trending phases in the process, as well as the negative correlation statistical “Empirical Rule”, this is interpreted as follows: there with the SPX. This is evident by the VIX peaks and valleys that is a probability of 68.2% (approximately) that the expected coincide with key SPX lows and highs, respectively. In this chart range of the SPX returns over the next 30 days will lie within we have indicated relatively high and low VIX readings with +/-7.22%, or within the range of (1,206.18 – 1,393.82). reference to the 30% and 16% levels, respectively, based on visual inspection over the period presented. A common strategy among 3.2 Statistical inferences traders when using the VIX is to go long on equities when the VIX Testing the VIX interpretation according to the Empirical Rule rebounds from key highs and sell (or short) equities when the VIX and using historical daily closing values for the VIX from January rebounds from key lows. 1990 to December 2011, we present the outcomes in table 2.

Figure 1: Upper window: The SPX index—Daily values——Normal scale Lower window: The VIX index—Daily values—Candlestick chart—Semi-logarithmic scale

1650 1600 1550 Periods marked in red indicate key highs on the 1500 SPX which coincided with VIX lows as identified 1450 by the 16% level 1400 1350 1300 1250 1200 1150 1100 1050 1000 950 900 850 Periods marked in blue indicate key lows on the 800 SPX which coincided with VIX highs as identified 750 by the 30% level 700 650 600 100 90 80 70

60

50

40

30

20

ulAug SepOct NovDec 2008 MarApr MayJun JulAug SepOct NovDec 2009Mar AprMay JunJul AugSep OctNov Dec 2010Mar AprMay JunJul AugSep OctNov Dec2011Mar Ap rMay JunJul AugSep Oct

IFTA.ORG PAGE 7 IFTA JOURNAL 2013 EDITION

Table 2: Statistical results TableFigure 3: Calculating 2 introduces the SPXthe IVPRprojection when range visually based plotted on the VIXover values the

Standard Date SPX Close VIX close VIX adj to Lower Upper IVPR 0.5 1 1.5 2 2.5 3 30 days IVPR Deviation Volatility Inside 2,417 4,177 5,111 5,395 5,472 5,498 05/25/2010 1,074.03 34.61 9.99% VIX range 05/26/2010 1,067.95 35.02 10.11% 43.8% 75.7% 92.6% 97.7% 99.1% 99.6% 05/27/2010 1,103.06 29.68 8.57% Outside 3,104 1,344 410 126 49 23 VIX range 05/28/2010 1,089.41 32.07 9.26% 56.2% 24.3% 7.4% 2.3% 0.9% 0.4% 06/01/2010 1,070.71 35.54 10.26% Outside lower 06/02/2010 1,098.38 30.17 8.71% 1,060 494 221 112 47 23 VIX range 06/03/2010 1,102.83 29.46 8.50% 19.2% 8.9% 4.0% 2.0% 0.9% 0.4% 06/04/2010 1,064.88 35.48 10.24% Outside upper 2,044 850 189 14 2 0 06/07/2010 1,050.47 36.57 10.56% VIX range 06/08/2010 1,062.00 33.7 9.73% 37.0% 15.4% 3.4% 0.3% 0.0% 0.0% 06/09/2010 1,055.69 33.73 9.74% Total 5,521 5,521 5,521 5,521 5,521 5,521 observed data 06/10/2010 1,086.84 30.57 8.82% Results are plotted Empirical Rule 38.20% 68.20% 86.60% 95.40% 98.80% 99.80% 06/11/2010 1,091.60 28.79 8.31% 30 days forward as 06/14/2010 1,089.63 28.58 8.25% implied As observed from table 2, the VIX estimates managed to contain the SPX action quite well over the period (5,521 days) 06/15/2010 1,115.23 25.87 7.47% under study. 06/16/2010 1,114.61 25.92 7.48% Accordingly, this paper proposes extending the statistical 06/17/2010 1,116.04 25.05 7.23% and visual capability of the VIX by representing the statistical interpretation of the VIX values over the SPX in the form of a 06/18/2010 1,117.51 23.95 6.91% projection range that moves dynamically as the SPX progresses 06/21/2010 1,113.20 24.88 7.18% forward in time. 06/22/2010 1,095.31 27.05 7.81% 4. Introducing the Implied Volatility 06/23/2010 1,092.04 26.91 7.77% Projection Range (IVPR) 06/24/2010 1,073.69 29.74 8.59% 06/25/2010 1,076.76 28.53 8.24% In this section, we introduce the IVPR and convey the 06/28/2010 1,074.57 29 8.37% means to visually plot it over the SPX. With reference to the interpretation previously presented, we regress the results over 06/29/2010 1,041.24 34.13 9.85% time (shown in table 3) and plot the resulting SPX range(s) over 06/30/2010 1,030.71 34.54 9.97% the SPX values on the chart (visualized by figure 2). 07/01/2010 1,027.37 32.86 9.49%

Figure 2: Upper window: The SPX index vs. the IVPR—Daily values— 07/02/2010 1,022.58 30.12 8.69% Candlestick chart—Normal scale. Lower window: The VIX index—Daily 07/06/2010 1,028.06 29.65 8.56% values—Candlestick chart—Normal scale 07/07/2010 1,060.27 26.84 7.75% Upper SPX VIX range shifted 1450 07/08/2010 1,070.25 25.71 7.42% 966.72 1,181.34 forward by 30-days 1400

1350 07/09/2010 1,077.96 24.98 7.21% 959.99 1,175.91

1300

1250 07/12/2010 1,078.75 24.43 7.05% 1,008.55 1,197.57 1200 07/13/2010 1,095.34 24.56 7.09% 988.55 1,190.27 1150 1100 07/14/2010 1,095.17 24.89 7.19% 960.86 1,180.56 1050

1000 07/15/2010 1,096.48 25.14 7.26% 1,002.72 1,194.04 Lower SPX VIX range shifted forward by 30-days 950 50 07/16/2010 1,064.88 26.25 7.58% 1,009.04 1,196.62

45 40 07/19/2010 1,071.25 25.97 7.50% 955.81 1,173.95 35

30 25 07/20/2010 1,083.48 23.93 6.91% 939.57 1,161.37 20

15

7 14 21 28 4 11 18 25 2 9 16 23 31 6 13 20 27 5 11 18 25 1 8 15 22 29 6 12 19 26 3 10 17 24 31 7 14 21 28 5 12 19 27 3 9 16 23 30 6 13 20 27 5 07/21/2010 1,069.59 25.64 7.40% 958.68 1,165.32 March April May June July August September October November December 2012 February Ma

PAGE 8 IFTA.ORG IFTA JOURNAL 2013 EDITION

Figure 3: The SPX index vs. the IVPR—Daily values—Candlestick chart—Normal scale

1300 1290 1280 1300 1270 1290 1260 1280 1250 1270 1240 1260 1230 1250 1220 1240 1210 1230 1200 1220 1190 1210 W-shape formation 1180 1200 1170 1190 1160 W-shape formation 1180 1150 1170 1140 1160 1130 1150 1120 1140 1110 1130 1100 1120 1090 1110 1080 1100 1070 1090 1060 1080 1050 Subsequent rally 1070 1040 1060 1030 Subsequent rally 1050 1020 1040 1010 Lower low outside 1030 1000 lower IVPR 1020 990 1010 Lower low outside 980 1000 lower IVPR 970 990 960 980 950 970 Lower low failed 940 960 to break below 930 950 lower IVPR 920 Lower low failed 940 910 to break below 930 900 lower IVPR 920 15 22 29 5 12 19 26 3 10 17 24 1 7 14 21 28 6 12 19 26 2 9 April May June July August 910 900 15 22 29 5 12 19 26 3 10 17 24 1 7 14 21 28 6 12 19 26 2 9 April May June July August

Figure 4: The SPX index vs. the IVPR—Daily values—Candlestick chart—Normal scale Higher high failed 1630 M-shape formations to break above 1620 upper IVPR 1610 Higher high failed 16301600 M-shape formations Higher high outside to break above 16201590 higher IVPR upper IVPR 16101580 16001570 Higher high outside 15901560 higher IVPR 15801550 15701540 15601530 15501520 Higher high failed 15401510 to break above 15301500 Higher high outside upper IVPR 15201490 Higher high failed higher IVPR 15101480 to break above 15001470 Higher high outside upper IVPR 14901460 higher IVPR 14801450 14701440 14601430 14501420 14401410 14301400 14201390 14101380 14001370 1360 Subsequent decline 1390 Subsequent decline 13801350 13701340 13601330 Subsequent decline 1320 Subsequent decline 1350 13401310 13301300 13201290 13101280 13001270 12901260 12801250 12701240 12601230 12501220 12401210 12301200 12201190 12101180 12001170 11901160 11801150 1170 5 11 18 25 2 9 16 23 30 6 13 20 27 4 11 18 26 3 16 22 29 5 12 20 26 5 12 19 26 2 9 16 23 30 7 14 21 29 4 11 18 25 2 9 16 23 30 6 13 2 1160 September October November December 2007 February March April May June July August 1150

5 11 18 25 2 9 16 23 30 6 13 20 27 4 11 18 26 3 16 22 29 5 12 20 26 5 12 19 26 2 9 16 23 30 7 14 21 29 4 11 18 25 2 9 16 23 30 6 13 2 September October November December 2007 February March April May June July August

IFTA.ORG PAGE 9 IFTA JOURNAL 2013 EDITION

Figure 2 introduces the IVPR when visually plotted over the SPX 5.2 During the SPX trendless phases and trend reversals: (upper window) and depicts the VIX (lower window). As a result of the visualization process, the IVPR boundaries 5.2.1 Trendless phases are plotted 30-days forward as implied by the VIX calculation. During trendless or sideways phases in the SPX, the buying Accordingly, the VIX implies a probability of 68.2% and selling power in the market are generally not sufficient (approximately) that the expected range of the SPX returns over to build-up or sustain a structured trend (up or down). It was the next 30 days will lie within terminus points of the IVPR. observed that all price excursions from both IVPR boundaries were unsustainable. 5. Technical interpretations of the IVPR We propose using both IVPR boundaries during such phases Additional to the common uses/applications of the VIX as overbought (OB) and oversold (OS) levels for the SPX. As such, previously stated, we introduce further interpretations by using reaching or breaking the upper boundary, followed by a pull- the IVPR. back inside, is regarded as a selling (trading) opportunity and vice versa for the lower IVPR. This tactic is similar to that used 5.1 During the SPX trending phases with -based envelopes during sideways trends.11 During up-trending phases, volatility (the VIX) trends down. Meanwhile the SPX trends close to the upper IVPR and Figure 5: The SPX index vs. the IVPR—Daily values—Candlestick chart— occasionally trends above it for a relatively short period. As Normal scale the SPX forms higher highs, the SPX exceeds the projected 445

440 estimation of the upper IVPR. However, in specifically 435 observed cases, failing to reach the upper IVPR during up- 430 425 trends implies weakness in the trend and suggests a counter- 420 415 trend correction. This can sometimes lead to a change in 410 405 the overall SPX trend direction, especially if the VIX itself 400 395 reverses direction. We recommend adopting the technique 390

385 associated with the M-shape formations as a trading pattern 380 that aims at highlighting such weaknesses. The M (& W) shape 375 370 formations were proposed by Bollinger through his works on the 365 360 (B-Bands) as two useful techniques to identify 355 350 10 weaknesses in the underlying trends. 345 340 Moreover during up-trends, the SPX rarely makes any IVPR aids in identifying OB/OS conditions 335 during SPX sideways or trendless phases 330 attempts towards the lower IVPR. In fact, all the SPX attempts 325 towards the lower IVPR during a structurally-maintained 320 April May June July August September October November December 1992 February March April May June July August September October uptrend and excluding trend reversal events are considered key lows in that up-trend. These dips are regarded as buying opportunities in the market. The opposite holds true during down-trending phases. The Figure 5 shows how the IVPR boundaries can be used to VIX trends up, while the SPX trends close to/and occasionally help identify OB and OS levels once a sideways phase has been exceeds the projected estimation of the lower IVPR, as it identified over the SPX. registers lower lows. In certain cases, failing to reach the lower IVPR during down-trends implies weakness in the trend and 5.2.2 Trend reversals suggests a counter-trend correction. This can sometimes lead A trend reversal phase may be viewed as a temporary to a change in the overall SPX trend direction, especially if the sideways or trendless condition in which an exchange in power VIX itself reverses direction. Similarly, we propose the trading between the buyers and sellers occur. This causes an existing technique associated with the Bollinger W-shape formations trend structure (up-trend or down-trend) to reverse to the as an appropriate tactic to be used during such events. On the opposite trend structure. other hand, during down-trends, rare attempts for the SPX We have observed that during structural trend reversals in towards the upper IVPR are observed and are considered key the SPX, volatility of the SPX also tends to change in behavior. highs/selling opportunities in the market. During reversals from down trends, the following occurs: Figure 3 shows the IVPR plotted over the SPX. Unlike the ƒƒ The undergoing rise in volatility (up trending VIX) begins to May/June low, the lower low created in July failed to move below reverse from high levels (above 30%) at the same time or prior the lower IVPR, implying that the market failed to exceed its to the actual reversal of the SPX. prior 30-day volatility expectations and implied lesser volatility ƒƒ The SPX begins to fall short of moving below the lower IVPR going forward. This was followed by a counter-trend rally that boundary, or—at least—makes brief attempts at the lower sustained till August. This trading pattern is comparable to a IVPR when compared to previous lows in that same bear trend. W-shaped B-Band pattern. ƒƒ The SPX performs successful attempts to reach the upper Figure 4 shows the IVPR plotted over the SPX. The chart IVPR, signaling new strength in buying power. suggests two cases of M-shaped B-Band formations and their subsequent countertrend declines.

PAGE 10 IFTA.ORG IFTA JOURNAL 2013 EDITION

Figure 6: The SPX index vs. the IVPR—Daily values—Candlestick chart—Normal scale ƒƒ The IVPR boundaries fail to mark lower lows indicating that volatility is 1250

receding. 1200 ƒƒ The trend reversal is confirmed A higher SPX high (reversal confirmation) accompanied by a breach of the upper IVPR 1150 when the SPX successfully moves and sustains outside the upper IVPR This SPX high was associated with a breach of 1100 the upper IVPR; an event unseen in this 2-year following an initial structural trend- bear trend except during May 2001. 1050

reversal pattern. 1000

950 Figure 6, shows the bottoming phase which occurred on the SPX following 900 the 2000–2002 bear trend. Additional 850

to the classic reversal structure which 800 consisted of a higher low (in March A higher 750 2003), followed by a higher high (in SPX low May/June), we have highlighted the 700 behavior of the SPX with the IVPR. The 650 last registered lower high of the SPX May June July August September October November December 2003 February March April May June July August September (November 2002) managed to breach above the IVPR on the near term frame. Figure 7: The SPX index vs. the IVPR—Daily values—Candlestick chart—Normal scale This was followed in March 2003 by a A lower This SPX high failed to relatively brief breach below the IVPR SPX high breach the upper IVPR 1650 and a higher SPX low was created during 1600 that same month. This higher SPX low was confirmed by a higher IVPR low. 1550 Finally, the higher SPX high which 1500 occurred later in May/June 2003 was 1450 associated with a breach of the upper 1400 boundary of the IVPR. Figure 7 shows the topping phase 1350 which occurred on the SPX prior to the 1300 2000–2002 bear trend. A lower high (in 1250 September 2000), followed by a lower A lower low (in December 2000) constituted a SPX low 1200 trend reversal pattern. Confirming this A lower SPX low (reversal confirmation) 1150 accompanied by a breach of the lower IVPR reversal, the last registered SPX peak 1100 during this period (March 2000) was accompanied by a breach above the 1050 ber No vem ber December 2000February March April May June July August September October November December 2001February March April upper IVPR. The following (lower high), however, failed to move outside the Figure 8: The NASDAQ 100 index vs. the IVPR—Daily values—Candlestick chart—Normal scale upper IVPR (a confirmed weakness). This 2450 was followed by a lower low in December 2400 that succeeded to move below the lower 2350 IVPR to confirm the market reversal. 2300 2250

2200

Conclusion 2150

Without any doubt, the VIX is 2100 an invaluable market indicator that 2050 provides essential market clues and 2000 1950 expresses the SPX trend conditions from 1900 the volatility standpoint. In this paper, 1850 we have recognized the virtues of the 1800 VIX as a standalone index and proposed 1750 1700 a new method that aims at extending 1650 the statistical and visual capability of 1600 the VIX, namely, the IVPR. The IVPR 1550 transforms the stationary/statistical 1500 1450 294 11 18 25 2 9 16 23 30 6 13 20 27 4 10 17 24 1 8 15 22 29 5 12 19 26 3 10 17 24 31 7 14 22 28 4 11 19 25 3 10 17 24 31 7 14 21 28 5 12 19 26 2 9 16 23 June July August September October November December 2008 February March April May June

IFTA.ORG PAGE 11 IFTA JOURNAL 2013 EDITION

interpretation of the VIX values into Figure 9: The Russell 2000 index vs. the IVPR—Daily values—Candlestick chart—Normal scale dynamic boundaries that project 830 820 810 the SPX expectations in the future, 830800 820790 810780 800770 derived from a statistical standpoint. 790760 780750 770740 The IVPR is yet still not without 760730 750720 740710 730700 limitations. Although the IVPR 720690 710680 700670 690660 managed to contain the SPX 680650 670640 660630 650620 data according to predetermined 640610 630600 620590 statistical probabilities, the SPX will 610580 600570 590560 580550 often exceed the IVPR boundaries 570540 560530 550520 540510 and sustain over the NT horizon. We 530500 520490 510480 attribute this to a TA-based premise 500470 490460 480450 470440 stating that markets move in defined 460430 450420 440410 12 430400 trend structures. In part, this is 420390 410380 400370 evident when the SPX values track 390360 380350 370340 360330 one of the IVPR boundaries during a 350320 340310 330300 320290 trending phase and not the other. 310 August September October November December 2009 February March April May June July August Sep 300 290

Nevertheless, the IVPR offers a August September October November December 2009 February March April May June July August Sep more comprehensive statistical and visual capability when compared to the VIX index (on a standalone Figure 10: The SPX index vs. the IVPR @ 2 standard deviations—Daily values—Candlestick basis). This advantage is especially chart—Normal scale valuable when trying to understand the relationship between the VIX and 1800 1750 1800 the SPX, and ultimately aids—to some 1700 1750 1650 1700 1600 extent—in forecasting the expected 1650 1550 1600 1500 moves of the SPX in the future. 1550 1450 1500 1400 1450 1350 1400 1300 1350 References 1250 1300 1200 [1, 2] Chicago Board Options Exchange 1250 1150 1200 1100 (CBOE), VIX White Paper, 2009. Pages: 1 and 1150 1050 1100 4 respectively. 1000 1050 950 [3, 4, 5, 6, 7] Whaley, Robert E., IVPR at 2 standard deviation 1000 900 950 IVPRwill contain at 2 standard more of deviation the SPX action 850 Understanding the VIX, The Journal of 900 800 will contain more of the SPX action 850 Portfolio Management, Spring 2009. Pages: 750 800 700 98, 99, 100 and 101 respectively. 750 650 700 600 [8, 9] Rhoads, Russel, Trading VIX 650 550 600 Derivatives: Trading and Hedging Strategies 500 550 450 Using VIX Futures, Options, and Exchange 500 400 450 350 Traded Notes, Wiley, 2011. Page 23. 400 300 350 10 Bollinger, John A., Bollinger on Bollinger ly August September October November December 2008 February March April May June July August September October November December 2009 February March 300 Bands, McGraw-Hill, 2001. Page 94. ly August September October November December 2008 February March April May June July August September October November December 2009 February March [11, 12] Murphy, John J., Technical Analysis of the Financial Markets, New York Institute of Finance, 1999. Pages: 207 and 3 respectively.

Bibliography Software and data Bollinger, John A., Bollinger on Bollinger Bands, McGraw-Hill, 2001. Data courtesy of Bloomberg and Reuters. Chicago Board Options Exchange (CBOE), VIX White Paper, 2009. Charting software courtesy of Equis Hull, John C., Options, Futures, and Other Derivatives, Prentice Hall, 2000. International MetaStock v.9.1. Mason, Robert D., Marchal, William G., Lind, Douglas A., Statistical Techniques in Business & Economics, McGraw-Hill/Irwin, 2002. Murphy, John J., Technical Analysis of the Financial Markets, New York Institute of Finance, 1999. Pring, Martin J., Technical Analysis Explained: The Successful Investor‘s Guide to Spotting Investment Trends and Turning Points, McGraw-Hill, 2002. Rhoads, Russell, Trading VIX Derivatives: Trading and Hedging Strategies Using VIX Futures, Options, and Exchange Traded Notes, Wiley, 2011. Whaley, Robert E., Understanding the VIX, The Journal of Portfolio Management, Spring 2009.

PAGE 12 IFTA.ORG IFTA JOURNAL 2013 EDITION

Psychological Barriers in Asian Equity Markets By Shawn Lim, CFTe, MSTA, and Bryan Lim

In this study, we investigate the presence of psychological explanations that have been offered for these round number barriers in the equity indexes of 10 Asian Markets over a 10‑year effects include psychological preferences for round numbers period from 2001–2011. This investigation was conducted (Ziemba et al 1986), coordination on limited price set (Harris through the use of uniformity tests, barrier proximity tests 1991), convenience (Mitchell 2001), odd pricing (Schindler and tests on the predictability of stock returns. We have found and Kirby 1997), bounded rationality and aspiration levels evidence for barriers at the 1000 level for 6 of these markets (Sonnemans 2006). (JKSE, KLSE, N225, STI, KS11, TWII) and at the 100 level for 4 of these markets (AORD, JKSE, KLSE, STI). However, while Data there may be evidence of psychological barriers, there is little This study uses closing prices for the following 10 Asian evidence for the predictability of stock returns induced by the Equity Indices for the 10-year period from 1 Jan 2001 to 31 presence of these psychological barriers. Dec 2011 as obtained from Yahoo Finance. Table 1 provides a summary of the indices used and their range over the test period. Introduction “Hang Seng Index: Investors hope for support at 19,000”, Table 1: Summary of the 10 Asian Equity Indices and their range from 1 Jan 01 to 31 Dec 11 “Nikkei rebounds after slipping below 8,500”, “KOSPI may test support at 1,900 points”, and the list goes on. This is Symbol Name Market High Low just a small snapshot of news headlines taken in May 2012, but the underlying theme is a distinctively familiar one. AORD All Ordinaries Australia 6853.6 2673.3 Regions of round numbers have often been regarded with SSEC Shanghai Composite China 6092.06 1011.5 special significance in the financial press, with the approach HSI Hang Seng Index Hong Kong 31638.22 8409.01 or penetration of these levels often taken by financial commentators to be of particular importance and is hence BSESN BSE 30 India 21004.96 2600.12 often used as a barometer of market sentiment. JKSE Jakarta Composite Indonesia 4193.44 337.48 However, while the evidence that people (or at least the KLSE KLSE Composite Malaysia 1594.74 553.34 press) view levels around round numbers as important is indisputable, is the added attention truly warranted? N225 Nikkei 225 Japan 18261.98 7054.98 Empirical studies applied to US markets have found some STI Straits Times Index Singapore 3875.77 1213.82 evidence of psychological barriers (Donaldson and Kim KOSPI Composite KS11 Korea 2228.96 468.76 1993) while evidence for return predictability has been weak Index (Koedijk and Stork 1994). While there has been substantial TWII Taiwan Weighted Taiwan 9809.88 3446.26 empirical research on the presence of these effects in Western markets, there has been considerably less research focused on Asian markets and this study attempts to supplement M-Values this growing body of knowledge by testing for the presence In order to test for the presence of psychological barriers, of psychological barriers around round numbers in Asian it is first necessary to introduce the concept of M-values as is Equity Indices. There are 3 ways that tests for psychological often used in empirical tests for barriers. significance are often conducted; tests on the distribution M-values are two digit numbers, ranging from 00 to 99 and of digits, tests on the frequency of digits, and the behaviour represent a point, with 00 representing the region around a of returns around presupposed barriers. This study tests for round number. We use 2 sets of M-values for each index, M 1000 barriers using these 3 categories of tests as applied to the 10 for M-values to test the presence of barriers at the 1000 level selected Asian Equity Indices. and M 100 for M-values to test the presence of barriers at the 100 Psychological barriers refer to regions of support or level. These are defined as follows: resistance around round numbers, hence at the 11300, 11400, mod 100. 11500, etc levels for barriers at the 100 level and at 12000, mod 100. 13000, 14000, etc for barriers at the 1000 level. As there is ⌊ ⌋

no fundamental reason for these levels to be of particular Where P t  is⌊ the⌋ integer part of Pt. For example, if prices are importance, the presence of regions of support or resistance 24998.02 and 54738.12, M 100 are 98 and 38 respectively. around these levels have been attributed to behavioural biases, hence the term ‘psychological barriers’. Some mod 100. ( ) mod 100. ⌊( ) ⌋ IFTA.ORG PAGE 13 ⌊ ⌋

IFTA JOURNAL 2013 EDITION

For example, if prices are 24998.02 and 54738.12, M 1000 are 99 and 73 respectively. When examining barriers at the 100 level, we would expect to see the index closing less frequently at the xx00.xx level if the barriers existed, and for the 1000 level we would expect to see the index closing less frequently at the x00x.xx level. This is what the respective M-values represent and is the rationale for using them in the tests specified throughout the rest of this paper.

Frequency Distribution

The following charts show the distribution of the M-values for the 10 markets:

PAGE 14 IFTA.ORG IFTA JOURNAL 2013 EDITION

Presence of Psychological Barriers

Uniformity Test To investigate the presence of psychological barriers, we conduct tests for positional and transgressional effects at the M 1000 and M 100 levels by carrying out a chi-square test using 3 different set ups and the resulting test statistics are reported in Table 2 and Table 4. We regard the thousand and hundred levels as potential psychological barriers and hence test the M-values at the corresponding levels. If there are no particular areas with any significance, we would then expect the M-values to follow a uniform distribution, i.e. there would be no particular reason for an index to close at, for example, 11900 (M 1000 = 90, M 100 = 00) more frequently than 11870 (M 1000 = 87, M 100 = 70) over the 10 year period investigated and hence we would expect to see the index closing at approximately an equal number of times at the 99 M-value as at the 45 M-value and so forth. If there

IFTA.ORG PAGE 15 If there are no particular areas with any significance, IFTAwe JOURNALwould then 2013 expect EDITION the M-values to follow a uniform distribution, i.e. there would be no particular reason for an index to close at, for example, 11900 ( = 90, = 00) more frequently than 11870 ( = 87, = 70) over the 10 year period investigated and hence we would are regions of resistance or support close to round numbers, with that adopted by Koedijk and Stork (1994). expect to see the index closing athowever, approximately we would then an equalexpect numberto see less ofM-values times closeat the to 99 M-valueWe as test at bythe means 45 M-value of a OLS regression test whether

If there are no particular areas andwith soany forth. significance, If there we are would regions thenthe of expect 00resistance region the and M-values or for support the distributionto follow close a oftouniform all round the M-values numbers, to not however, the we distribution would thenof M-values expect is linked to the presence of distribution, i.e. there would be noto particular see less reason M-values for an close index to closethefollow at,00 foraregion uniform example, and distribution. 11900 for the ( Hence,distribution= 90, the first = oftest all (‘Absolute the M-values psychologicalto not follow barriers. a uniform A vector P(M) with length 100 is 00) more frequently than 11870 ( = 87, = 70) over the 10 yearTest’) period tests theinvestigated distribution and of hence allWe the test weM-values would by meansagainst aof uniform a OLS regressioncreated, which test registers whether the relativethe distribution frequency of of each M-values M-value is linked to the presence of distribution. Hence, the first test (‘Absolute Test’) tests the distribution of all the M-values against a uniform expect to see the index closing at approximately an equal number of timesdistribution at the with99 M-value the hypothesis as at psychologicalthe h0 :45 the M-value 100 M-values barriers. follow A a vector P(M)occurring with along length with 1003 dummy is created, variables, which D1, D2 and registers D3 and which the relative frequency of each M- and so forth. If there are regions distributionof resistance orwith support the hypothesi close to rounds uniform :numbers, the distribution100 however, M-values against we follow would h1value: the athen 100uniform occurring M-Valuesexpect distribution doalong not follow with against a3 dummy equal: the variables,1 if the100 M-value M-Values , of the anddo index at which closing equalis in one 1 of if thethe M-value of the index at closing is uniform distribution. The Chi-Squared Test static for each index following ranges: 98, 99, 00, 01, 02 for D1; 93, …, 97 or 3, …, 7 for to see less M-values close to thenot 00 followregion aand uniform for the distribution. distribution of The all Chi-Squaredthe M-values Testto not static followin onefor a each uniformof the index following is computed ranges: as 98, follows: 99, 00, 01, 02 for ; 93, …, 97 or 3, …, 7 for ; 85, …, 92 or 8, …, 15 for . We is computed as follows: D 2; 85, …, 92 or 8, …, 15 for D3. We then regress the P(M) vector distribution. Hence, the first test (‘Absolute Test’) tests the distribution of all the M-values against a uniform then regress the P(M) vector againstagainst these these 3 3variables: variables: distribution with the hypothesis : the 100 M-values follow a uniform distribution against : the 100 M-Values do not follow a uniform distribution. The Chi-Squared Test static for each index is computed as follows: ( ) Where stands for the number of observations in each ∑ category i (i=0..99 for ‘Absolute Test’, i=1..10 for ‘Barrier Where Oi stands for the number of observations in each If there ( are) no psychological barriers and the M-values

Test’ and i=1,2 for ‘Remainder Test’)category and i (i = 0…99 stands for ‘Absolute for the Test’, number i=1…10 of for observations ‘Barrier Test’ expectedfollow ( a uniform distribution,). what we would expect is an ( ) and i = 1,2 for ‘Remainder Test’) and Ei stands for the number intercept of C = 1 and , and β 1, β 2 and β 3 = 0. If the β values Where stands for the number of observations in each ∑ category i (i=0..99 for ‘Absolute Test’, i=1..10 for ‘Barrier

The number of degrees of freedom for each test is computed as df = n-1. are significant and not equal ( ) to zero, what this implies is that Test’ and i=1,2 for ‘Remainder Test’) and stands for the number ofof observations observations expectedexpected ( ).. The number of the relative frequency of M-values at the respective levels is

degrees of freedom for each test is computed as df = n-1. greater (lower) than 1 if the β values are positive (negative). For The number of degrees of freedom for each test is computed as df = n-1. ( ) The second test (‘Barrier Test’) follows the methodology example, if C = 1 and β 1 = –0.2 and β 1 is statistically significant, adopted by Koedijk and Stork (1993) and splits the M-values what this suggests is that the relative frequency of occurrence into 10 disjunct categories of equal size, i.e. 06-15, 16-25,…, of a M-value if it is in the 98, 99, 0, 1, 2 region is 1 – 0.2 = 0.8, 96-05. We register the number of times the index closes with i.e. significantly less than we would expect if the M-values an M-Value in these categories and perform a chi-square test are uniformly distributed and supportive of the presence of with 10 categories in the manner described above. This test for psychological barriers at round numbers. The results of this uniformity is similar to the ‘Absolute Test’ but widens the range regression are shown in Table 3 and Table 5 below. of the potential areas of significance and is consistent with the conventional wisdom of areas of being Positional Effects price bands and not absolute price levels. In this section we investigate if the indices close more or less While the first and second tests pick up whether the frequently around round numbers by performing the uniformity distribution of M-values follows a uniform distribution, they say tests and barrier proximity tests on the M-values of the closing little about the location of these deviations and whether they prices across the 10 years from 1 Jan 2001 to 30 Dec 2011. are indeed around regions of round numbers, as per the purpose of the investigation. For example, if the index closes much Uniformity Tests less frequently in the 40-50 M-value region but is uniformly From the uniformity tests, we see strong evidence against distributed over the remaining values, the null hypothesis a uniform distribution of M-values at the level for 8 out of 10 would be rejected in the first two tests but this would not be the of the indices in the absolute test (AORD, SSEC, BSESN, JKSE, effect we are trying to test for. Hence, a third test (‘Remainder KLSE, STI, KS11, TWII), with the results confirmed in the barrier Test’) is introduced that attempts to separate this effect by tests and in 5 out of 8 of these indices in the remainder tests splitting the M-values into 2 categories (96-05, 06-95) and a chi- (SSEC, JKSE, KLSE, STI, KS11). For the AORD, BSESN and TWII squared-goodness-of-fit test with the expected number in each indices that show evidence for non-uniformity in the absolute category based on the expected number if it followed a uniform and barrier but not remainder test, this could be indicative distribution has also been conducted and the results reported of non-uniformity in this distribution of M-values, not due to below. The barrier tests and remainder tests have also been significantly less (or more) values in the region of interest, but conducted for a wider 20 point band (90-09,etc) but the results outside the round number region. For the 100 level we find some were qualitatively similar to the tests for the 10 point band and evidence against uniformity in the barrier tests for 2 out of 10 of hence have been omitted in the presentation below. the indices (N225, STI).

Barrier Proximity Test Barrier Proximity Tests One major limitation of the uniformity tests is the lack of From the regression, we see evidence of psychological information on directionality, in that while it may present evidence barriers at the 1000 level in 6 out of 10 of the indices (SSEC, for a non-uniform distribution due to unexpected deviations in our JKSE, KLSE, N225, STI, KS11) consistent with the results from region of interest, it fails to show if that is because the index closes the uniformity test, as well as evidence of barriers at the 100 more frequently in those regions (price clustering) or because the level in the 3 out of 10 of the indices (JKSE, KLSE, TWII). We index closes less frequently (evidence of barriers). This information also find evidence of clustering around the 93-97, 3-7 levels as is picked up in our second test, the barrier proximity test, as indicated by the positive and significant β 2 values in the AORD described by Donaldson and Kim (1993). We employ a variant of and TWII M 1000 regressions, which could be an explanation for the methodology that yields more interpretable information in line the results they exhibited in the remainder tests.

PAGE 16 IFTA.ORG IFTA JOURNAL 2013 EDITION

Table 2: Chi-squared test statistics from the uniformity tests on closing prices Transgressional Effects In this section we test for the Index Absolute Tests Barrier Tests Remainder Tests presence of psychological barriers by M 1000 M 100 M 1000 M 100 M 1000 M 100 investigating if the regions around AORD 263.0792 90.95103 128.1602 13.46813 1.490377 0.006762 round numbers have been transgressed SSEC 190.039 100.3166 74.8177 14.86272 12.42461 2.481287 less frequently than other regions. For HSI 105.3825 115.4372 14.63206 6.12204 0.292451 0.972475 instance, if the index jumps from 1420 to BSESN 140.3906 67.92364 27.82085 2.997063 0.375918 0.127916 1532, the M 1000 values from 43 to 53 are JKSE 921.707 86.17156 631.0271 6.863106 32.06673 0.529898 transgressed and the M 100 values from KLSE 1204.126 121.0935 675.8046 12.65403 101.7024 1.281192 21 to 32 are transgressed. We conduct N225 102.7389 100.0682 9.578635 18.88131 3.85905 0.079789 chi-squared tests for uniformity and STI 233.121 116.7521 113.2472 18.25046 20.70616 2.916069 perform the barrier proximity tests KS11 717.5943 82.83391 490.082 11.41339 38.64807 0.842507 using the same specifications as above TWII 144.1781 96.86387 62.41943 6.841795 2.596026 1.444853 on the distribution of M-values that Results significant at the 95% confidence level are in yellow have been transgressed for each index and the results are presented in an Table 3: Regression parameters for the barrier proximity tests on closing prices identical format to Tables 2 and 3 in Tables 4 and 5. Index c Pc β 1 Pβ 1 β 2 Pβ 2 β 3 Pβ 3 M 1000 0.99 0.00 0.04 0.79 0.25 0.01 -0.13 0.12 Uniformity Test AORD M 100 1.01 0.00 -0.06 0.50 0.01 0.85 -0.08 0.11 For the uniformity tests, we see M 1000 1.05 0.00 -0.29 0.01 -0.25 0.00 -0.07 0.34 evidence against a uniform distribution SSEC M 100 0.98 0.00 0.00 0.98 0.13 0.04 0.04 0.46 of M-values in all the indices at the 1000 M 1000 0.98 0.00 0.07 0.42 0.05 0.47 0.10 0.06 level and for 8 out of 10 indices at the HSI M 100 1.00 0.00 0.16 0.08 -0.13 0.06 0.02 0.75 100 level (AORD, SSEC, HSI, BSESN, JKSE, M 1000 1.00 0.00 0.05 0.67 0.10 0.21 -0.06 0.38 KLSE, N225, STI) in the barrier tests. BSESN M 100 1.00 0.00 0.09 0.20 -0.02 0.68 -0.01 0.82 This is confirmed in 7 out of 10 indices at M 1000 1.16 0.00 -0.58 0.02 -0.46 0.01 -0.50 0.00 the 1000 level (AORD, HSI, JKSE, KLSE, N225, STI, KS11) in the remainder tests JKSE M 100 1.02 0.00 -0.15 0.07 -0.01 0.85 -0.09 0.07 and in 6 out of 10 indices in the absolute M 1000 0.97 0.00 -0.63 0.03 -0.20 0.35 0.49 0.01 tests (AORD, JKSE, KLSE, STI, KS11, KLSE M 100 1.01 0.00 -0.19 0.06 0.04 0.58 -0.03 0.64 1000 TWII). At the 100 level, this is confirmed M 1.03 0.00 -0.26 0.00 -0.09 0.17 -0.04 0.47 in 6 out of 8 indices in the remainder 100 N225 M 1.02 0.00 0.02 0.82 -0.10 0.12 -0.05 0.32 tests (AORD, HSI, BSESN, JKSE, KLSE, 1000 M 1.06 0.00 -0.35 0.01 -0.31 0.00 -0.08 0.27 STI) and in 2 out of 8 indices in the 100 STI M 1.01 0.00 -0.09 0.33 0.04 0.53 -0.08 0.16 absolute tests (KLSE, STI). The stark M 1000 1.05 0.00 -0.30 0.20 -0.43 0.01 0.04 0.76 difference in the number of indices KS11 M 100 1.01 0.00 -0.03 0.79 0.05 0.38 -0.06 0.19 that demonstrate evidence against a M 1000 0.98 0.00 0.04 0.69 0.15 0.06 0.02 0.75 uniform distribution in the absolute TWII M 100 1.01 0.00 0.03 0.72 -0.13 0.05 -0.01 0.81 test compared to in the barrier and in

Results significant at the 90% confidence level are in yellow. PA represents the P-value of the t-test the remainder tests particularly at the run with the hypothesis: h0: A = 0, H1: A ≠ 0 100 level suggests that for some of these indices the individual M-values may be Table 4: Chi-squared test statistics for the uniformity tests on M-value transgressions approximately uniformly distributed, Index Absolute Tests Barrier Tests Remainder Tests but when grouped into categories M 1000 M 100 M 1000 M 100 M 1000 M 100 around the barrier levels there is AORD 248.0737 100.1657 105.9686 88.91875 10.85873 31.79044 evidence against a uniform distribution, SSEC 97.82411 31.10837 59.21055 17.73228 0.529747 0.046776 which can be seen as evidence for the HSI 51.86512 35.74191 36.30685 27.80382 6.650109 9.455114 presence of psychological barriers in the BSESN 33.93022 37.25957 23.1322 32.05841 0.174333 6.604553 region around round numbers but not at JKSE 548.9661 86.78622 403.9064 62.89023 48.19487 10.54079 the exact round number level, consistent KLSE 263.1542 220.4746 133.578 116.1296 69.24603 23.0023 with the conventional wisdom of N225 90.12974 23.93532 71.15175 17.56112 7.453099 0.117918 support and resistance existing as STI 328.871 127.7222 214.8117 94.26897 30.37677 6.990322 regions around a level instead of at a KS11 756.551 36.22413 606.6621 7.233078 15.31477 1.150354 single fixed level. TWII 396.3475 10.93114 355.3461 6.762931 0.329328 0.006915 Results significant at the 95% confidence level are in yellow

IFTA.ORG PAGE 17 IFTA JOURNAL 2013 EDITION

Barrier Proximity Test Table 5: Regression parameters for the barrier proximity tests on M-value transgressions

For the barrier proximity tests, we Index c Pc β 1 Pβ 1 β 2 Pβ 2 β 3 Pβ 3 see evidence supporting the presence of M 1000 1.00 0.00 0.13 0.08 0.16 0.00 -0.12 0.01 psychological barriers in 5 out of 10 of the AORD M 100 1.02 0.00 -0.08 0.00 -0.08 0.00 -0.05 0.00 indices at the 1000 level (JKSE, KLSE, N225, M 1000 1.00 0.00 -0.02 0.70 -0.04 0.26 0.03 0.36 STI, TWII) and evidence of price clustering SSEC M 100 1.00 0.00 0.01 0.60 -0.01 0.47 0.01 0.28 around round numbers in 2 out of 10 of the M 1000 0.99 0.00 0.05 0.00 0.03 0.00 0.02 0.00 indices at the 1000 level (AORD, HSI) with HSI M 100 0.99 0.00 0.04 0.00 0.03 0.00 0.02 0.00 the remaining indices with intercepts that M 1000 1.00 0.00 0.01 0.44 0.00 0.87 0.01 0.21 are not statistically significant, suggesting BSESN M 100 0.99 0.00 0.04 0.00 0.03 0.00 0.02 0.00 the absence of psychological barriers. At M 1000 1.14 0.00 -0.53 0.00 -0.37 0.00 -0.46 0.00 the 100 level, we see evidence supporting 100 the presence of psychological barriers in JKSE M 1.02 0.00 -0.04 0.01 -0.08 0.00 -0.05 0.00 1000 6 out of the 10 indices (AORD, JKSE, KLSE, M 1.04 0.00 -0.68 0.00 -0.44 0.00 0.27 0.00 100 STI, KS11, N225 (with weak evidence)) and KLSE M 1.01 0.00 -0.06 0.26 -0.14 0.00 0.05 0.14 1000 evidence supporting price clustering in M 1.02 0.00 -0.10 0.10 -0.04 0.00 -0.08 0.00 2 out of 10 indices (HSI, BSESN) with the N225 M 100 1.00 0.00 0.00 0.74 -0.01 0.06 0.00 0.56 remaining indices with intercepts that are M 1000 1.05 0.00 -0.30 0.01 -0.27 0.00 -0.05 0.39 not statistically significant. STI M 100 1.02 0.00 -0.03 0.13 -0.07 0.00 -0.06 0.00 M 1000 1.00 0.00 -0.14 0.49 -0.20 0.20 0.16 0.19 Return Predictability KS11 M 100 1.01 0.00 -0.02 0.24 -0.02 0.04 -0.02 0.04 Having examined the presence of psycho- Having examined the presence of psychological barriers, we proceed to investigate if these levelsM 1000can 1.01be used0.00 to -0.01 0.85 -0.01 0.81 -0.07 0.09 logical barriers, we proceed to investigate TWII M 100 1.00 0.00 0.00 0.67 0.00 0.38 0.00 0.95 predict stock returns.Having We examined employ thethe methodology presence ofif these psychologicalof Koedijk levels can and be barriers, used Stork to predict(1994) we proceed stock and test to the investigate following if specification these levels can be used to Results significant at the 90% confidence level are in yellow. PA represents the P-value of the over the sample: predict stock returns. We employreturns. the methodology We employ the methodologyof Koedijk andof Stork (1994)t-test run and with test the hypothesis: the following h0: A = 0, specificationH1: A ≠ 0 over the sample: Koedijk and Stork (1994) and test the follow- ing specification over the sample: And where rt stands for the stock return, st for the index at time t, (1) stands for the value of the first dummy variable at time t, (2) stands ( ) ( ) ( ) Dt Dt (3) for the value of the second dummy variable at time and t and Dt stands Where ( ) ( ) ( ) for the value of the third dummy variable at time t. The dummy variables Where Where are specified in the same way as the regression that was run for the barrier proximity test. From the regression, we find little evidence to support return

( ) Table 6: Regression parameters for the test on return predictability ( ) Index c M 1000 0.000023 0.92 -0.000348 0.70 -0.000272 0.66 0.000874 0.14 -0.022171 0.24 AORD M 100 0.000048 0.84 -0.021365 0.26 0.000869 0.45 0.001191 0.07 -0.000713 0.21 M 1000 0.000191 0.62 -0.000634 0.71 -0.001853 0.13 0.000021 0.98 0.004302 0.82 SSEC M 100 0.000110 0.78 -0.000380 0.84 -0.000816 0.44 0.000090 0.92 0.003468 0.86 M 1000 -0.000161 0.67 0.000781 0.58 0.001611 0.12 0.000206 0.81 -0.025871 0.18 HSI M 100 0.000189 0.61 -0.002124 0.23 0.001279 0.25 -0.000929 0.28 -0.026532 0.16 M 1000 0.000634 0.10 -0.000085 0.95 -0.000510 0.62 -0.000786 0.38 0.078057 0.00 BSESN M 100 0.000195 0.60 -0.004174 0.02 0.001333 0.22 0.001692 0.05 0.076570 0.00 M 1000 0.000759 0.02 0.001957 0.27 -0.000756 0.52 -0.000205 0.83 0.117766 0.00 JKSE M 100 0.000689 0.05 0.000338 0.86 0.000841 0.39 -0.000277 0.74 0.116878 0.00 M 1000 0.000078 0.77 0.000792 0.64 0.002787 0.00 0.000166 0.75 -0.132408 0.00 KLSE M 100 0.000539 0.04 -0.002514 0.07 -0.000857 0.24 -0.000231 0.71 -0.134908 0.00 M 1000 -0.000143 0.70 0.000264 0.87 0.000102 0.93 -0.000392 0.65 -0.03418 0.08 N225 M 100 -0.000512 0.16 0.001925 0.28 -0.000608 0.58 0.002081 0.02 -0.034389 0.07 M 1000 0.000147 0.61 0.000045 0.97 0.000885 0.34 -0.000606 0.37 0.010741 0.57 STI M 100 0.000203 0.48 -0.000377 0.82 -0.001220 0.13 0.000367 0.60 0.010346 0.59 M 1000 0.000749 0.05 -0.000864 0.61 -0.000260 0.85 -0.001428 0.09 0.022893 0.23 KS11 M 100 0.000496 0.19 -0.005436 0.00 0.001097 0.30 0.000016 0.99 0.023296 0.22 M 1000 0.000279 0.42 0.000370 0.78 -0.002336 0.01 0.000563 0.48 0.059239 0.00 TWII M 100 0.000097 0.77 0.000732 0.64 0.001230 0.23 -0.000666 0.40 0.060986 0.00 Results significant at the 90% confidence level are in yellow. represents the P-value of the t-test run with the hypothesis: h0: A = 0, H1: A ≠ 0

PAGE 18 IFTA.ORG IFTA JOURNAL 2013 EDITION predictability for most of the indices with most of the parameter effects is larger and hence the evidence would have to be estimates for the dummies of the psychological barriers not stronger for there to be evidence of barriers. In addition, the statistically significant. Only in the BSESN at the 100 level, KLSE test for transgressional effects specifically looks at movements at the 1000 and 100 level, the KS11 at the 100 level and the TWII from one day to the next and hence would capture the presence at the 1000 level is there some evidence of the predictability of of barriers in a more compelling manner than the test for stock returns induced by the presence of psychological barriers. positional effects. Discussion Conclusion Table 7 summarizes the results from all the tests with a final In this study, we have examined the presence of conclusion on whether there is sufficient evidence for positional psychological barriers in the equity indexes of 10 Asian Markets effects, transgressional effects and psychological barriers. over a 10 year period from 2001-2011. We have found evidence When there were conflicts in the results within each section, for barriers at the 1000 level for 6 of these markets (JKSE, they were resolved in the following manner: KLSE, N225, STI, KS11, TWII) and at the 100 level for 4 of these ƒƒ If there was evidence of a non-uniform distribution but no markets (AORD, JKSE, KLSE, STI). However, while there may be evidence of barriers in the regression test, it was concluded evidence of psychological barriers, there is little evidence for that there was no evidence for that section. the predictability of stock returns induced by the presence of ƒƒ If there was evidence of barriers in the regression test but no these psychological barriers. evidence of non-uniformity in all the 3 tests, it was concluded that there was no evidence for the section. References ƒ Aggarwal, R., & Lucey, B. M. (2007). Psychological barriers in gold ƒ If there was evidence of barriers in the regression test and prices? Review of Financial Economics, 217-230 evidence of non-uniformity in only one of the 3 tests, if Chen, M. H., & Tai, V. W. (2011). Psychological barriers and prices behaviour of taifex futures. 2011 International Conference of Taiwan that non-uniformity test was the barrier test, then it was Finance Association concluded that there was no evidence for the section; but if Donaldson, R. G., & Kim, H. Y. (1993). Price barriers in the dow jones industrial average. Journal of Financial and Quantitative Analysis, 28(3), 313-330. the non-uniformity test was the remainder test then it was Dorfleitner, G., & Klein, C. (2009). Psychological barriers in European concluded that there was evidence for the section. stock markets: Where are they? Global Finance Journal, 268-285. Harris, L. (1991). Stock price clustering and discreteness. Review of Financial Studies 4, 389-415. For conflicting results between the section for Johnson, E., Johnson, N. B., & Shanthkumar, D. (2008). Round numbers and security returns. transgressional effects and positional effects, they were Koedijk, K. G., & Stork, P. A. (1994). Should we care? Psychological resolved in the following manner: barriers in stock markets. Economics Letters , 427-432. ƒ Schindler, R.M. and Kirby, P.N. (1997). Patterns of rightmost digits ƒ If there was evidence of transgressional effects but not used in advertised prices: implications for nine-ending effects. Journal of positional effects, it was concluded that there was evidence of Consumer Research 24, 192-201. Ley, E. & Varian, H.R. (1994). Are there psychological barriers in the psychological barriers. Dow-Jones index? Applied Financial Economics, 4, 217-224. ƒƒ If there was evidence of positional effects but not Mitchell, J. (2001) Clustering and Psychological Barriers: The Importance of Numbers. The Journal of Futures Markets, 21, 395–428. trangressional effects it was concluded that there was no Sonnemans, J. (2006). Price clustering and natural resistance points in evidence of psychological barriers. the dutch stock market: a natural experiment. European Economic Review, 1937-1950. Ziemba, W.T., Brumelle, S.L., Gautier, A. and Schwartz, S. L. (1986). Dr. Z’s This is because the data set for the testing of transgressional 6/49 lotto guidebook. Vancouver, Canada: Dr. Z Investments.

Table 7: Summary of results for the various tests

IFTA.ORG PAGE 19 IFTA JOURNAL 2013 EDITION

Regime-Switching Trading Bands Using A Historical Simulation Approach By Ka Ying Timothy Fong, CFTe, MFTA

Abstract DMI indicator itself should be used for generating a reliable Bollinger Bands have been one of the greatest tools trading signal. The DMI tool offers us two key lessons to developed for technical analysis. In statistical terms, Bollinger consider in developing a new technical analysis tool. The first Bands involve the construction of a 95% confidence interval key lesson learned is that a comprehensive technical analysis around the moving average of a stock’s price and they capture tool or trading system should have two key components. One the mean plus and minus two standard deviations. Having said component acts as a filter to help the trader determine whether that, Bollinger Bands are subject to a number of limitations or not the tool or indicator itself should be used. The other including the assumption that prices are normally distributed component, of course, is the main indicator itself. The second and ambiguity in terms of generating explicit trading key lesson learned is the importance of assessing whether the decisions when prices move in trends. The key objective of price is trending or oscillating in technical analysis. this research paper is to develop the next-generation trading This paper will describe the development of the next- bands that not only use empirical price distributions without generation trading bands that not only capture the skewed making a normality assumption but also include the use of and fat-tailed nature of stock price distributions but also use autocorrelation to determine whether prices are moving in a statistical filter to determine whether the price is in the a trending or oscillating regime so that a better buy or sell trending or oscillating regime, in the short, term so that traders decision can be made. The regime-switching trading bands using can make better buy or sell decisions. historical simulation will be applied to different test cases and the performance of these next-generation trading bands will be 2. Review of the Use of Bollinger Bands summarized in this paper. and Other Trading Bands 1. Introduction Bollinger Bands (BB) are one of the useful decision-making Buying low and selling high is one of the most fundamental tools developed in technical analysis. They are typically strategies in trading. Trading bands are intuitive and easy-to- calculated as the 20-day simple moving average of closing use tools that can help traders in determining entry and exit prices plus and minus two standard deviations over that 20- points for their investments. The most popular and well-known day period. The lower bound forms a support while the upper trading bands are the Bollinger Bands, and Section 2 provides a bound forms a resistance. It is important to highlight the fulsome review of Bollinger Bands and other trading bands that linkage between Bollinger Bands and the concept of confidence currently exist. intervals in statistics. In statistical terms, Bollinger Bands are Bollinger Bands were developed by in the basically 95% confidence interval around the moving average 1980s. Given the relatively limited computing power and and they capture the mean plus and minus two standard technology during that period of time, it would be logical to deviations as long as the distribution is normally distributed. A develop a tool that is easy to calculate. One of the implicit buy signal is generated when prices are trending down and hit assumptions made in the use and interpretation of Bollinger the lower band. Similarly, a sell signal is generated when prices Bands is the normality of the distribution of stock prices. are trending up and hit the upper band. Interestingly, there is a great deal of empirical evidence that One of the implicit assumptions used in the interpretation suggests that asset prices, such as equity prices, are better and application of Bollinger Bands is normality and a significant characterized by skewed and fat-tailed distributions rather than amount of empirical evidence suggesting that security prices normal distributions. Therefore, it would be useful if we could such as stock prices are not normally distributed. Instead, they develop trading bands that could capture the empirical behavior show skewed and fat-tailed distribution exhibiting various without making any simplifying assumptions about the shape of degrees of skewness and kurtosis. Therefore, symmetrical the distribution. bands around the moving average such as Bollinger Bands may No single technical analysis tool is perfect and it is logical to not capture the skewness and kurtosis of the price movements complement an existing tool (i.e. Bollinger Bands in this case) adequately. with other metrics that could potentially fine-tune trading Bollinger Bands also make another implicit assumption signals. The Directional Movement Indicator (DMI) developed that stock prices tend to be mean-reverting as a buy signal by Welles Wilder is an interesting trading tool as it not only is generated when the price hits the lower band, while a sell has an indicator component but also a metric called Average signal is generated when the price hits the upper band. In other Directional Index (ADX) that determines whether or not prices words, prices that go outside the bands are considered to be too move in trends, which in turn will determine whether the extreme and therefore are expected to be pulled back to the

PAGE 20 IFTA.ORG IFTA JOURNAL 2013 EDITION moving average. A commonly identified limitation of Bollinger intuitive interpretation in the context of normal distribution. If Bands is the lack of an appropriate signal when prices move in the underlying prices follow a normal distribution, the 68-95-99 trends and in turn track along either the upper or lower band. empirical rule could be used to interpret the results, i.e. 68% of Therefore, a statistical indicator or metric that could measure the price data is within one standard deviation from the mean, whether the prices are oscillating or reverting would be very 95% of the price data is within two standard deviations from useful for improving the accuracy of trading signals generated the mean, etc. Therefore, prices going outside the bands are by these bands. considered “abnormal” and are expected to revert to the mean. Other trading bands that are typically covered in standard Table 1 below shows a summary of key features of the four technical analysis textbooks are Moving Average Envelopes commonly used and well-documented trading bands as well as (ENV), Keltner’s Channel (KC) and (DC). the proposed regime-switching trading bands (or Fong’s Bands, Moving Average Envelopes can be obtained by adding or to keep it simple). subtracting a pre-determined fixed percentage, e.g. 5% to a A key observation from the review of four commonly simple or exponential moving average. An obvious advantage used and well-documented trading bands described is that of ENV is its computational simplicity. However, the selection none of them can be considered as a comprehensive trading of the fixed percentage is perhaps too arbitrary, and there is no system (which is previously defined as a tool that has two key intuitive statistical interpretation of the pre-defined percentage components). Therefore, this paper will introduce the following and the resulting envelope. three specific new dimensions into our regime-switching Keltner’s Channel is made up of two bands plotted around trading bands: 1) historical simulation approach to generate an Exponential Moving Average (EMA) of typical prices. For the trading bands so that the empirical price distribution can the upper band, the (ATR) is calculated be captured, 2) an autocorrelation filter to make the bands over 10 days, doubled and added to a 20-day EMA. A similar a trading system by indicating whether the price is trending procedure can be used to calculate the lower band. There are or oscillating, and 3) a swing confirmation filter to deal with two key differences between Bollinger Bands and Keltner’s the situation where prices have a higher tendency to increase Channel. Firstly, Bollinger Bands use closing prices in the momentum rather than reverting to the mean. calculation while Keltner’s Channel uses typical prices in the calculation. Secondly, Bollinger Bands use standard deviation to 3. The Significance of Historical measure dispersion but Keltner’s Channel uses ATR to measure Simulation, Autocorrelation and variability. One should also observe that Keltner’s Channel has attempted to measure dispersion or variability by using Swing Filter in Fine-Tuning the ATR (which could capture gaps in the price series) rather than Decision-Making Process for the Use standard deviation, but it does not have a mechanism to tell of Trading Bands under what conditions the channels themselves should be used. Lastly, Donchian Channel (DC) is an indicator developed by Historical simulation is a non-parametric approach that Richard Donchian. It is formed by taking the highest high and is often used in the context of Value-at-Risk calculation for the lowest low over the last n days. If the stock price is above its trading books at banks. This approach makes no parametric highest high for the last n days, then a buy signal is generated. assumptions about the price distribution and also involves the On the other hand, if the stock price is below its lowest low for use of percentiles in measuring risk. One of the key assumptions the last n days, then a sell signal is generated. The key difference in historical simulation is that past history will repeat itself between Donchian Channel and the other three trading bands in the future. This assumption is consistent with one of the (i.e. BB, ENV and KC) is that trading signals generated by three key principles or foundations of technical analysis that Donchian Channel are based on breakout of the channels while “history will repeat itself” (i). For risk measurement purposes, trading signals from other trading bands are based on mean- banks focus on the tail or the lower percentile of the return reversion away from the channels. distribution. In the context of this research paper, the 5th One could see why Bollinger Bands are superior among percentile and the 95th percentile of the closing prices for the last these four trading bands as Bollinger Bands give rise to some 20 days will form the upper and lower bands, and the dispersion

Table 1

IFTA.ORG PAGE 21 IFTA JOURNAL 2013 EDITION

of stock prices will be measured by the interpercentile range In this case, the sample autocorrelation will be -1. The sign of (i.e. the difference between the 5th percentile and the 95th the sample autocorrelation tells the trader whether the prices percentile). We can also apply this concept to any percentile move in trends or not. Positive autocorrelation implies that and for any historical look-up period. More importantly, prices move in trends (i.e. trending regime) while negative depending on the skew and kurtosis of the price history, these autocorrelation implies that prices oscillate (i.e. oscillating bands that are generated by historical simulation capture the regime). The magnitude of the sample autocorrelation gives entire empirical distribution (including fat tails) and may be an indication of the strength of a trend. In the context of asymmetrical (unlike the Bollinger Bands). identifying a change from a trending regime to an oscillating The second key principle of technical analysis is the belief regime, if the autocorrelation changes from a positive sign or observation that “prices move in trends” (i). An objective to a negative sign, the direction of a price movement is more method of measuring whether prices are moving in trends or likely to reverse (more detailed explanation of the application oscillating is very important in technical analysis. For example, of the bands can be found in Section 4) and therefore the use of Welles Wilder’s Average Directional Index (ADX) is a well-known historically simulated trading bands is more likely to identify indicator that measures the strength of the trend as part of the optimal entry and exit points for a trade. DMI trading system. Perhaps a more direct and intuitive way One of the problems encountered through the use of the of measuring whether the price is trending or oscillating is to Bollinger Bands is that the price can track along the bands use the concept of autocorrelation. Autocorrelation is definitely when prices show exceptional momentum. In other words, the a well-known concept for statisticians or econometricians in price could touch the upper or lower band multiple times over time series analysis. However, autocorrelation is somewhat a short period of time, resulting in a false or premature buy or underused in technical analysis. It is not a standard topic in sell signal. In this case, even the autocorrelation filter might technical analysis textbooks used for professional technical not help us resolve the problem completely when prices show analysis designations, such as the CFTe program administrated extremely strong momentum. In this situation, a swing filter, by IFTA. Also, most technical indicators that are commonly first introduced by Arthur Merrill in his book, Filtered Waves, in available on popular websites such as Yahoo Finance or 1977, will be useful in generating a correct buy or sell signal. The stockcharts.com do not take autocorrelation into consideration. swing filter is basically a pre-determined percentage of price A brief overview of autocorrelation would be definitely movement and is generally considered as a breakout trading helpful in illustrating its usefulness in the trading system tool. In other words, prices are assumed to continue on its trend proposed in this paper. Autocorrelation, also known as serial until prices reverse more than the pre-defined threshold or correlation, measures the correlation of the data points over trigger. In the situation where prices hit the trading band with time. The following is the sample autocorrelation formula: positive autocorrelation, it means that the price is breaking out to levels observed in the extreme tails of the historical price distribution. The swing filter serves as a means to confirm the price reversal so that those extra miles from the trend can be captured more fully in the use of the trading bands. Detailed description of test cases is provided in Section 5.

where P is the closing price of a stock, h is the time lag, N is 4. The Methodology for Regime- the number of observations and P is the average closing price Switching Trading Bands using a over the respective period. In the context of measuring trends for daily prices over a short period of time, we are dealing with Historical Simulation Approach a time lag (h) of 1 specifically. It means that we are measuring (Putting them all together) whether an up-day is likely to be followed by another up-day “Transitions between a rising and falling trend are often and vice versa, where an up-day is defined as a day where the signaled by price patterns” (ii). Many of these patterns involve closing price is higher than the closing price on the previous day. the consolidation of prices manifested in the form of a zig- The following two diagrams illustrate the key interpretation of zag or whipsaw. These consolidation movements generally autocorrelation in the context of technical analysis: show negative autocorrelation, as shown in Section 3. The use of autocorrelation eliminates a cognitive recognition or assessment of these price patterns by the traders and it measures objectively whether the price is trending or oscillating, i.e. whether the price is going through the “transition” signaling a potential turn or reversal. In other Figure 1 Figure 2 words, the autocorrelation is positive when prices are trending and the autocorrelation is negative when prices are oscillating Figure 1 shows a situation where prices move in a perfect (i.e. giving a reversal signal). In the context of our regime- uptrend and the autocorrelation for this case is +1. On the other switching trading bands, if the autocorrelation is changing from Figure 1 Figure 2 hand, Figure 2 shows a situation where prices are oscillating in positive to negative by crossing the zero line (when the price is such a way that a daily price change in one direction is followed hitting the lower band or upper band), it indicates that the price by a daily price change of equal size in the opposite direction. may be transitioning or consolidating and therefore it is more

PAGE 22 IFTA.ORG IFTA JOURNAL 2013 EDITION

likely for the historically simulated trading bands to give the following diagram (Figure 3) illustrates how autocorrelation traders the optimal entry/exit points right before the reversal of could facilitate the decision-making process for buying with the price. trading bands (and the converse for selling is true as well): In the application of the regime-switching trading bands, A decision-tree generated from the regime-switching trading when prices move in a downtrend towards or hit the lower bands is outlined in Figure 4 below: band (for example), it will only trigger a buy signal when the In terms of the parameters and other assumptions for the price is in an oscillating regime (i.e. negative autocorrelation). bands, we will use the 5th and the 95th percentiles over the last Then, the swing filter rule of x% will be used to generate a buy 20 days as our trading bands, with an autocorrelation filter, signal, meaning that the prices need to revert to the upside for using the last 10 days of closing price data in the application more than x% before a legitimate buy signal is generated. The of the regime-switching trading bands, using a historical

Figure 3

Stock Price

negative autocorrelation when If the price reverts to the upside prices oscillate or consolidate for more than x%, it confirms during a transition period → that the price is breaking out use the lower band as entry from the consolidation pattern point to buy and a buy signal will be positive generated. autocorrelation when prices move in a downward trend Lower Band

Figure 4

Price hits or gets Price hits or gets close to the lower close to the upper band. band.

If autocorr is If autocorr is If autocorr is If autocorr is positive (i.e. non-positive positive (i.e. non-positive (i.e. trending (i.e. oscillating trending oscillating regime), then regime), then regime), then regime), then sell do nothing. buy only when do nothing. only when the the price price reverts reverts up by down by more more than x%. than x%.

IFTA.ORG PAGE 23

Price hits or gets Price hits or gets close to the lower close to the upper band. band.

If autocorr is If autocorr is If autocorr is If autocorr is positive (i.e. non-positive positive (i.e. non-positive (i.e. trending (i.e. oscillating trending oscillating regime), then regime), then regime), then regime), then sell do nothing. buy only when do nothing. only when the the price price reverts reverts up by down by more more than x%. than x%.

IFTA JOURNAL 2013 EDITION

simulation approach. The detailed methodology of how the 9 9 p pi ∑ i+1 bands and the autocorrelation are calculated mathematically ∑ i=1 a = i=1 b = will be discussed below. where 9 and 9 . It is important to note that a As mentioned in previous sections of this paper, the and b represent the respective averages over slightly different generation of the bands will involve the use of percentiles. time periods i.e. {R1, R2, …, R9} for the calculation of a vs. th The value of the P percentile of an ascending ordered price {R2, R3, …, R10} for the calculation of b. The application of the data series containing n data points with values such that regime-switching trading bands using empirical data and

v1 ≤ v2 ≤ … ≤ vn is defined as vp . First of all, the rank for a given the associated results on their performance will be discussed percentile P is calculated as follows: further in the next section. 5. Application of Regime-Switching

Trading Bands and Empirical Results where rank is further broken down into an integer Data from January 1, 2010 to December 31, 2010 is used in the component i and a decimal component d such that the sum of following three empirical test cases: (a) the S&P 500 index; (b) the two components is equal to the rank. Then the value of the a sector ETF (GLD); and (c) a single-name stock (Suncor). [Due to th P percentile (VP )will be calculated as follows: limited space we only show case a and b - the editor] We will apply both the regime-switching trading bands (i.e. Fong’s Bands) and Bollinger Bands to each of the above three cases. In order to test the effectiveness of the regime-switching trading bands, we will benchmark their performance against the Bollinger Bands by calculating the total cumulative return for both approaches under the three different test cases by Given the above parameterization and methodology, the lower implementing the following trading strategies: 1) If a buy signal band (L) will be calculated as the 5th percentile and the upper is generated using the bands, then we will put all the money band (U) will be calculated as the 95th percentile given the last to buy the equity product. 2) We will keep the long position 20 days of closing prices which are ranked in an ascending until a legitimate sell signal is generated. 3) When a sell signal

order such that p1 ≤ p2 ≤ … ≤ p20. By using the above percentile is generated, we will sell first to close an opening position and concept, the lower band and the upper band on a given day will then establish a new short sell position with all the money in the be calculated as follows: account. 4) We will not cover our short position until a legitimate buy signal is generated. 5) Similarly, immediately right after the short is covered, we will use all the money to establish a long position in the equity product as a result of a new buy signal. In summary, we will be alternating net long and net short positions using the legitimate buy and sell signals generated by the bands. Regarding the exact generation of trading signals using Bollinger Bands, a buy signal is generated when the closing price on a particular day is less than or equal to the lower Bollinger Band, and the trader will initiate a long position at the opening price on the next day. Similarly, a sell signal is generated when the closing price on a particular day is greater than or equal to Regarding the calculation of autocorrelation, only 10 out of the upper Bollinger Band, and the trader will initiate a short 20 closing prices are used in order to make the autocorrelation position at the opening price on the next day. value more sensitive to recent closing prices and preserve As outlined in the previous section, a buy signal using the the principle of harmonicity developed by J.M. Hurst at the regime-switching trading bands when the following conditions same time. Specifically, a time lag of 1 and the last 10 days of are met in sequence: 1) the closing price on a particular day

closing prices (i.e. R1, R2, …, R10) will be used to calculate the daily is less than or equal to the lower band, 2) the autocorrelation autocorrelation value for the regime-switching trading bands estimate is negative or close to zero, 3) the x% swing filter is th where pt is the closing price on the t day looking backwards. triggered when any percentage change on a single day, as well as The following formula shows how the daily autocorrelation any consecutive daily changes, exceeds the x% threshold, and 4)

value (r1)is calculated: a legitimate buy signal is generated and the trader will initiate a long position at the opening price on the next day. Similarly, a sell signal using the regime-switching trading bands when the following conditions hold in sequence: 1) the closing price on a particular day is greater than or equal to the upper band, 2) the autocorrelation estimate is negative or close to zero, 3) the x% swing filter is triggered when any percentage change on a single day, as well as any consecutive daily changes, exceeds the x%

PAGE 24 IFTA.ORG IFTA JOURNAL 2013 EDITION

threshold, and 4) a legitimate sell signal As a result, it is appropriate to use a index was trending from the middle of is generated and the trader will initiate a shorter time window (i.e. 5 days instead February to end of April as well as from short position at the opening price on the of 10 days) for calculating autocorrelation the beginning of September to end of next day. to fully capture the strong tendency to October. On the other hand, the index In terms of the parameters and other oscillate or mean-revert. On the other was oscillating between early May and assumptions for the bands (as mentioned hand, Bollinger Bands with default setting end of August. in Section 4), we will use the 5th and the (i.e. the simple moving average plus and Table 2 and Table 3 summarize the 95th percentiles over the last 20 days as minus two standard deviations using the trading results for the S&P 500 index our trading bands with an autocorrelation last 20 days of closing prices) will be used using regime-switching trading bands filter using last 10 days of closing price as the benchmark for our comparison of and Bollinger Bands respectively. The data and a 2% swing confirmation filter performance. use of regime-switching generated for the regime-switching trading bands. an annual cumulative return of 32.8% The 2% swing filter is triggered when any 5a) Equity index—S&P 500 while the use of Bollinger Bands only percentage change on a single day, as well It is important to note that the index generated 16.4%. The results are actually as any consecutive daily changes, exceeds showed both trending and oscillating not surprising as we expect that the the 2% threshold. As you will notice in behavior during the time window used autocorrelation filter will be able to Section 5c, Suncor showed a very secular in this case study. A close examination deal with both trending and oscillating trend and a strong tendency to oscillate. of Graph 1 and Graph 2 reveals that the behavior demonstrated by the index during 2010 effectively and explicitly. A minor technical point to note is that Graph 1 the closing prices were very close to S&P 500 Time Series with Fong's Bands the upper Bollinger Band during the 1300 first week of January but they did not

1250 hit the band according to our definition described earlier in this section (as 1200 the closing prices were not equal to or 1150 greater than the upper band).

1100 Table 2

1050

1000 5-Jul-10 7-Jun-10 4-Jan-10 8-Nov-10 1-Mar-10 19-Jul-10 2-Aug-10 1-Feb-10 6-Dec-10 11-Oct-10 25-Oct-10 21-Jun-10 12-Apr-10 26-Apr-10 18-Jan-10 22-Nov-10 15-Mar-10 29-Mar-10 16-Aug-10 30-Aug-10 15-Feb-10 13-Sep-10 27-Sep-10 20-Dec-10 10-May-10 24-May-10

Autocorrelation of Lag 1 with 10-day Moving Window

1.00 Table 3 0.50 0.00 -0.50 -1.00 5-Jul-10 7-Jun-10 4-Jan-10 8-Nov-10 1-Mar-10 19-Jul-10 2-Aug-10 1-Feb-10 6-Dec-10 11-Oct-10 25-Oct-10 21-Jun-10 12-Apr-10 26-Apr-10 18-Jan-10 22-Nov-10 15-Mar-10 29-Mar-10 16-Aug-10 30-Aug-10 15-Feb-10 13-Sep-10 27-Sep-10 20-Dec-10 10-May-10 24-May-10

Graph 2

S&P 500 Time Series with Bollinger's Bands 1350 5b) Exchange Traded Fund (ETF)—GLD 1300 It is important to mention that GLD 1250 (i.e. gold ETF) showed an exceptional 1200 upward trend during the year of 2010 1150 and it was overbought for a significant 1100 period of time. Graph 3 and Graph 1050 4 illustrate that the prices of GLD 1000 were tracking along both the regime- switching trading bands and Bollinger 5-Jul-10 7-Jun-10 4-Jan-10 1-Mar-10 8-Nov-10 1-Feb-10 19-Jul-10 2-Aug-10 6-Dec-10 21-Jun-10 11-Oct-10 25-Oct-10 18-Jan-10 12-Apr-10 26-Apr-10 15-Mar-10 29-Mar-10 22-Nov-10 15-Feb-10 16-Aug-10 30-Aug-10 10-May-10 24-May-10 13-Sep-10 27-Sep-10 20-Dec-10 Bands from early April to mid-May as well as early August to mid-October due

IFTA.ORG PAGE 25 IFTA JOURNAL 2013 EDITION to its extreme trending behavior. closing price was not equal to or greater showed very strong tendency to oscillate Table 4 and Table 5 summarize the than the upper band). in a secular pattern. Regime-switching trading results for GLD using regime- This test case is another great trading bands appeared to be able to switching trading bands and Bollinger example demonstrating that the cope with both trending and oscillating Bands, respectively. The use of regime- autocorrelation filter is an effective behavior quite effectively as they switching generated a reasonable tool in dealing with extreme trending generated profits for strong trending and cumulative return or profit of 17.4% behavior. strong oscillating equity products. On the while the use of Bollinger Bands resulted Based on the three test cases, regime- other hand, although the use of Bollinger in disappointing negative return or switching trading bands generated Bands generated a higher positive return loss of 6.3%. The use of Bollinger Bands consistent positive returns regardless of for the “oscillating” test case for Suncor generated premature trading signals as the underlying equity product (i.e. index than the use of regime-switching trading they cannot cope with trending behavior vs. sector ETF versus single-name equity) bands (i.e. 54.8% instead of 26.0%), effectively. If you look at the first period for all cases. On the other hand, Bollinger Bollinger Bands cannot effectively deal of explicit uptrend from early April Bands only generated positive returns with assets that are trending such as to mid-May more closely, the regime- for two of the three cases. In terms of GLD, leading to a negative return of switching trading bands did not generate dealing with trending vs. oscillating asset 6.3%. Our test cases demonstrated that a sell signal until June 21 so that a prices, we observe that GLD showed the regime-switching trading bands can significant portion of the strong uptrend heavy trends in 2010 while Suncor effectively resolve one of limitations of could be captured. On the other hand, Bollinger Bands generated a premature signal on April 7 and therefore the Graph 3 strong uptrend was not captured by GLD Time Series with Fong's Bands the trading strategies using Bollinger 140 Bands. Similarly, during second period 135 of explicit uptrend from early August 130 to mid-October, the regime-switching 125 trading bands capitalized on the trend 120 and generated a sell signal towards the 115 end of this trending period on October 110 19. However, Bollinger Bands generated 105 an early sell signal on September 14 100 in the middle of the strong uptrend. A 5-Jul-10

minor technical point to note is that the 7-Jun-10 4-Jan-10 8-Nov-10 1-Mar-10 19-Jul-10 2-Aug-10 1-Feb-10 6-Dec-10 11-Oct-10 25-Oct-10 21-Jun-10 12-Apr-10 26-Apr-10 18-Jan-10 22-Nov-10 15-Mar-10 29-Mar-10 16-Aug-10 30-Aug-10 15-Feb-10 13-Sep-10 27-Sep-10 20-Dec-10 closing price on January 11 was very close 10-May-10 24-May-10 to the upper Bollinger Band but did not hit the band according to our definition Autocorrelation of Lag 1 with 10-day Moving Window described earlier in this section (as the 1.00 0.50 Table 4 0.00 -0.50 -1.00 5-Jul-10 7-Jun-10 4-Jan-10 8-Nov-10 1-Mar-10 19-Jul-10 2-Aug-10 1-Feb-10 6-Dec-10 11-Oct-10 25-Oct-10 21-Jun-10 12-Apr-10 26-Apr-10 18-Jan-10 22-Nov-10 15-Mar-10 29-Mar-10 16-Aug-10 30-Aug-10 15-Feb-10 13-Sep-10 27-Sep-10 20-Dec-10 10-May-10 24-May-10

Graph 4

GLD Time Series with Bollinger's Bands 140 135 Table 5 130 125 120 115 110 105 100 5-Jul-10 7-Jun-10 4-Jan-10 8-Nov-10 1-Mar-10 19-Jul-10 2-Aug-10 1-Feb-10 6-Dec-10 11-Oct-10 25-Oct-10 21-Jun-10 12-Apr-10 26-Apr-10 18-Jan-10 22-Nov-10 15-Mar-10 29-Mar-10 16-Aug-10 30-Aug-10 15-Feb-10 13-Sep-10 27-Sep-10 20-Dec-10 10-May-10 24-May-10

PAGE 26 IFTA.ORG IFTA JOURNAL 2013 EDITION

Bollinger Bands mentioned in Section 2 of our paper. those who believe in technical analysis since our observation It is also worth noting that two of the three test cases do is actually consistent with one of the three key principles in not have fat-tails at all. In fact, the tails were even thinner than technical analysis, i.e. prices move in trends! the ones in a normal distribution as their excess kurtosis were negative during 2010 (as indicated in Table 6). As a result, the 6. Conclusion historical simulation aspect of the regime-switching trading The proposed trading bands involve a non-parametric bands, which was designed to capture the fat-tails, did not approach with the use of autocorrelation and swing filter to fine- contribute too much to the positive returns for GLD and Suncor. tune trading decisions. Based on our empirical tests, our regime- On the other hand, S&P 500 showed some excess kurtosis (i.e. fatter tails) and it is also interesting to observe at the same time switching trading bands using historical simulation generated that the regime-switching trading bands had the best absolute substantial positive returns for different types of equity performance and relative performance for this particular test products, and they improved trading performance for asset case. This evidence suggests that the historical simulation prices that have a strong tendency to move in trends relative to approach could add value to the regime-switching trading band the Bollinger Bands. This research does not only highlight the for asset prices that show skewed and fat-tailed distributions. significance of the use of autocorrelation in technical analysis but also outlines the next-generation trading bands for traders. Table 6 References i. Murphy, JJ, Technical Analysis of the Financial Markets, New York Institute of Finance, New York, 1999, Chapter 1, p. 2. ii. Pring, MJ, Technical Analysis Explained: The Successful Investor’s Guide to Spotting Investment Trends and Turning Points, McGraw Hill, New York, 2002, Chapter 5, p. 64. Another interesting observation is that the autocorrelations Bibliography are above the zero line (i.e. positive) most of the time for all the Murphy, JJ, Technical Analysis of the Financial Markets, New York three cases. This observation could suggest that equity indices, Institute of Finance, New York, 1999. equity ETFs and individual stocks tend to move in trends from Pring, MJ, Technical Analysis Explained: The Successful Investor’s Guide to day to day over a short period time as trends are statistically Spotting Investment Trends and Turning Points, McGraw Hill, New York, 2002. StockCharts.com, ChartSchool (http://stockcharts.com/school/doku. characterized by positive autocorrelation. This is great news to php?id=chart_school)

intalus.com TRADESIGNAL® GET AHEAD OF THE MARKET.

• The ultimate solution for Technical Analysis and algorithmic trading • Design and back-test almost any trading strategy before you risk your money • Scan thousands of securities according your own criteria or use some from hundreds of pre-installed indicators and strategies • Supports Bloomberg, Thomson Reuters, Trayport, Morningstar, your own data and others • Supports EasyLanguage™ • Made in Germany

ORDER YOUR FREE TRIAL TODAY | http://tradesignal.intalus.de

GRE T PERFORMANCE. WORLDWIDE. MADE IN GERMANY. SINCE 1990. BERLIN. BREMEN. FRANKFURT. HONG KONG. LONDON. LUXEMBOURG. MOSCOW. PARIS. ZURICH. FOLLOW US www.facebook.com/intalus www.twitter.com/intalus www.linkedin.intalus.com www.youtube.com/intalus www.blog.intalus.com

INTALUS is a registered trademark of Intalus GmbH, Germany | www.intalus.com. Tradesignal is a registered trademark of Tradesignal GmbH, Germany | www.tradesignal.com. All other trademarks are the property of their respective owners.

IFTA.ORG PAGE 27 IFTA JOURNAL 2013 EDITION

Using a Volatility Adjusted Stop Loss (VASL) to Enhance Trading Returns By Edward Rowson, CFTe, MFTA

Abstract of values drawn from a normal distribution are within one How can two traders using the same technical trading strategy standard deviation σ away from the mean; about 95% of the have such widely different investment returns? values lie within two standard deviations; and about 99.7% are This is a question that has played on my mind ever since I within three standard deviations. started trading and using technical analysis. During the course of this investigation a range of standard Within the discipline of technical analysis we are frequently deviations will be tested, and their ability to limit losses and presented with new and dynamic ways of interpreting charts increase profits will be compared. It is hard to judge before to help spot trends or reversal patterns but little time is spent conducting the study what to expect as using a large standard looking at the utilisation of these skills once acquired. There deviation stop loss would allow positions to run for longer as seems to be a disproportionate focus on the interpretation of the stop wouldn’t be particularly ‘tight’ to the price but at the price action over the implementation of a trade based on that same time this would result in being stopped out after large interpretation. This paper will focus on the latter point, and retracements from the high price thus giving back a lot of any more specifically on whether implementing VASL methodologies potential gains. On the other hand, a very close stop loss created within several simple trading strategies can produce enhanced using a low standard deviation would result in being stopped out returns when compared to a variety of standard stop loss close to any high price and maximise the gains but at the same strategies. Further investigation will be carried out to time mean that positions would be stopped out earlier as part of investigate whether optimisation of a VASL strategy is possible. the market ‘noise’. The results of this investigation should shed Hopefully this will illustrate the importance of employing a light on which technique performs best. risk management strategy when trading. Previous MFTA papers published in the annual Journal have looked at stop loss calculation. Most recently David Linton Introduction introduced in his MFTA paper, which was published in the 2008 It is the intention of this paper to investigate whether a IFTA Journal, titled ‘Optimisation of Trailing stop losses’, the dynamic approach to stop loss calculation, by studying the benefits of using trailing stop losses to optimise trading returns. historic volatility of the underlying security, performs better The concept of trailing stop losses will be incorporated in this than the more traditionaly used stop loss strategies. Across the study both within the volatility adjusted stop loss strategy industry traders and portfolio managers employ a wide variety and within the control environment to which the results will of stop loss methodologies and it would be impossible to test all be compared. The control environment will be, as mentioned of them in this paper so the investigation will concentrate on the previously, the ‘Percentage Drawdown’ stop loss strategy which most widely used stop loss strategy, specifically the ‘Percentage will be tested across a number of varying technical momentum Drawdown’ stop loss strategy. This strategy will act as the trading strategies which will be outlined in the control control to test the performance of the (VASL) methodology environment section. against. By back testing the different methodologies across a There are two variations for calculating the trailing stop loss. number of simple technical trading strategies it is intended to Both will be tested to make sure the most profitable control conclude whether there is any empirical evidence that supports is in place to judge the comparative performance of the VASL the use of a more dynamic approach to stop loss calculation. It methodology. The first shall be referred to as the ‘Dynamic’ is important to note that we are not testing the success of the approach, and is the one explained in David Linton’s paper which trading strategies in this paper but the success of the differing can be summarised by the following formula: stop loss strategies to limit the losses on losing trades and thus maximise the overall return on capital used. The overall Dynamic control stop loss investigation being ‘does the use of a VASL strategy enhance Initial stop loss = Price * (100-stop %) trading performance’? If (Price*(100-stop %)> SL) SL=Price*(100-stop %)-i.e. raise stop level For the sake of this paper volatility is quantified as the If Price

PAGE 28 IFTA.ORG IFTA JOURNAL 2013 EDITION this was the day that the price closed below the 10% trailing stop Moving average crosses which, on the 16th of May 2010, was 97.43 SAR. The technique of trading in the direction of the long-term trend The second variation for the calculating of a trailing stop can be implemented with two simple moving averages. The slower loss can be described as the ‘Static’ approach. This is where the moving average, using a longer calculation period, identifies the Initial stop loss spread, described above, is calculated to give an primary trend. The faster moving average is used for timing. The absolute number in terms of the price change, this price change following moving average cross indicators were used: then acts as the trailing stop loss, in equal increments to any 10/20 day simple moving average cross—Buy when the subsequent price rise. The Static stop loss can be summarised by 10 day moving average (fast trend line) crosses from below to the following formulas: above the 20 day (low trend line) moving average (MAV). 20/60 day simple moving average cross—Buy when the 20 Static control stop loss day crosses from below to above the 60 MAV. Initial stop spread = price*(% drawdown/100) 60/200 day simple moving average cross—Buy when the 60 For a long position the stop loss = Entry price – Initial stop spread day crosses from below to above the 200 MAV. Subsequent stop loss = closing high price since position inception Overbought/oversold RSI – Initial stop spread The Index (RSI) is a measurement that The main drawback to using the Static technique is that if a expresses the relative strength of the current price movement security doubles in price then the absolute drawdown will only as increasing from 0 to 100. For this paper an RSI of 25 will represent half the original move that the initial stop loss was indicate oversold and 75 overbought. based, this could increase the risk of being whipsawed as the Buy when the RSI rises above 25 from an oversold position stock price increases. Sell when the RSI falls below 75 from an overbought position Using another example: a long position in Marco Telecom was initiated on the 31st of August 2010 at the closing price of 145.5 Control environment parameters MAD, a 7% stop loss was used giving a stop loss of 135.3 MAD, The control environment has a number of constants so that thus the initial stop spread was equal to 10.2 MAD (the the results can be easily compared, these include: difference between the entry price and the stop loss). ƒƒ Each security will be back tested using the above trading The highest close was on the 28th of February 2011 at 159.90, strategies over the period commencing 1st January 2006 to the the trailing stop flat lined at 149.70 MAD (highest close – initial 12th of September 2011 inclusive. stop spread). The stop loss was triggered on the 20th of May 2011 ƒƒ The trailing stop loss from 1% to 20% inclusive will be tested when the price closed below the stop loss at 146.55. in increments of 1%, so 1%, 2%, 3% etc. Both techniques will be investigated to see if there is any ƒƒ The standardised position size of $500,000 rounded to the empirical evidence to support either technique as the preferred nearest share will be used. benchmark. ƒƒ The total return of the securities will be calculated so the stop losses aren’t triggered by the securities going ex-dividend. The Control Environment ƒƒ The moving averages will be calculated from the total return As previously mentioned, to test the theory that VASL can priced in USD not the underlying price. enhance trading profits, it is important to have a credible control ƒƒ Only buy signals will be tested. to test the hypothesis against. Unfortunately it is not possible to ƒƒ No commission or slippage will be included (the use of the top test every stop loss strategy used by traders, so for the purpose nine stocks by market capitalisation and liquidity will be used of this paper, we will look at one of the most popular stop loss to offer a realistic environment to open and close these trades). strategies: the ‘Percentage Drawdown’ stop loss will be tested. ƒƒ Closing prices will be used to open trades, mark the high for As already discussed, there are two variations to the the trailing stop loss, and used to close the trade. This may ‘Percentage Drawdown’ stop loss, the Static and the Dynamic. result in the trade being closed well below the actual stop loss Both variations will be initially tested to see if there is any should there be a large move on that day. As all trades will be marked difference. Whichever technique is most successful treated the same the results will still be comparable. will be used as a benchmark to compare the volatility adjusted ƒƒ Only trades that have been opened and subsequently closed returns against. will be included in this analysis; those trades that remained The control environment will consist of running a number open as at the 12th of September 2011 will not be included. of technical momentum strategies across the top nine stocks ƒƒ Multiple trades can be opened if the technical buy signal is (by market capitalisation) in the MENA (Middle East and North triggered while an existing trade is still open; each trade is African) region. Using the daily closing price the back test will treated independently. be run from the 1st of January 2006 to the 12th of September ƒƒ As the securities total return price is calculated daily in USD 2011 inclusive. The prices will be converted to USD and will using the daily FX rate, the positions will have currency be the total return price since the beginning of the period (1st exposure. January 2006) so that any dividend distribution is included; the following technical trading strategies were tested:

IFTA.ORG PAGE 29 IFTA JOURNAL 2013 EDITION

The Results compare the value for skewness with twice the standard error Interpretation of the control environment results of skewness. If the value of skewness falls within this range, the When comparing the success of the two controls it is skewness is considered not material. Twice the standard error of important to compare the relevant statistics, remembering that skewness = 2*0.183 = 0.366. the fundamental aim of any stop loss strategy is to minimise Using the data of the distribution of return for the Static losses, both on a total bases as well as the average loss. Secondly and Dynamic control environment the resulting skew for each is how that stop loss strategy impacts the overall return on control was: capital of the trading strategy. Key metrics to consider in the Static stop loss skew = 2.19 evaluation of the best control include 1) annualised return on Dynamic stop loss skew = 2.21 capital 2) average winning P&L to average losing P&L 3) the number of trades that exceed two standard deviations on the As the difference between the skewness of the two negative side and 4) the average return to capital risk employed. approaches is minimal and falls within the standard error of The Table below summarises the results of the two strategies: skewness no conclusion can be drawn from this analysis in its isolation. Hopefully it will prove to be an interesting factor for Table 1: Summary of results comparison when compared to the VASL Strategy. Static SL Dynamic SL To work out which control offers the best benchmark for Total Notional $5,376,500,000 $5,277,000,000 the volatility adjusted methodology a points system will be used, the control with the highest score will be the benchmark. Average Notional $500,000 $500,000 A number of key metrics have been compared and scored; the Total P&L $230,096,264 $253,849,884 results are shown in Table 2: Average P&L $21,398 $24,052 Average Return on Capital 4.3% 4.8% Table 2: Score table for control stop losses Average Annualised 23.6% 21.9% Static SL Dynamic SL Return on Capital Average P&L $21,398 $24,052 Total Return on Risk 4,794.34 5,234.36 Average Annualised Average Return on Risk 0.45 0.50 23.6% 21.9% Total Number of Trades 10753 10554 Return on Capital No of Winning Trades 4694 4445 Average Return on Risk 0.45 0.50 No of Losing Trades 6059 6109 Average Gain on Winning Trade $101,526 $114,230 Hit Ratio 43.7% 42.1% Average Loss on Losing Trade -$40,678 -$41,562 Winning Trades Profit $476,563,681 $507,752,715 % of trades greater than 2 STD 7.1% 6.3% Losing Trade Losses -$246,467,418 -$253,902,832 Average Winning P&L/Average 2.5 2.7 Average Gain on Winning Trade $101,526 $114,230 Losing P&L Average Loss on Losing Trade -$40,678 -$41,562 Skewness * 2.19 2.21 Average Winning P&L/ 2.5 2.7 Average Losing P&L Points Total 2½ 4½ Standard Deviation $102,621 $124,682 * As the difference in skewness is minimal the point will be shared No. of Trades Exceeding 766 659 Average Gain + 2SD Based on the above, the Dynamic stop loss will be the No. of Trades Less than control environment for the study. 1 1 Average Gain - 2SD From analysing the table of total distribution data and the Average Holding Period for subsequent skew analysis, it seems to support the potential 66 80 all Trades limitation of the Static stop loss control that was observed earlier. Average Holding Period for The data shows that the Dynamic control had a greater number of 104 134 Winning Trades trades that resulted in returns greater than 70%: 276 trades for the Dynamic control compared to 219 trades for the Static control. Average Holding Period for 37 41 This suggests that there is a limitation to using an absolute value Losing Trades trailing stop—as the share price rises the trailing stop becomes a Figures 1 and 2 illustrate the differing performance based on lower percentage of the current price, thus a smaller retracement these metrics: is required to stop out a Dynamic stop. This has had a detrimental The last thing to look at before making a decision on which effect on the most profitable winning trades. control should be used as the benchmark is to study the distribution of the returns between each strategy, and whether The Volatility Adjusted Stop Loss (VASL) methodology there is a difference in the ‘skewness’ of the returns. Intuitively it doesn’t make sense to maintain a rigid stop loss There is a standard error for whether a calculated skew methodology. For example, is it correct to have a similar stop is relevant; to calculate whether a skew is relevant, we can loss when trading in a utilities stock as that used trading a high

PAGE 30 IFTA.ORG IFTA JOURNAL 2013 EDITION

Figure 1: Comparing average annualised return on capital beta technology company? I believe the 60.0% answer is ‘no’, the stop loss should be adjusted to reflect the volatility of the 50.0% underlying. 40.0% It has been concluded from

30.0% the investigation into the control environment that the Dynamic 20.0% calculation proved most successful. 10.0% With this mind, the investigation will

Average Annualise Rtn on Capital % Rtn Average Annualise Capital on concentrate on looking at whether a 0.0% 1% SL 2% SL 3% SL 4% SL 5% SL 6% SL 7% SL 8% SL 9% SL 10% SL11% SL12% SL13% SL14% SL15% SL16% SL17% SL18% SL19% SL20% SL Dynamic calculated volatility adjusted stop loss better protects the trader from S top L oss Dynamic Static the downside without sacrificing the potential upside. The calculation for the stop loss is shown below: Figure 2: Comparing average annualised return on initial risk 1.80 Calculation of the Volatility 1.60

1.40 Adjusted Stop Loss (VASL)

1.20 1.00 Initial Stop Loss = Entry Price – σ 0.80

0.60

Retun Initial on RIsk 0.40

0.20

0.00

1% SL 2% SL 3% SL 4% SL 5% SL 6% SL 7% SL 8% SL 9% SL 10% SL 11% SL 12% SL 13% SL 14% SL 15% SL 16% SL 17% SL 18% SL 19% SL 20% SL

Stop Loss Static Stop Loss Dynamic Stop Loss n = Being the number of historic data points, and thus reflecting the Figure 3: Comparing the average annualised returns on capital for the various Volatility investment horizon of the trade, a Adjusted Stop Losses long term investor may use 200 days,

55% whereas a hedge fund trader interested 50% in monthly performance might use 20 45% days. This can also be hours or minutes 40% 35% for day traders; however it is worth 30% remembering that for an accurate 25% calculation of standard deviation a 20% sample of no less than 20 data points

Return Capital on 15% 10% is preferred. An analysis of time frame 5% optimisation will be looked at in more 0% detail later. 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00 The stop loss is recalculated on a

Number of Standard Deviations Volatility Stop Losses daily basis using the above formula and is always run from the highest closing price since the position was opened (for Figure 4: Comparing the average annualised returns on risk for the various Volatility Adjusted long trades) Stop Losses Thus 2.00 1.80 Subsequent stop loss = closing high price 1.60 since position inception – σ 1.40 1.20

1.00 Theory Put To The Test! 0.80 As with the control test, the 0.60 assumptions and the environment are 0.40 the same, including the position size and 0.20 calculation of daily prices in USD taking 0.00

0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00 into account dividends. The back test was run across the Volatility Stop Losses standard deviation curve, from 0.5

IFTA.ORG PAGE 31 IFTA JOURNAL 2013 EDITION standard deviations to 3 standard Table 4: Comparable percentage stop losses for the VASL and the control. deviations, in increments of 0.25 so 0.50, Standard Deviations 0.50 0.75 1.00 1.25 0.75, 1.0, 1.25 etc. Average percentage Stop Loss 4.83% 6.83% 8.63% 10.29% Interpretation of the results and Nearest Comparable Control 5.00% 7.00% 9.00% 10.00% comparing these to the control VASL AAROC 46.9% 40.0% 39.4% 35.3% As with the control back test, there Corresponding Control AAROC 32.8% 23.9% 23.7% 22.4% are some key metrics that are useful to VASL AROR 1.46 1.79 0.99 0.45 look at when evaluating the performance Corresponding Control AROR 0.51 0.42 0.50 0.49 of the VASL strategy. Before making any comparisons with the benchmark’s Standard Deviations 1.50 1.75 2.00 2.25 performance, it is interesting to obverse the varying performance of the differing Average percentage Stop Loss 12.35% 14.70% 14.50% 16.55% volatility levels used against each other. Nearest Comparable Control 12.00% 15.00% 15.00% 17.00% Figure 3 and Figure 4 show the average VASL AAROC 32.7% 29.0% 31.6% 25.7% annualised return on capital and the Corresponding Control AAROC 20.3% 18.3% 18.3% 27.2% average return on risk for the varying VASL AROR 1.00 0.96 1.35 1.51 stop losses tested. Corresponding Control AROR 0.43 0.39 0.39 0.63 Figure 3 and Figure 4 show similar trends to that of the controls tested, with the main difference being that Standard Deviations 2.50 2.75 3.00 they seem to be shifted up the chart, i.e. Average percentage Stop Loss 17.25% 11.07% 10.00% the VASL curve has a maximum value Nearest Comparable Control 17.00% 11.00% 10.00% of 46.9% and minimum value of 22.3%, VASL AAROC 22.3% 24.0% 22.9% whereas the control has a maximum Corresponding Control AAROC 27.2% 19.0% 22.4% value 53.7% and minimum 16.4%. If we VASL AROR 0.53 1.14 1.68 eliminate the extreme high and low Corresponding Control AROR 0.63 0.41 0.49 values, the average annualised return on AAROC – Average Annualised Return on Capital capital for the VASL test would be 31.2% AROR – Average Return on Risk whereas the control average would be 25.5%. A similar situation can be seen Table 5: Splits into regions (North Africa & Turkey and the Middle East) the results of the when comparing the return on risk; the control and VASL. range for the VASL is from 1.79 to 0.45 and for the control 0.83 to 0.39. Using a Middle East Volatility Adjust Control similar method of comparing the two, the Average Annualised Return on Capital 18.83% 14.26% average results (excluding the highest Average Return on Risk 1.95 0.41 and lowest numbers) are 1.17 and 0.48, North Africa & Turkey Volatility Adjust Control respectively. It is also interesting to note Average Annualised 12.43% 11.39% that there seems to be some differential Return on Capital in the average return on risk along the Average Return on Risk 0.07 0.61 standard deviation curve, average return on risk falls dramatically for 1.25 and 2.5 standard deviations with defined peaks Figure 5 : Comparing the average annualised returns on capital and the average return on risk at 0.75 and 2.25 standard deviations, for both the control and the VASL strategies whereas the trend for the control seems stable. This suggests the possibility for 2.00 1.80 further optimisation of the VASL. 1.60 Table 3 compares the results of the 1.40 Volatility Adjusted Stop Loss (VASL) vs. 1.20 the control : 1.00 0.80 Table 3: Comparable table of results for the 0.60 Return Risk Capital on VASL & the control. 0.40 0.20 Volatility Adjust Control 0.00 0% 5%

Average Annualised 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 28.2% 21.9% Return on Capital Return on Capital %

Average Return on Risk 1.17 0.50 Volatility Adjusted Stop Loss Control Stop Loss Skewness 2.91 2.21 Linear (Volatility Adjusted Stop Loss) Linear (Control Stop Loss)

PAGE 32 IFTA.ORG IFTA JOURNAL 2013 EDITION

The overall outperformance of utilising Figure 6: Comparing the average annualised returns on capital and the average return on risk for the VASL strategies. the VASL strategy can be best illustrated 2.00 in Figure 5, where the combination of 0.75 1.80 average annualised return on capital 3.0 1.60 and average return on risk have been 2.25 0.5 1.40 2.0 2.75 compared for both the control and the 1.20 1.00 1.50 1.0 VASL strategies. 1.75 Though the VASL strategy as a whole 0.80 0.60 2.50 looks to have outperformed the control, Return Risk Capital on 0.40 1.25 there are a few standard deviations that 0.20 did not, making it important to optimise 0.00 the VASL strategy to make sure the most 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% effective stop loss strategy is being Return on Capital % Volatility Adjusted Stop Loss Linear (Volatility Adjusted Stop Loss) employed. Before looking at the potential of optimising the VASL, it is worth Figure 7: Comparing how varying time frames influences the average annualised returns on reaffirming the performance of the VASL capital for the most successful VASL strategies compared to the benchmark. 60% The trend from the VASL results 50% suggests that the wider the stop loss (in terms of standard deviations) the lower 40% 30% the performance (measured by average capital annualised return on capital & average 20% return on risk). If this is the case it is 10% worth investigating what the average Return Average Annualised on 0% percentage drawdown was for each 5 10 15 20 30 40 50 60 100 120 150 200 250 300 standard deviation strategy. This way we Volatility Time Frame can make a direct comparison with those 0.5 0.75 2 2.25 percentage drawdowns from the control environment. If the average percentage Figure 8: Comparing how varying time frames influences the average return on risk for the drawdown from the VASL is comparable most successful VASL strategies with one of the control’s and has superior 0.90 results, it can be said that the VASL 0.80 0.70 strategy has ‘added value’ and acts as 0.60 conclusive empirical evidence to support 0.50 the use of a VASL strategy over the use of a 0.40 Percentage Drawdown stop loss strategy. 0.30 0.20

Table 4 shows the average percentage Average Return Risk on 0.10 stop loss for each of the standard 0.00 deviations tested and the comparable 5 10 15 20 30 40 50 60 100 120 150 200 250 300 control stop loss. Out of 22 data points the Volatility Time Frame control only managed to outperform the 0.5 0.75 2 2.25 VASL 4 times; these have been highlighted. Table 5 splits the performance of the Figure 9: Analysing the consistency of returns across multiple time frame control and the VASL results into regions, 1.00 both the Middle East and North Africa 0.90 20 Days & Turkey are shown. This table adds 0.80 further weight to support the utilisation 0.70 of a VASL strategy, as it shows that the 0.60 strategy performs across markets— 0.50 though it is important to note that in 0.40 0.30

the North Africa and Turkey region the Return Risk Capital on annualised return on capital is higher 0.20 than the control, but is produced on a 0.10 lower average return on risk. 0.00 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60%

Optimising the VASL strategy Return on Capital % 0.5 0.75 2 2.25 Within the VASL strategies it is Linear (0.5) Linear (0.75) Linear (2) Linear (2.25)

IFTA.ORG PAGE 33 IFTA JOURNAL 2013 EDITION important to distinguish the optimal standard deviation to it became apparent that the relative performance of the use as the results across the stop losses investigated varies longer term stop losses, for example the 3 standard deviation somewhat. Figure 6 illustrates the profitability within the VASL or the 20% drawdown stop loss, may have been negatively strategies. impacted by the decision to exclude trades that were still open From this it can be concluded that optimal strategies to use as the end of the time series, as these trades might have been are the 0.5, 0.75, 2.0 and 2.25 standard deviation within the carrying large unrealised profits which would impact the VASL strategy. overall results. To clarify whether the decision to exclude the open trades made a material difference, the back tests were How does time frame influence the results? re-run to include the unrealised P&L as of the 12 th of September Having found what the optimal volatility is within the 2011. The impact of this change on the VASL results was volatility adjusted stop loss, it is worth investigating if and how minimal, with the 1.75 standard deviation stop loss seeing the adjusting the time frame within the formula affects the results. largest impact with a change in the average annualised return After all a successful stop loss strategy should be able to be on capital from 29.0% to 28.7%. incorporated by traders as well as fund managers, both of which Further enhancement of trading returns could also result may have differing time horizons and investment objectives. if entry price was considered: an investigation into whether Taking the optimal stop losses of 0.5, 0.75, 2.0 and 2.25, an optimal entry price is possible by measuring where the the back tests were re-run with varying time frames and the stock is trading compared to its historical average, i.e. if the average annualised return and average return on risk were stock is trading between 2.0 and 2.5 standard deviations recorded. The results are illustrated in Figure 7 & Figure 8. from its 20 day average does that further optimise the The above investigation suggests that the most profitable performance. Also an investigation into whether position time frame is up to 20 days, after which the average return on sizing can impact overall returns, i.e. adjusting the size of risk declines steadily. A similar trend is seen for the average position to match the underlying securities volatility should annualised return on capital though there is a marked pick up also be a consideration. Unfortunately both these variables after 60 days for the 0.5 and 0.75 standard deviation strategies. are sufficiently complex to warrant their own investigation Though it is important to remember what was mentioned when and can’t be covered in this paper. discussing the measure of standard deviation, for a credible Given the need of further analysis, it can be concluded measure at least 20 data points are required. With this in mind from this investigation that employing a VASL strategy over the time frames of 5 to 15 will be excluded from analysis. more traditional stop loss strategies can enhance trading It is the aim to see which standard deviation performs performance. The best performing VASL strategy returned a 46% best across the varying time frames so both a Trader or Fund average annualised return on capital with a 146% average return Managers, who may have differing investment horizons can on risk, compared to the best control strategy which produced utilise the VASL strategy. Figure 9 best illustrates which 53.7% and 83% respectively, with the best performing VASL strategy performs most consistently; the graph suggests that averaging a $5,222 profit per trade compared to $3,994 average a standard deviation of 0.75 performs consistently across all profit for the control. Both the control and the VASL tests have time frames. Though a short term trader might consider using produced positive results, which is worth putting into context, a 0.5 measure with a 20 day time frame as it is by far the most when compared to a simple buy and hold strategy which would successful strategy on a standalone basis. have only returned a 5% annualised return on capital. I hope this paper has demonstrated that all traders and Conclusion fund managers alike should incorporate some sort of risk To summarise, a detailed analysis has been carried out to find management into their investment process. the most profitable control to test the theory: that a volatility adjusted stop loss strategy can enhance returns. The control Contact the author for questions: [email protected] test included back testing over 10,000 trades going back 5 years, across 9 different securities, and using 4 different technical References trading strategies. It was found that using the Dynamic control LINTON, D. Optimisation of trailing stop losses, IFTA Journal 2008 Edition. P38-46 produced the most profitable control environment. This control then acted as a benchmark to compare the volatility adjusted Bibliography stop loss performance against. KAUFMAN, P.J. New Trading Systems and Methods. Wiley, 2005 The results gave conclusive evidence supporting the use of MURPHY, J. Technical Analysis of the Financial Markets. New York Institute of Finance, 1999. a Volatility Adjusted Stop Loss (VASL) strategy over that of the FAITH, C.M. Way of the Turtle. McGraw-Hill. 2007 popular Percentage Drawdown stop loss strategy. BOLLINGER, J. Bollinger on Bollinger Bands. McGraw-Hill. 2002 Further investigation and comparison against other stop MARBER,B. Marber on Markets: How to make money from charts. loss strategies is needed to make any conclusions on whether Harriman House, 2007. the VASL strategy is the most successful stop loss strategy to Software and Data employ though. Further analysis could also be conducted to see Bloomberg (www.bloomberg.com) whether using real time prices (over end of day prices) yields Updata Technical Analyst Advance for Bloomberg (www.updata.co.uk) different results. Athena Systems (www.athenasystems.com) Back testing Spreadsheet (developed by Athena Systems with special On completing this investigation and reviewing the results thanks to Mr. Scott Sykowski).

PAGE 34 IFTA.ORG IFTA JOURNAL 2013 EDITION

Momentum Indicators: An Empirical Analysis of the Concept of Divergences By Stephan Belser, CFTe, MFTA

Abstract Divergence, or the non-confirmation of price movement by H1: A more meaningful extreme (high or low) in price should be a related market or , has long been a useful followed by a stronger price reaction in the opposite direction. part of the technician’s repertoire. Identifying divergences between price and technical indicators is an important aspect H2: The longer the duration between the divergence pairs, the of technical analysis. Especially when a trend enters a mature stronger the anticipated price reaction. phase the concept of divergences can help to identify potential reversals or trend changes. H3: The longer the duration between the divergence pairs, the The following paper provides an empirical investigation longer the period until the maximum price reaction sets in. about the concept of divergences using the DAX Index and the Gold Spot Price. The aim is not only to identify the divergence H4: Bullish divergence signals are as profitable as bearish between price and indicator development, but also to measure divergence signals. the resulting returns concerning the appearance of an expected price development. Furthermore, the study tries to answer the H5: RSI divergences are as profitable as MACD divergences. questions, how long it takes until the maximum return will be noticeable after an expected price development occurred and Definitions andM ethodology if there is a relationship between the type of divergence and the return realisation. This paper will show which assumptions Divergences about the forecasting qualities of a trading signal in the MACD Divergences act as a technician’s early warning system, or RSI indicator are possible. allowing valuable insight into the internal strength of a price move. One of the common uses of momentum indicators is the Introduction identification of trend completion when price and momentum In technical analysis a large variation of indicators is used begin to diverge. Divergence can be especially helpful for in order to describe the dynamics of market trends. On the traders as a leading indicator when assessing possible future one hand, an attempt is being made to depict overbought trend reversal situations in markets that are still trending but and oversold markets via the usage of an indicator’s extreme the strength of the trend is waning. A bullish divergence usually value. On the other hand, trend reversals or corrections are occurs in a down trend when new lows in price do not result anticipated through the concept of divergence between price in a new low in the indicator. This signifies that the prevailing and indicator development. The most commonly known downward trend is weakening and a trader should look for other indicators, which reveal the concept of divergence, are RSI and possible signs of a pending reversal to the upside. A bearish MACD. Many of the technical analysts agree on the fact that divergence usually occurs in an up-trend when new highs in divergences are actually working, but only a few empirical price do not result in a new high in the indicator. This signifies studies exist which prove the veracity of this concept. Within that the prevailing upward trend is weakening and a trader the debate the question is raised whether oscillators generate should look for other possible signs of a pending reversal to the too many false signals and hence reduce the forecasting downside. qualities of an indicator signal overall. However, before an One of the central premises of this method is that any evaluation of forecasting qualities is possible, it needs to move in price must be confirmed by the indicator. If the price be analysed whether or not a divergence is actually able to is making new highs while the indicator is unable to break old predict a reversal. If this is the case, the price ones, the strength of the primary trend is called into question. development after a divergence will become important and the following questions turn relevant: In how many cases does the bullish price is making a new low while the oscillator expected price reaction occur? How long does it take until the = maximum price reaction sets in? To identify the most effective divergence concurrently makes a higher low way of trading divergences, the following five hypotheses should help to get profitable divergence signals: bearish prices make a higher high but the indicator = divergence makes a lower high

IFTA.ORG PAGE 35 IFTA JOURNAL 2013 EDITION

Figure 1 demonstrates a typical divergence. In it, as the price by the trader’s comfort level and familiarity with the indicator’s sets new records, the indicator failed to better highs. particular nuances. In this study the RSI and the MACD are used to identify divergences because these two indicators are Indicators of Choice calculated and interpreted in a different way. For a valid signal Various indicators can be used to confirm the price to occur the oscillator should be in overbought or oversold development of a security or index. The choice is often governed territory as the divergence begins to take shape. By screening

Figure 1: Example of a divergence between price and indicator movement

Figure 2: Divergences (1. or 2.) can occur in different time frames

PAGE 36 IFTA.ORG IFTA JOURNAL 2013 EDITION potential trades in this way, the technician can avoid many new market situation exists a different duration between marginal patterns and unnecessary losses. the appearance of the price extreme and the extreme of the indicator in the analysed trend move. Unfortunately, Identification of divergences divergences are also quite subjective and often noticed only in The first step of this analysis is to track divergences. hindsight. Hypothesis 2 says that a longer duration between However, their discovery is rather difficult because for every the observed data points the more valuable should the trading

Figure 3: Detection of divergences looking 40 days back

Figure 4: The trading signal

IFTA.ORG PAGE 37 IFTA JOURNAL 2013 EDITION signal be. Therefore different time frames were observed to top or bottom. To capture potential cycle highs or lows in price get an idea in which time intervals the trader should look for the data point in the middle of each observed time frame are divergences (Figure 2). investigated. Rolling periods are used to address every new In order to be most flexible in tracking divergences, the market situation and potential new trading signals. (local) highs and lows are detected based on rolling 10-/20- In using different time frames it can not be excluded out that a /40-/80-day or week periods which are used as time intervals single divergence is detected in more than one observed time for the comparison of price and indicator movements. In the frame. This can be possible if there isn’t a more meaningful optimum case a divergence occurs hand in hand with a cycle combination of data points in the larger time interval.

Table 1: DAX Index—Bearish divergences on a daily basis

Cases of Average Trading profit after max. average trading profit Indicator No. of div- expected price days after time interval ergences 10 days 20 days 40 days 80 days max. in% reaction trading signal MACD 60 8.33% -0.62% -0.44% -0.45% 0.08% 40 0.47% 10 days RSI 43 9.30% -0.31% -0.40% -0.21% 0.09% 43 0.87% MACD 38 10.53% -0.49% -0.43% -0.18% 0.33% 36 0.81% 20 days RSI 60 10.00% -0.39% -0.39% -0.31% 0.05% 44 0.79% MACD 54 7.41% -0.70% -0.60% -0.49% -0.03% 41 0.26% 40 days RSI 83 8.43% -0.54% -0.52% -0.36% -0.11% 44 0.43% MACD 71 11.27% -0.57% -0.30% -0.28% 0.08% 38 0.56% 80 days RSI 101 9.90% -0.49% -0.43% -0.29% -0.05% 46 0.47%

Table 2: Gold Spot Price—Bearish divergences on a daily basis

Cases of Average Trading profit after max. average trading profit Indicator No. of div- expected price days after time interval ergences 10 days 20 days 40 days 80 days max. in% reaction trading signal MACD 61 24.59% -0.15% 0.53% 0.59% 0.54% 51 1.48% 10 days RSI 25 20.00% 1.71% 0.11% 0.53% 0.42% 59 1.71% MACD 22 31.82% 0.23% 0.20% 0.38% 0.94% 58 1.31% 20 days RSI 38 18.42% -0.07% 0.20% 0.19% 0.70% 50 1.15% MACD 32 34.38% 0.38% 0.56% 0.94% 1.09% 49 1.95% 40 days RSI 49 20.41% -0.08% 0.24% 0.30% 0.66% 54 1.22% MACD 35 22.86% 0.24% 0.28% 0.30% 0.77% 60 1.19% 80 days RSI 56 19.64% -0.14% 0.18% 0.31% 0.60% 53 1.13%

Table 3: DAX Index—Bullish divergences on a daily basis

Cases of Average Trading profit after max. average trading profit Indicator No. of div- expected price days after time interval ergences 10 days 20 days 40 days 80 days max. in% reaction trading signal MACD 47 36.17% 0.61% 1.54% 2.12% 4.28% 64 5.07% 10 days RSI 16 25.00% 0.19% 0.79% 1.51% 2.69% 60 4.24% MACD 16 37.50% 0.43% 1.51% 1.75% 3.93% 61 4.47% 20 days RSI 26 34.62% 0.81% 1.84% 3.35% 4.10% 67 5.97% MACD 18 33.33% 0.55% 1.43% 2.68% 3.92% 62 4.44% 40 days RSI 39 30.77% 0.78% 1.91% 2.74% 3.57% 63 5.06% MACD 19 15.19% -0.31% -0.31% 0.33% 1.54% 73 1.83% 80 days RSI 47 29.79% 0.72% 1.56% 2.25% 3.40% 65 4.68%

PAGE 38 IFTA.ORG IFTA JOURNAL 2013 EDITION

To find divergence signals, first peaks in price and in Divergences in daily data momentum were defined and identified. A peak was defined as a high in price or momentum such that the day on which Bearish divergences the high took place was preceded by lower or equal values in The analysis of the bearish divergences shows for the DAX the following five data points. Once price peaks and dips were Index that in all cycles the maximum price movement sets in identified the indicator value and the closing price built the within a narrow time range. Furthermore, the return analysis first pair of data points. Once a matching pair was found, the depicts that the maximum price reaction for the DAX Index algorithm looked back up to 40 closing prices (depending on the becomes already noticeable between the 36th and the 46th day analysed cycle) for an earlier pair of lower peaks or higher dips after the trading signal (Table 1). in price and indicator values. The two pairs of matching peaks In all analysed time frames RSI identifies more divergences or dips were then checked for divergence (Figure 3). compared to the MACD except of the 10-day interval. In this study a bearish divergence is defined as a higher peak In this 10-day interval the maximum trading profit for the in price matched by a lower or equal peak in the indicator and RSI appears. Most profitable considering both indicators is the bullish divergence as a lower dip in price matched by a higher or 20-day interval. Here were higher trading profits compared to equal dip in momentum. The trading signal is set when a valid the other intervals evident. divergence in the analysed time interval appears. Per definition In the cases of expected price moves the MACD indicator no new extreme in the upcoming five days or weeks is allowed achieves enhanced results in the 20- and 80-day interval. In (Figure 4). contrast to that for the RSI in the 10- and 40-day intervals Once divergences were found the next step is to see how better results are noticeable. It is remarkable that in expanding often prices move into the anticipated direction. For this the time interval improved results are not achievable. the following points in time are examined: 10, 20, 40 and 80 Comparing the maximum profit there are no major differences periods after the trading signal, as well as the point in time at between both indicators. which the maximum average profit occurs. So trend reversals The results for the Gold Spot Price are dissimilar. Maximum or at least a correction within a trend are of interest. trading profit appears in the 40-day interval for the MACD Furthermore it is examined after how many days the model with a return of 1.95%, whereas the best window for the maximum price movement sets in. This is done by calculating all RSI indicator is the 10-day time frame (Table 2). Most profitable trading profits after a recognized trading signal if the defined for the Gold price in terms of expected price reaction is the 40- stop was not hit. In this study profitability is understood as the day interval. average trading profit in the analysed time frame. Comparing the maximum profit of the Gold price to the DAX Index there are significant higher returns after bearish Money Management divergences for Gold. Additionally the maximum price reaction Money management is based on a from-entry stop. The set between the 49th and the 60th day after the signal and so risk per trade is defined by the difference between the entry considerably later versus the DAX Index. price and the stop level. Volatility in weekly data is regularly substantially higher compared to daily data. Therefore the Bullish divergences stop-level for daily data is tighter versus weekly data. The The analysis of bullish divergences shows positive returns methodology does not allow a risk of more than 1% per trade in for the DAX Index in nearly all time intervals. The returns only daily data and 5% in weekly data. differ in the amplitude of the maximum trading profit and for the time at which this price reaction occurs. Slightly negative Period of Investigation and Object of Study results are for the 80-day interval given (Table 3). The study examines daily and weekly data of the German It is noteworthy that the maximal return sets in between the blue chip stock market index DAX and the Gold Spot Price 60th and 73rd day after the divergence completely independent of between 01/01/1980 and 03/31/2011. This window is large the observed time frames. This leads to the assumption that the enough to offer a great variety of up- and down-trends inside maximum price reaction can always be expected after 60 to 73 different overall market conditions (secular structure). Within days regardless of the quoted low and analysed cycle. this time frame, the closing prices of daily and weekly data The period of time until the maximum trading profit sets are analyzed in terms of bullish and bearish divergences. in increases generally in the longer time intervals for both Furthermore, it is examined how many bullish and bearish indicators. Considering the 80-day interval, the maximum divergences occur and on which trading days after the return is recorded in the MACD indicator after 73 days and in the divergence the maximum average price movement sets in. RSI indicator after 65 days. In numbers, this observation equals As indicators, the RSI with a default of 14 and the MACD with a plus of 1.83% after a trading signal in the MACD indicator and a a default of 26, 12, 9 are used. Concerning the RSI indicator plus of 4.68% in the RSI indicator. Once more the 20-day interval it is worth mentioning that its signal is only used if the value is best for both indicators taking the probability of expected on the first extreme (extreme in the indicator / first pair) price reaction and maximum trading profit into account. records more than 65 or less than 35 points while forming the For the signals in the Gold price movement a positive profit divergence, in order to avoid worthless signals in the neutral development can be seen in all time intervals (Table 4). Most range of the indicator. profitable results came out in the 20-day interval for the MACD and in the 40-day interval for the RSI. By looking on larger

IFTA.ORG PAGE 39 IFTA JOURNAL 2013 EDITION time frames trading results for bullish divergences can not Furthermore, cases of expected price reaction for bullish be improved. The 80-day interval is obviously too large for daily divergences lie usually higher than for bearish divergences. tracking the most valuable signals in both indicators. Finally the maximum price development after positive divergences for Gold Divergences in weekly data occurs remarkably earlier than the development after bullish Bearish divergences divergences in the DAX Index. But it needs to be pointed out Looking on the DAX Index only bearish divergence signals that the average trading profit after Gold divergences is smaller within the MACD indicator are valuable for 10-, 20- and 40-week compared to the returns after DAX divergences. time frames. The biggest price reaction occurs in the 10-week

Table 4: Gold Spot—Bullish divergences on a daily basis

Cases of Average Trading profit after max. average trading profit Indicator No. of div- expected price days after time interval ergences 10 days 20 days 40 days 80 days max. in% reaction trading signal MACD 68 22.06% -0.38% 0.34% 0.52% 1.31% 53 2.22% 10 days RSI 22 18.18% 0.13% 0.22% 0.40% -0.21% 42 0.76% MACD 21 38.10% -0.18% 1.14% 1.06% 2.62% 59 4.25% 20 days RSI 24 16.67% -0.45% -0.32% -0.08% -0.50% 46 0.25% MACD 32 34.38% 0.10% 1.42% 1.41% 2.74% 57 4.17% 40 days RSI 47 25.53% 0.02% 0.77% 0.92% 1.09% 54 3.08% MACD 46 39.13% 0.14% 0.78% 1.28% 1.88% 54 3.95% 80 days RSI 57 26.32% -0.11% 0.54% 0.61% 0.98% 54 2.61%

Table 5: DAX Index—Bearish divergences on a weekly basis

Cases of Average Trading profit after max. average trading profit Indicator No. of div- expected time interval ergences days after price reaction 10 0 weeks 2 weeks 40 weeks 80 weeks max. in% trading signal MACD 11 18.18% -3.72% -2.98% 1.31% 2.05% 54 4.20% 10 weeks RSI 7 0.00% -5.00% -5.00% -5.00% -5.00% -- -5.00% MACD 6 33.33% -2.62% 1.66% 5.40% 3.11% 21 9.60% 20 weeks RSI 9 11.11% -4.55% -1.74% -1.86% -2.80% 21 -1.09% MACD 7 14.29% -4.42% -0.80% -0.96% -2.17% 21 0.03% 40 weeks RSI 15 6.67% -4.73% -3.04% -3.12% -3.86% 21 -2.65% MACD 10 10.00% -4.60% -2.06% -2.17% -3.02% 21 -1.48% 80 weeks RSI 16 6.25% -4.75% -3.16% -3.23% -3.76% 21 -2.80%

Table 6: Gold Spot—Bearish divergences on a weekly basis

Cases of Average Trading profit after max. average trading profit Indicator No. of div- expected time interval ergences days after price reaction 10 0 weeks 2 weeks 40 weeks 80 weeks max. in% trading signal MACD 5 20.00% 0.50% -0.10% 1.56% 5.10% 78 5.86% 10 weeks RSI 4 25.00% -1.50% -1.03% -1.64% 0.00% 80 0.00% MACD 3 33.33% -2.22% -3.73% -3.56% -1.01% 78 -0.62% 20 weeks RSI 5 20.00% -2.20% -1.82% -2.31% -1.00% 80 -1.00% MACD 1 0.00% -5.00% -5.00% -5.00% -5.00% -- -5.00% 40 weeks RSI 6 16.67% -2.66% -2.35% -2.76% -1.66% 80 -1.66% MACD 3 0.00% -5.00% -5.00% -5.00% -5.00% -- -5.00% 80 weeks RSI 7 28.57% -1.07% -1.08% -0.11% 1.86% 77 2.19%

PAGE 40 IFTA.ORG IFTA JOURNAL 2013 EDITION interval 54 weeks after the trading signal and quotes a trading RSI signal in the 80-week time frame are profitable (Table 6). profit of 4.20%. In contrast to that, the price developments The maximum profit for these two trades comes in very late following a bearish RSI signals have been consequently stopped compared to the results for the DAX Index. out a few weeks after the signal. In this case negative trading results in all time intervals were received. Bullish divergences Bearish divergences appeared not frequently higher in longer The analysis of the 10-week cycle in the DAX Index reveals intervals. Thus, 10 bearish signals occured in the MACD and within the whole observation period only two signals in the RSI 16 bearish divergences are identified in the 80-week interval. and nine in the MACD indicator. For both RSI cases the stop was In the case of a bearish divergence, it can be stated that the hit so apparently a 10-week interval is too short for the RSI to expected negative price development always sets in very fast identify bullish divergences. For the bullish MACD divergences after the signal occurred (Table 5). positive trading results are noticeable for all observed points in In the 20- and 40-week time frames the maximum trading time. Maximum price development following the MACD signal profit appeared after 21 weeks. In all time intervals higher occurs after 72 weeks and quotes 29.85% (Table 7). probabilities for the expected price reaction for the MACD During the investigation of the 20-week interval one bullish indicator are discoverable compared to the RSI, since there were divergence in the MACD indicator and three divergences in the RSI only negative trading results for the RSI indicator. So the RSI is indicator revealed. Maximum price development occurs for the obviously not a good indicator to detect bearish trend reversals MACD 65 weeks after the divergence and notes 6.85%. For the RSI on a weekly basis. a maximum trading profit of 12.13% after 59 days is receivable. In weekly data, is in case of a MACD divergence, the 20-week For the 40-week interval the analysis presents two bullish time frame is the most valuable interval. A longer time frame to divergences in the MACD indicator and five in the RSI indicator. track divergences is generally less successful. In terms of price developments the results are consistently When analysing Gold it is remarkable that the trading results positive within both indicators. For the MACD the maximum are pretty disappointing after a bearish divergence in weekly price development occurs after 80 weeks quoting 21.04% and data. Only the MACD signal in the 10-week interval and the 15.69% for the RSI 70 weeks later.

Table 7: DAX Index—Bullish divergences on a weekly basis

Cases of Average Trading profit after max. average trading profit Indicator No. of div- expected time interval ergences days after price reaction 10 0 weeks 2 weeks 40 weeks 80 weeks max. in% trading signal MACD 9 66.67% 1.51% 3.44% 9.28% 25.81% 72 29.85% 10 weeks RSI 2 0.00% -5.00% -5.00% -5.00% -5.00% -- -5.00% MACD 1 100.00% 6.85% 7.07% 2.73% 0.73% 65 6.85% 20 weeks RSI 3 33.33% -2.49% -0.90% 4.72% 9.42% 59 12.13% MACD 2 50.00% 1.38% 8.93% 12.92% 21.04% 80 21.04% 40 weeks RSI 5 40.00% -0.94% 3.03% 8.00% 14.07% 70 15.69% MACD 3 66.67% 4.65% 14.09% 23.00% 26.22% 60 28.42% 80 weeks RSI 6 50.00% 1.08% 6.60% 13.86% 17.82% 60 20.27%

Table 8: Gold Spot—Bullish divergences on a weekly basis

Cases of Average Trading profit after max. average trading profit Indicator No. of div- expected time interval ergences days after price reaction 10 0 weeks 2 weeks 40 weeks 80 weeks max. in% trading signal MACD 7 85.71% 3.77% 6.13% 9.12% 19.59% 50 25.59% 10 weeks RSI 4 50.00% 1.78% 1.24% 0.74% 6.05% 43 13.06% MACD 6 50.00% 3.80% 4.84% 6.58% 5.01% 33 12.15% 20 weeks RSI 4 50.00% 1.78% 1.24% 0.74% 6.05% 43 13.06% MACD 10 50.00% 2.40% 3.28% 3.51% 3.96% 35 10.11% 40 weeks RSI 8 50.00% 4.14% 3.95% 2.38% 2.70% 32 12.96% MACD 10 40.00% 0.77% 2.04% 1.59% 1.98% 40 7.72% 80 weeks RSI 13 46.00% 2.22% 2.82% 1.20% 2.83% 35 9.94%

IFTA.ORG PAGE 41 IFTA JOURNAL 2013 EDITION

The analysis of the 80-week interval reveals results which the highest probability of expected price reaction, as well as the are similar to the 40- week interval. Data for both indicators largest trading profits for both indicators, occurs. For the Gold show a consistently positive price development. The maximum price the results get poorer the longer the time interval gets. return of 28.42% occurs 60 weeks after the MACD signal while This is a completely different result compared to the DAX Index. the RSI signal initiates a maximum price performance after 60 weeks, noting a profit of 20.27%. Conclusions All in all MACD identifies fewer divergences compared to the This analysis—based on the DAX Index and the Gold Spot RSI indicator in the 20-, 40- and 80-week cycle. Therefore the Price—shows that changes in market trends can be identified probability of success is superior for the MACD signals. with the help of the concept of divergence. At the same time The maximum price reaction for the Gold price sets in a tight stop management is necessary because a divergence between 32 and 50 weeks after the trading signal (Table 8). In the only indicates a trend reversal. To summarize, the following 10-week time frame the most profitable results are evident. Here statements about profitable trading divergences can be made.

Figure 5: Maximum trading profit for daily data in different time frames

Figure 6: Maximum trading profit for weekly data in different time frames

PAGE 42 IFTA.ORG IFTA JOURNAL 2013 EDITION

Per definition, hypothesis H1 estimates that a more Interestingly the duration to the maximum trading profit meaningful extreme should be followed by a stronger expected appears independent from the analysed time frame. For the DAX price reaction in the opposite direction. Concurrently, the Index the maximum trading profit for daily and weekly bearish duration between the observed pairs usually expands as divergences comes in earlier compared to bullish divergences. described in hypothesis H2. However, the analysis can not offer For the Gold price the maximum trading profit for daily bearish better results in observing larger time frames and consequently and bullish divergences appeared approximately after the same is not able to approve hypothesis H1 and hypothesis H2. In duration. Comparing daily data of the Gold price and the DAX the majority of all cases the 20- and 40-day or week intervals Index maximum trading profits in bearish divergences appear provide the best results. Divergences in larger time intervals do much later for Gold compared to the DAX. In the case of bullish not lead to more profitable trading results, neither in daily nor divergences the opposite effect was measurable. In weekly data in weekly data. Therefore, we do not have to analyse larger time maximum trading profits in bullish divergences emerge much frames to track divergences (Figure 5 and 6). earlier for Gold compared to the DAX (Figure 7 and 8).

Figure 7: Duration to the maximum trading profit in daily data for different time frames

Figure 8: Duration to the maximum trading profit in weekly data in different time frames

IFTA.ORG PAGE 43 IFTA JOURNAL 2013 EDITION

The results show that bullish divergences are basically more Literature profitable than bearish divergences for both indicators. This is Bleymüller, Josef / Gehlert, Günther / Gülicher, Herbert (2004): Statistik für Wirtschaftswissenschaftler, 14. Auflage, München. true for daily and weekly data. A possible explanation is the fact Breuer, Wolfgang / Gürtler, Marc / Schuhmacher, Frank (2002): that down-trends are normally more erratic than up-trends. Risikoverfahren: in Coche, Joachim / Stotz, Olaf: Asset Allocation, Köln, S. 165-191. Table 9: Summary of daily data Fama, Eugene F. (1976): Foundations of finance: portfolio decisions and securities prices, Oxford. DAX GOLD Florek, Erich (2000): Neue Trading Dimensionen, FinanzBuch Verlag GmbH 2000. bearish bullish bearish bullish Garz, Hendrick / Günther, Stefan / Moriabadi, Cyrus (2002): Portfolio- divergence divergence divergence divergence Management, Theorie und Anwendung, 1. Auflage, Frankfurt am Main. Most Gast, Christian (1998): Asset Allocation – Entscheidungen im Portfolio- Management, Bern. profitable 20-day 20-day 40-day 40-day Gügi, Patrick (1995): Einsatz der Portfolioopimierung im Asset-Allocation- time frame Prozess: Theorie und Umsetzung in der Praxis, Bern, Stuttgart, Wien. Preferred RSI or Heckmann, Tobias (2009): Markttechnische Handelssysteme, RSI MACD MACD quantitative Kursmuster und saisonale Kursanomalien, Josef EUL Verlag, indicator MACD 1. Auflage 2009. Expected Murphy, John (2004): Technische Analyse der Finanzmärkte, Finanzbuchverlag, 3. aktualisierte Auflage, 2004. max. trading 36-46 days 60-73 days 49-60 days 42-57 days Paesler, Oliver (2006): Technische Indikatoren simplified, profit after Finanzbuchverlag, 2006. Pflüger, Patrick (2011): Möglichkeiten und Grenzen einer Aktienkursprognose mittels der technischen Analyse, GRIN Verlag, 1. Table 10: Summary weekly data Auflage 2011. Poddig, Thorsten / Dichtl, Hubert / Petersmeier, Kerstin (2003): DAX GOLD Statistik, Ökonometrie, Optimierung: Methoden und ihre praktischen bearish bullish bearish bullish Anwendungen in der Praxis, Uhlenbruchverlag, Bad Soden, 3. Auflage. divergence divergence divergence divergence Pring, Martin J. (2003): Momentum Explained, 2, McGraw Hill 2003. Rose, Rene (2006): Enzyklopädie der technischen Indikatoren Rene Rose Most (Hrsg.) FinanzBuch Verlag GmbH, 1. Auflage 2006. profitable 20-week 80-week 10-week 10-week Schmidt-von Rhein, Andreas (1996): Die moderne Porfoliotheorie im time frame praktischen Wertpapiermanagement: Eine theoretische und empirische Analyse aus Sicht privater Kapitalanleger, Bad Soden. Preferred MACD or Spremann, Klaus (2000): Portfoliomanagement, München. MACD MACD RSI indicator RSI Vogel, Friedrich (2000): Beschreibende und schließende Statistik: Formeln, Definitionen, Erläuterungen, Stichwörter und Tabellen, 12. Expected Auflage, München. 21-54 59-80 77-80 32-50 max. trading Wilder, Welles (1978): New Concepts in Technical Trading Systems. weeks weeks weeks weeks profit after Accordingly in an up-trend more bearish divergences may be Articles stopped out while prices continue to climb. Nevertheless there Cartwright, D. (1991): RSI as an Exit Tool, Technical Analysis of Stocks & Commodities, vol.9, no.4, 1991, pp.160-162. are some consistent differences between the DAX Index and the Drinka, Thomas P. & Kille, Steven L. (1987): ‘Profitability of Selected Gold price. Technical Indicators’, Technical Analysis of Stocks & Commodities, vol.5, no.9, Comparing RSI and MACD the study demonstrated that 1987, pp.288- 291. Drinka, Thomas P. & Kille, Steven L. (1987): RSI profitability with money MACD divergences are on principle more profitable than RSI management Profitability of Selected Technical Indicators, Technical divergences. Furthermore it is to conclude that in daily Gold Analysis of Stocks & Commodities, vol.5, no.9, 1987, pp.162-164. price data better results for the MACD indicator arise, whereas Ehlers, John F. (1986): Optimizing RSI with Cycles, Technical Analysis of Stocks & Commodities, vol.4, no.1, 1986, pp.26-28. mixed outcomes for the DAX Index can be recorded. In weekly Ehlers, John F. (1991): MACD Indicator Revisited, Technical Analysis of data the investigation showed for Gold in the shorter intervals Stocks & Commodities, vol.9, no.10, 1991, pp.391-395. better MACD results and for the larger time frames superior RSI Hall, Herbert S. (1991): The Common (But Useful) RSI, Technical Analysis of Stocks & Commodities, vol.9, no.8, 1991, pp.325-327. trading profits are measurable (Table 9 and 10). McWhorter, Lawson W. (1994): Price/Oscillator Divergences, Technical These findings show for the DAX Index and the Gold Spot Analysis of Stocks & Commodities, vol.12, no.1, 1994, pp.45-48. Price how prices develop after divergences. First, this analysis Merrill Arthur A. (1991): Testing Indicators Technical Analysis of Stocks & Commodities, vol.9, no.5, 1991, pp.187-188. helps to predict the duration between the trading signal after Nicholas, John (1984): Momentum Indicators and Market Cycles, Technical a divergence and the resulting maximum trading profit. And Analysis of Stocks & Commodities, vol.2, no.6, 1984, pp.209-211. second, assumptions about the forecasting probability of Pring, Martin J. (1997): Reverse Divergences and Momentum, Technical Analysis of Stocks & Commodities, vol.15, no.12, 1997, pp.568-571. receiving an expected return can be made. Star, Barbara (1996): Hidden Divergence, Technical Analysis of Stocks & This study illustrates the power of price to oscillator Commodities, vol.14, no.7, 1996, pp.285-289. divergences. They are difficult to identify, but are usually worth the effort. It is also important to be aware of the overall Software technical picture (including chart patterns and trendlines) Bloomberg MS-Excel as well as likely areas of support and resistance. Combining classical chart analysis, proper risk management techniques and a healthy respect for the dominant trend with the concept of divergences can lead to prosperous trading profits.

PAGE 44 IFTA.ORG IFTA JOURNAL 2013 EDITION

Momentum Success Factors By Gary Antonacci

Abstract With respect to fixed income, Jostova, Niklova and Philipov Momentum is the premier market anomaly. It is nearly (2010) show that momentum strategies are highly profitable universal in its applicability. Rather than focus on momentum among non-investment grade corporate bonds. High yield, applied to particular assets or asset classes, this paper explores non-investment grade corporate bonds have, by far, the highest momentum with respect to what makes it most effective. We do volatility among bonds of similar maturity. This may indicate this first by introducing a hurdle rate filter before we can initiate that credit default risk is also a proxy for volatility risk. long positions. This ensures that momentum exists on both an The real estate market and long-term Treasury bonds are absolute and relative basis and allows momentum to function as also subject to high volatility due to their sensitivity to interest a tactical overlay. We then explore the factor most rewarded by rate risk and economic conditions. Gold is subject to high momentum—extreme past returns, i.e., price volatility. volatility as well, due to its response to economic stress and We identify high volatility through the paired risk premiums uncertainty. In this paper, we will examine momentum with in foreign/U.S. equities, high yield/credit bonds, equity/ respect to high volatility associated with all four markets— mortgage REITs, and gold/Treasury bonds. Using modules of equities, bonds, real estate, and gold. asset pairs as building blocks lets us isolate volatility related Before proceeding, we need to distinguish between risk factors and successfully use momentum to effectively relative and absolute momentum. When we consider two harvest risk premium profits. assets, momentum is positive on a relative basis if one asset has appreciated more than the other has. However, Introduction momentum is negative on an absolute basis if both assets Momentum is the tendency of investments to persist in have declined in value. their relative performance. Assets that perform well over Most momentum researchers use long and short a 6 to 12 month period tend to continue to perform well into positions to examine both the long and short side of a market the future. The momentum effect of Jegadeesh and Titman simultaneously. They are therefore only concerned with relative (1993) is one of the strongest and most pervasive financial momentum. It makes little difference whether the studied phenomena. Researchers have verified its existence in U.S. markets go up or down, since short momentum positions hedge stocks (Fama and French (2008)), industries (Moskowitz and long ones and vice versa. Relative momentum can help one Grinblatt (1999), Asness, Porter and Stevens (2000)), styles identify when assets will remain strong relative to others, but if (Lewellen (2002), Chen and DeBondt (2004)), foreign stocks a market as a whole is in a downtrend, then all related assets are (Rouwenhorst (1998), Chan, Hameed and Tong (2000), Griffen, likely to sustain losses. Ji and Martin (2005)), emerging markets (Rouwenhorst (1999)), When looking only at long side momentum, however, it country indices ( Bhojraj and Swaminathan (2006), Fama and is desirable to be long only when both absolute and relative French (2011)), commodities (Pirrong (2005), Miffre and Rallis momentum is positive, since momentum results are highly (2007)), currencies (Menkoff, Sarno, Schmeling, and Schrimpf regime dependent. Fortunately, there is a way to put the (2011)), international government bonds (Asness, Moskowitz odds in one’s favor with respect to momentum profits from and Pedersen (2009)), corporate bonds (Jostova, Nikolova and long positions. Positive momentum means an asset that has Philipov (2010)), and residential real estate (Beracha and Skiba outperformed over the past twelve months is likely to continue (2011)). Since its first publication, momentum has been shown to doing so. To determine absolute momentum, we see if an asset work going forward in time (Grundy and Martin (2001), Asness, has outperformed Treasury bills over the past year. Since Moskowitz, and Pedersen (2009)) and back to the Victorian age Treasury bills are expected to always remain positive, if our (Chabot, Ghysels and Jagannathan (2009)). chosen asset shows positive relative strength with respect to There has also been considerable study of exogenous factors Treasury bills, then it too is likely to continue showing a positive that influence momentum. In a recent paper, Bandarchuk, Pavel return. In our momentum match ups, if our selected assets do and Hilscher (2011) reexamine some of the factors that have not show positive relative strength with respect to Treasury previously been shown to impact momentum in the equities bills, then we select Treasury bills as an alternative investment market. These include analyst coverage, illiquidity, price level, age, until our other assets are stronger than Treasury bills. Treasury size, analyst forecast dispersion, credit rating, r squared, market- bill returns thus serve as both a hurdle rate before we can to-book, and turnover. The authors show that all these factors invest in other momentum assets, as well as a safe, alternative are proxies for extreme past returns, or high volatility. Greater investment until our assets show both relative and absolute momentum profits simply come from more volatile assets. positive momentum.

IFTA.ORG PAGE 45 IFTA JOURNAL 2013 EDITION

Besides incorporating a safe alternative when market that time frame.4 One often skips the most recent month during conditions are not favorable, our module approach has another the formation period in order to to disentangle the momentum important benefit. It imposes diversification on our momentum effect from the short-term reversal effect returns that may be portfolios. If one were to throw all assets into one large pot, as related to liquidity or microstructure issues with equity returns. is often the case with momentum investing, and select the top Momentum results for non-equity assets are actually better if few momentum candidates, there is a good chance some of the one does not skip a month, since they suffer less from liquidity selected assets would be highly correlated with one another. issues. Because we are dealing with gold, fixed income and real Asset pair modules help ensure that different asset classes (and estate, as well as equities, we adjust our positions monthly but risk factors) receive portfolio representation. without skipping a month. We first apply momentum broadly to the MSCI U.S. and 2. Data and Methodology EAFE+ stock market indices in order to create a baseline equities All monthly return data begins in January 1974, unless momentum portfolio. In bonds, we incorporate credit risk otherwise noted, and includes interest and dividends. For volatility using the High Yield Bond Index, which has an average equities, we use the MSCI US, MSCI EAFE, and MSCI ACWI exUS duration of just over four years. We match High Yield Bonds indices. These are all free float adjusted market capitalization with the Barclays Capital U.S. Intermediate Credit Bond Index, weightings of large and midcap stocks. The MSCI EAFE Europe, the next most volatile intermediate term fixed income index. Australasia and Far East Index includes twenty-two major Real estate has the highest volatility over the past five years developed market countries, excluding the U.S. and Canada. of the eleven U.S. equity market sectors tracked by Morningstar. The MSCI ACWI exUS, i.e., MSCI All Country World Index ex US, Real Estate Investment Trusts (REITs) make up most of this includes twenty-three developed market countries (all but the sector. The Morningstar real estate sector has both mortgage U.S.) and twenty-one emerging market countries. MSCI ACWI and equity based REITs. We similarly use both. exUS data begins in January 1988. We create a composite data Our final high volatility risk factor focuses on economic series called EAFE+ that is comprised of the MSCI EAFE Index stress and uncertainty. For this, we use the Barclays Capital U.S. until December 1987 and the MSCI ACWI exUS after that time.2 Treasury 20+ year Bond Index and gold. Investors generally hold The Bank of America Merrill Lynch High Yield Cash Pay Bond these as safe haven alternatives to equities and fixed income Index that we use begins in November 1984. Data prior to that securities subject to credit default risk. is from Steele System’s Corporate Bond High Yield Average. All other bond indices are from Barclays Capital. REIT data is from the 3. Equity/Sovereign Risk National Association of Real Estate Investment Trusts (NREIT). Equities are the mainstay of momentum investing. Gold returns using the London PM gold fix are from the World Therefore, our first momentum module is composed of the MSCI Gold Council. Treasury bill returns are from newly issued 90-day U.S. and EAFE+ indices. It gives us broad exposure to the U.S. auctions as reported by the U.S. Treasury. No deductions have equity market, as well as international diversification. Volatility been made for transaction costs. The average number of switches comes from the equity risk premium, as well as from sovereign per year for our modules is 1.4 for foreign/U.S.equities, 1.2 for risk. Table 1 presents the summary statistics from January 1974 high yield/credit bonds, 1.6 for equity/mortgage REITs, and 1.6 for through December 2011 for the equity indices, our momentum gold/Treasuries, making momentum transaction costs negligible. strategy, and momentum excluding the use of Treasury bills as a The average annual expense ratio for a representative group of hurdle rate and alternative. exchange-traded funds corresponding to the indices we use is .25%, and their annual transaction costs are .05%. Table 1 Equities 1974-2011 The most common metric for evaluating investment Momentum Momentum US EAFE+ strategies is the Sharpe ratio. It is most appropriate when you exT Bills have normally distributed returns or quadratic preferences. Annual Return 15.79 13.46 11.49 11.86 Yet the returns from financial assets usually are not normally Annual Std Dev 12.77 16.17 15.86 17.67 distributed. Tail risk may be much greater than one expects under an assumption of normality. Quadratic utility implies Annual Sharpe .73 .45 .35 .33 that as wealth increases, you become more risk averse. Such Max Drawdown -23.01 -54.56 -50.65 -57.37 increasing absolute risk aversion is not consistent with rational Skewness -.24 -.34 -.38 -.32 investor behavior. Yet despite its limitations, the Sharpe ratio is based on expected utility theory, while most alternative performance The average of the annual return of both equity indices is measures lack a theoretical underpinning. Therefore, we use 11.68%, and their average annual standard deviation is 16.77%. the Sharpe ratio as a risk adjusted metric, but also present The annual return and standard deviation of our momentum skewness and maximum drawdown as additional risk factors.3 strategy are 15.79% and 12.77%. This is a remarkable 400 Maximum drawdown here is the greatest peak to valley equity basis point increase in return and 400 basis point reduction erosion on a month end basis. in volatility from the market indices. Momentum doubles Most momentum studies use either a six or a twelve-month the Sharpe ratio and cuts the drawdown in half. Momentum formation period. Both perform well, but since twelve months results without the use of Treasury bills are better than the is more common and has lower transaction costs, we will use index averages’, but not nearly as good as the results that come

PAGE 46 IFTA.ORG IFTA JOURNAL 2013 EDITION from using momentum with Treasury bills as a trend filter and difference is the credit default risk of their respective holdings, alternative asset. as reflected in their average credit ratings.

Figure 1: Equities Momentum 1974-2011 Table 3: Intermediate Fixed Income Index Rating Duration Volatility Treasury AA 4.0 3.7 Government A 5.3 3.3 Government/ A 3.9 3.4 Credit Aggregate A 4.4 3.6 Bond Credit A 4.4 5.4 High Yield B 4.1 14.0

In Table 4, we see that applying momentum to both bond indices produces almost a doubling of the indices’ individual Sharpe Most momentum research on equities looks at individual ratios, from .51 and .54 to .97. securities sorted by momentum. All three of the fully disclosed, publically available momentum equity programs use momentum Table 4: Intermediate Term Fixed Income 1974–2011 Momentum High . Credit applied to individual stocks. It might be useful therefore to see Momentum how our module approach stacks up against individual stock exTBills Yield Bonds momentum. Annual Return 10.49 10.39 10.29 8.53 The AQR momentum index is composed of the top one-third Annual Std Dev 4.74 6.13 8.67 5.19 of the Russell 1000 stocks based on twelve-month momentum with a one-month lag. Positions are adjusted quarterly. The AQR Annual Sharpe .97 .74 .51 .54 small cap momentum index follows the same procedure with the Max Drawdown -8.20 -12.08 -33.17 -11.35 Russell 2000. Table 2 shows the AQR results, as well those of our Skewness -.10 .15 -.49 -.45 Equity module, from when the AQR indices began in January 1980.

Table 2: AQR Index versus Equity Module 1980–2011 Momentum gives the same profit as from high yield bonds alone, AQR . AQR . US MSCI Equity but with less than half the volatility, one-quarter the drawdown, Large Cap Small Cap Module and one-fifth the negative skewness. Our momentum strategy Annual Return 14.75 16.92 12.42 16.43 even has a lower standard deviation and drawdown than the Annual Std Dev 18.68 22.44 15.60 13.13 investment grade, credit bond index. Momentum without the use of Treasury bills does not give nearly as much improvement in Annual Sharpe .45 .46 .41 .75 reducing volatility or drawdown. Although investors most often Max Drawdown -51.02 -53.12 -50.65 -23.01 apply momentum to equity investments, fixed income investors Skewness -.55 -.61 -.61 -.22 should take note of the potential here for extraordinary momentum returns of an extra 196 basis points per year over intermediate term credit bonds, and with less volatility. The AQR indices show a modest advantage over the broad U.S. market index. However, our Equity module results are Figure 2: Credit Risk Momentum 1974–2011 considerably better. The differences here are understated, since AQR estimates that their index results should be reduced by transaction costs of .7% per year. 4. Credit Risk Table 3 lists the average credit rating, average bond duration, and annualized standard deviations over the past five years for the most common intermediate term fixed income indices maintained by Barclays Capital. The U.S. High Yield Bond Index has by far the highest volatility. Its standard deviation over the past five years is 14.0, compared to 5.4 for the next highest one belonging to the U.S. Intermediate Credit Bond Index. Since their average bond durations are about the same, the main cause of their volatility

IFTA.ORG PAGE 47 IFTA JOURNAL 2013 EDITION

6. Economic Stress One possible explanation for this impressive performance is that the credit default risk associated with high yield bonds may Figure 3: REIT Momentum 1974–2011 be less when these bonds are in a positive relative and absolute momentum situation. Their risk premium is still able to flow to investors under favorable market conditions identified through momentum, when their actual risks may not be very high. 5. Real Estate Risk We next look for additional asset classes with risk factors related to high volatility. Table 5 is a list of the eleven Morningstar equity sector indices with their annualized standard deviations over the five years ending 12/31/11.

Table 5: Morningstar Sectors Sector Volatility Real Estate 33.9 Basic Materials 29.7 Economic stress is another volatility-based risk factor. Gold Financial Services 29.4 and long-term Treasury bonds respond to that stress. Both Energy 27.2 often react positively to weakness in the economy. Economic weakness tends to produce falling nominal interest rates, Consumer Cyclicals 24.4 which raises bond prices. Gold is usually strong when long-term Industrials 24.1 Treasury yields fall. There is some differentiation and separation Technology 22.6 for momentum purposes, since gold responds more favorably to inflationary expectations, while Treasuries respond positively to Communication Services 21.0 deflationary pressures. Health Care 15.9 Gold is not highly correlated with most other assets, which Utilities 14.8 makes it particularly useful from a portfolio point of view. Consumer Defensive 12.6 Gold, like Treasuries, is not only a good hedge and diversifier; it is also a safe haven during times of economic turmoil (Bauer and McDermot (2010)). A safe haven is an asset that remains At the top of the list is real estate with a standard deviation uncorrelated or negatively correlated with another asset or of 33.9%. The Morningstar Real Estate sector includes both portfolio in times of market stress or turmoil. equity and mortgage REITS. We will also use both to give us Table 7 shows the economic stress module results. Gold’s some separation and differentiation for momentum selection average annual standard deviation of 20.00 since 1974 is almost purposes. the same as the 20.71 volatility of mortgage REITs, which is the Table 6 shows an annual rate of return of 16.78% from our highest of all our assets. Treasury bond annual volatility of 10.54 momentum strategy applied to REITs. This is the highest return is higher than the 8.67 volatility of the High Yield Bond Index. of our momentum modules so far. It is also significantly higher than the returns of the individual equity and mortgage REIT Table 7 Economic Stress Momentum 1974–2011 indices of 14.6% and 8.28%. The momentum standard deviation Momentum Momentum Gold Treasury and drawdown are substantially lower than the indices exTBills Bonds themselves. The momentum Sharpe ratio is .77, compared to Annual Return 16.65 16.31 9.22 9.90 .48 and .13 for the REIT indices. As with our other modules, the Sharpe ratio and volatility of momentum without Treasury bills Annual Std Dev 17.04 17.65 20.00 10.54 are less than the Sharpe ratio and volatility of the portfolio with Annual Sharpe .59 .56 .17 .39 Treasury bills. Max Drawdown -24.78 -36.82 -61.78 -20.08 Table 6: REITs 1974-2011 Skewness .68 .62 .60 .38 Momentum Equity Mortgage Momentum exTBills REIT REIT Momentum raises annual profits substantially to 16.65%, Annual Return 16.78 16.80 14.60 8.28 from a return of 9.22% with gold and 9.90% with Treasuries. The Sharpe ratio increases from .17 and .39 to .59. Annual Std Dev 13.24 16.56 17.39 20.71 Annual Sharpe .77 .62 .48 .13 7. Robustness Checks Max Drawdown -23.74 -48.52 -68.30 -42.98 We can divide our 38 years of data into two equal sub- periods. Table 8 shows performance from January 1974 through Skewness -.75 -1.13 -.72 -.22 December 1992 and from January 1993 through December 2011.

PAGE 48 IFTA.ORG IFTA JOURNAL 2013 EDITION

Table 8: Performance 1974–1992 and 1993–2011 Equities Equities Credit Credit REIT REIT Stress Stress 1974–1992 1993–2011 1974–1992 1993–2011 1974–1992 1993–2011 1974–1992 1993–2011 Annual Return 17.82 13.79 12.97 8.06 14.74 18.85 21.39 12.09 Annual Std Deviation 13.0 12.54 4.77 4.63 11.34 14.91 19.41 14.23 Annual Sharpe .68 .79 .96 1.00 .54 .96 .61 .59 Maximum Drawdown -17.81 -11.33 -4.94 -8.20 -17.91 -23.74 -24.27 -24.78 Skeness -.11 -.39 .98 -1.31 -.20 -1.01 .99 -.36

Table 9 Formation Periods—12 and 6 Month Equities Equities Credit Credit REIT REIT Stress Stress 12 Mo 6 Mo 12 Mo 6 Mo 12 Mo 6 Mo 12 Mo 6 Mo Annual Return 15.79 14.67 10.49 10.95 16.78 16.67 16.65 11.79 Annual Std Deviation 12.77 12.33 4.74 4.98 13.24 13.61 17.04 16.35 Annual Sharpe .73 .68 .97 1.01 .77 .74 .59 .35 Maximum Drawdown -23.01 -22.54 -8.20 -7.65 -23.74 -34.59 -24.78 -24.27 Skewness -.24 -.13 -.10 -.17 -.75 -.76 .68 .87

Table 10: Returns and Volatility 1974–2011 Volatility Return Utilization Rate Weighted Avg Return Momentum Return U.S. 15.86 11.49 37.7% Equities EAFE+ 17.67 11.86 39.7% TBill 1.19 5.89 22.6% 10.41 15.79 Credit 5.19 8.53 19.5% Credit Risk Hi Yield 8.67 10.29 55.3% TBill 1.19 5.89 25.2% 7.66 10.49 Equity 17.38 14.60 46.9% REITs Mortgage 20.71 8.28 26.8% TBill 1.19 5.89 26.3% 10.63 16.78 Gold 20.00 9.02 39.0% Stress Treasuries 10.54 9.90 43.2% TBill 1.19 5.89 17.8% 8.84 16.65 Average 9.39 14.93

Sharpe ratios remain high for all the modules during both sub- Table 9 compares performance using twelve-month and six­ periods. They are very consistent across both sub-periods for the month formation periods. equities, credit risk, and economic stress modules. Performance is very good for both periods. The stress module does better with a twelve-month formation period, while Figure 4 Economic Stress Momentum 1974–2011 equities, credit bonds, and REITs perform about the same using either six or twelve months. 8. Momentum Return versus Weighted Average Return Table 10 shows momentum return along with average return weighted by each asset’s percentage usage within a module. By comparing momentum returns to weighted average returns, we see that momentum and our timing filter create 59% higher profits.

IFTA.ORG PAGE 49 IFTA JOURNAL 2013 EDITION

Table 11: Results Summary 1974–2011 Annual Annual Annual Maximum Return Std Dev Sharpe Drawdown Skewness Kurtosis US 11.49 15.86 .35 -50.65 -.38** 4.83** Equities EAFE+ 11.86 17.67 .33 -57.37 -.32** 4.21** High Yield 10.29 8.67 .51 -33.17 -.49** 10.01** Credit Risk Credit Bond 8.53 5.19 .54 -11.35 .45** 9.53** Equity REIT 14.60 17.39 .48 -68.30 -.72** 11.57** REITs Mortgage REIT 8.28 20.71 .13 -42.98 -.22* 8.29** Economic . Gold 9.22 20.00 .17 -61.78 .60** 6.72** Stress Treasuries 9.90 10.54 .39 -20.08 .38** 4.81** Equities 15.79 12.77 .73 -23.01 -.24* 4.83** Momentum Credit Risk 10.49 4.74 .97 -8.20 -.10 8.96** Modules REITs 16.78 13.24 .77 -23.74 -.75** 8.33** Economic Stress 14.27 16.60 .48 -24.78 .73** 11.86** Composite –. Momentum 14.90 7.99 1.07 -10.92 -.45** 6.56** Equal Weight Non-Momentum 9.95 8.19 .50 -26.77 -.54** 7.00** **p<.01 * p<.05 for normality

Figure 5: Momentum versus Benchmarks 1974–2011 9. Module Characteristics The modules are in Treasury bills from 17.8% of the time with the economic stress module to 26.2% of the time with the REIT module. Singular match ups of Treasury bills with each asset, rather than with paired combinations of assets, would lead to higher Treasury bill utilization and lower expected profits. On the other hand, more than two assets within a momentum module could make it more difficult to isolate singular risk factors. We might find higher volatility by further segmenting a market or asset class. For example, we could split equities into individual countries and find additional volatility. However, this Figure 6: Composite Momentum 1974–2011 granularity would come at the cost of individual country risks dominating our desired risk factor of high volatility from sovereign markets. Greater segmentation might also reduce the benefits we get from diversification by using multiple rather than singular assets. Table 11 is a results summary of each asset and risk module, as well as the equally weighted composite of all four modules. As a benchmark, we also present the equal weighted portfolio of all nine assets (two per module plus Treasury bills) without the use of momentum. The composite momentum portfolio gives an annual return of 14.90% with a standard deviation of 7.99%. The Sharpe ratio of this portfolio is 1.07, versus

PAGE 50 IFTA.ORG IFTA JOURNAL 2013 EDITION

Sharpe ratios of .73, .97, .77, and .59 for Table 12: Sharpe Ratios Gold the individual equity, credit risk, REIT, Gold and economic stress modules. The return Treasury Bond of this composite momentum portfolio is Economic Stress Momentum exTBill 50% higher than the return of the equal Economic Stress Momentum weight, all asset benchmark portfolio. The momentum portfolio has double the MSCI EAFE+ Sharpe ratio (1.07 vs. 0.50) and less than MSCI US half the drawdown (-10.92 vs. -26.77). Equities Momentum exTBill These are impressive results using just Equities Momentum twelve-month momentum, a simple trend following filter, and a balanced Mortgage REIT portfolio of U.S and foreign equities, Equity REIT credit and high yield bonds, REITs, gold REIT Momentum exTBill and Treasury bonds. REIT Momentum The risk profile of our dynamic asset mix bears some resemblance to those of High Yield Bond static risk parity portfolios. Successful Intermediate Credit Bond Credit Risk Momentum exTBill risk parity programs can offer 300-400 Credit Risk Momentum basis points of additional annual return when leveraged to the same level of risk Composite Momentum (10.6 annual standard deviation) as a conventional balanced portfolio (See 0 0.2 0.4 0.6 0.8 1 1.2 Dalio (2011)). Our composite momentum portfolio, leveraged to the same level Table 13: Largest Equity Drawdowns of risk as a conventional balanced Date MSCI US MSCI World World 60/40 Composite portfolio, shows a remarkable 950 basis Momentum points of incremental return, while 3/74-9/74 -33.3 -30.8 -21.5 2.1 avoiding derivatives, counterparty risk, 4/02-9/02 -29.1 -25.6 -13.2 7.5 and tracking error. Table 12 shows the Sharpe ratios of 11/7-2/09 -50.6 -53.6 -34.7 -2.8 each of our assets and modules, as well World 60/40 is composed of 60% MSCI World Index and 40% US Aggregate Bond Index. as the composite momentum portfolio. Table 13 shows performance versus Table 14: Correlation Coefficients 1974–2011 several benchmarks during the three worst periods of equity erosion over the With Treasury Bill Hurdle Rate past 38 years of data. We see that the Credit Risk REITs Stress composite momentum portfolio, through its trend following characteristics, Equities .35 .29 .17 is itself a safe haven from market Credit Risk .40 -.05 adversity. REITs .10 10. Correlations Without Treasury Bill Hurdle Rate Table 14 shows the correlations of Equities .40 .45 .13 the modules, as well as the correlations Bonds .46 .06 if Treasury bills are not included in the REITs risk modules. We have already seen that .12 Treasury bills are very helpful in raising return and lowering volatility. Now we see that they also are beneficial from a 11. Portfolio asset classes is via Markowitz mean portfolio point of view, since they lower Considerations variance portfolio optimization. This most correlations. According to PIMCO uses quadratic programming algorithms ( Page (2010)), risk factor correlations Given the inequalites in the Sharpe to determine efficient portfolios that are lower than asset class correlations. ratios and correlations of our four offer the highest potential return at They are also more robust with respect modules, we may not want to allocate any given level of expected volatility, or, to regime shifts. Our lower risk module capital equally to all of them. The conversely, the lowest volatility at any correlations support those findings. traditional way to allocate varying given level of expected return. There amounts of capital across different are, however, several potential pitfalls

IFTA.ORG PAGE 51 IFTA JOURNAL 2013 EDITION with this approach. First, the process is very sensitive to the this volatility and convert it into extraordinary returns. inputs used. These are the assets’ past returns, volatilities, 4. Focused risk modules that isolate and target specific risk and correlations. Second, the optimization process depends factors are an efficient way to incorporate volatility into on the same simplifying assumptions as the Sharpe ratio, momentum-based portfolios. They also facilitate the i.e., that returns are normally distributed or that one has effective use of a hurdle rate/safe alternative asset. Modules quadratic utility preferences. It is because these assumptions provide flexibility, making it simple and easy to implement are unrealistic and/or the inputs are unpredictable, that there momentum-based portfolios. Otherwise, portfolio have been many attempted fixes to the Markowitz approach. construction could be problematic given the strong non- These include shrinkage of the estimated inputs, constraining normality of momentum income streams. the portfolio weights, estimating expected returns from an 5. Despite an abundance of momentum research, no one is asset-pricing model, bootstrapping outputs to correct for bias, sure why it works so well. The most common explanations and imposing shifts toward lower variance portfolios with have to do with behavioral factors, such as anchoring and less uncertainty. Yet the math can still go wrong and create the disposition effect. An alternative explanation is that allocation mistakes because of input instability. investor risk aversion is wealth dependent. Investors are Expected returns are the least predictable of the inputs. more risk averse under adverse conditions and less risk Yet momentum makes returns more consistent and predictable. averse under favorable conditions. This causes prices to go It may be tempting then to use Markowitz optimization for to extremes beyond their reasonable values. Volatility makes momentum portfolio construction. However, we need to bad conditions seem worse and good conditions seem better, keep in mind the non-normality of our momentum return which leads to overextension of price trends and higher distributions.5 momentum profits. Fortunately, our momentum modules can guide us to an attractive alternative to Markowitz mean variance optimization. Diversification is the closest thing to a free lunch in the Modules reduce the number of portfolio inputs from eight (two investment world. This is because investors using intelligent assets per module) to four. One can analyze possible portfolio diversification can earn the same returns with less risk than allocations using nothing more than a simple spreadsheet. those holding undiversified portfolios. Momentum investing, One can search for a high Sharpe ratio, a targeted level of which is still in its infancy, may offer even better opportunities volatility, or other objective functions. There is no need for for higher returns with less risk, if done intelligently. Just matrix inversions, Lagrange multipliers, or other complicated as the benefit of diversification diminishes when applied procedures associated with Markowitz optimization. indiscriminately, the value of long side momentum also diminishes if applied too broadly, or without trying to 12. Conclusions differentiate downside from upside market conditions. We have seen how risk factors indicating high volatility When applied effectively, momentum makes diversification contribute to momentum profitability. We also introduced more efficient by selectively utilizing assets only when their the hurdle rate/alternative asset concept to help ensure that momentum is strong, and they are therefore more likely to momentum is positive on an absolute, as well as a relative, basis. appreciate. A focused momentum approach bears market risk Our final contribution is the introduction of risk factor oriented only when it makes the most sense, i.e., when there is positive momentum modules that facilitate portfolio diversification and absolute as well as relative momentum. Momentum, serving as enable the construction of effective momentum portfolios for an alpha overlay with proven success factors, can capture the harvesting risk premium profits. high premia from volatile assets while defensively adapting to Using thirty-eight years of past performance data, regime change. momentum modules show significant performance improvements in all four areas we have examined—equities, References credit risk, real estate, and economic stress, as represented Asness, Clifford S., Burt Porter, and Ross Stevens, 2000, “Predicting Stock Returns Using Industry Relative Firm Characteristics,” working by gold and Treasuries. The Fama-French three-factor annual paper, AQR Capital Management. alphas of these four modules are 8.9, 4.2, 8.7, and 10.64 Asness, Clifford S., Tobias J. Moskowitz, and Lasse J. Pedersen, 2009, respectively. The ancillary conclusions we reach are as follows: “Value and Momentum Everywhere,” working paper, AFA 2010 Atlanta Meetings. 1. Investors should consider momentum investing based on Bandarchuk, Pavel and Jena Hilscher, 2011, “Sources of Momentum diversified risk factors rather than solely by asset class. Profits: Evidence on the Irrelevance of Characteristics,” working paper. 2. Long side momentum works best when used with a hurdle Baur, Dick and T.K. McDermott, 2010, “Is Gold a Safe Haven? International Evidence,” Journal of Banking and Finance 34, 1886- 1898. rate and safe alternative asset, such as Treasury bills, that Beracha, Eli and Hilla Skiba, 2011, “Momentum in Residential Real can neutralize market risk. This puts momentum on an Estate,” Journal of Real Estate Finance and Economics 43, 299-320. absolute, as well as a relative, basis. Momentum can and Bhojraj, Sanjeev and Bhaskaran Swaminathan, 2006, “Macromomentum: Returns Predictability in International Equity Indices,” should be used tactically, as well as a strategically, in order to Journal of Business 79, 429–451. take advantage of regime persistence. Chabot, Benjamin R., Eric Ghysels, and Ravi Jagannathan, 2009, “Price 3. Investors generally wish to avoid high volatility. There is Momentum in Stocks: Insights from Victorian Age Data,” working paper, National Bureau of Economic Research. now, in fact, a propensity toward low volatility investment Chan, Kalak, Allaudeen Hameed and Wilson H.S. Tong, 2000, portfolios. Yet momentum profits are greater when using “Profitability of Momentum Strategies in International Equity Markets,” high volatility assets. Momentum can help investors harness Journal of Financial and Quantitative Analysis 35, 153-175.

PAGE 52 IFTA.ORG Chen, Hsiu Lang and Werner DeBondt, 2004, “Style Momentum within the S&P 500 Index,” Journal of Empirical Finance 11, 483-507. Dalio, Ray, 2011, “Engineering Targeted Returns and Risks,” Bridgewater Associates. Fama, Eugene F. and Kenneth R. French, 2008, “Dissecting Anomalies,” Journal of Finance 63, 1653-1678. Fama, Eugene F. and Kenneth R. French, 2011, “Size, Value, and Certifi ed Financial Momentum in International Stock Returns,” working paper. Griffin, John, Xiuquing Ji, and J. Spencer Martin, 2005, “Global Momentum Technician Strategies: A Portfolio Perspective,” Journal of Portfolio Management 31, 23-39. Grundy, Bruce D and J Spencer Martin, 2001, “Understanding the Nature of the Risks and the Sources of the Rewards to Momentum Investing,” (CFTe) Program Review of Financial Studies 14, 29-78. Jegadeesh, Narasimhan and Sheridan Titman, 1993, “Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency,” IFTA Certifi ed Financial Technician (CFTe) consists of Journal of Finance 48, 65-91. the CFTe I and CFTe II examinations. Jostova, Gergana, Stanislova Nikolova, Alexander Philipov, and Christof W Stahel, 2010, “Momentum in Corporate Bond Returns,” working paper. Successful completion of both examinations culminates Lewellen, Jonathen, 2002, “Momentum and Autocorrelation in Stock in the award of the CFTe, an internationally recognised Returns,” Review of Financial Studies 15, 533-563. professional qualifi cation in technical analysis. Menkoff, Lukas, Lucio Sarno, Maik Schmeling and Andreas Schrimpf, 2011, “Currency Momentum Strategies,” working paper. Miffre, Joelle and Georgios Rallis, 2007, “Momentum Strategies in Commodity Examinations Futures Markets,” Journal of Banking and Finance 31, 1863-1886. The CFTe I exam is multiple-choice, covering a wide Moskowitz, Tobias J. and Mark Grinblatt, 1999, “Do Industries Explain range of technical knowledge and understanding Momentum?” Journal of Finance 54, 1249–1290. of the principals of technical analysis; it is offered in Page, Sebastian, 2010, “The Myth of Diversification: Risk Factors vs English, French, German, Italian, Spanish and Arabic; Asset Classes,” Insights, PIMCO Publications. it’s available, year-round, at testing centers throughout Pirrong, Craig, 2005, “Momentum in Futures Markets,” working paper. the world, from IFTA’s computer-based testing provider, Rouwenhorst, K. Geert, 1998, “International Momentum Strategies,” Pearson VUE. Journal of Finance 53, 267-284. Rouwenhorst, K. Geet, 1999, “Local Return Factors and Turnover in The CFTe II exam incorporates a number of questions Emerging Stock Markets,” Journal of Finance 54, 1439-1464. that require essay-based, analysis responses. The candidate needs to demonstrate a depth of knowledge Note from the Editor: and experience in applying various methods of The author Gary Antonacci is the winner of the NAAIM technical analysis. The candidate is provided with Wagner Award 2012. This paper was originally submitted for current charts covering one specifi c market (often an equity) to be analysed, as though for a Fund Manager. this contest. IFTA is thankful to Greg Morris and the National Association of Active Investment Managers for the permission The CFTe II is also offered in English, French, German, Italian, Spanish and Arabic, typically in April and to print this document. For further information please refer also October of each year. to http://www.naaim.org/resources/wagner-award/ References Curriculum 1. http://optimalmomentum.com The CFTe II program is designed for self-study, 2. Since these indices are based on capitalization, the MSCI ACWI exUS however, IFTA will also be happy to assist in fi nding receives only a modest influence from emerging markets. Our results qualifi ed trainers. Local societies may offer preparatory do not change significantly if we use only the MSCI EAFE Index. courses to assist potential candidates. Syllabuses, 3. Skewness relates directly to the symmetrical characteristics of Study Guides and registration are all available on the the return distribution. Positive skewness implies the potential for IFTA website at http://www.ifta.org/certifi cations/ greater variance of positive returns than negative returns. Risk registration/. averse investors generally prefer positive skewness over negative skewness. 4. The four disclosed momentum products available to the public use twelve-month momentum. They are AQR Funds, Russell Investments, To Register QuantShares, and Summerhaven Index Management. Please visit our website at http://www.ifta.org/ 5. The Jarque-Bera, Shapiro-Wilk, Lilliefors and Anderson-Darling tests certifi cations/registration/ for registration details. all have p values <.0001 for each of our modules, which strongly rejects normality. Cost IFTA Member Colleagues Non-Members CFTe I $500 US CFTe I $700 US CFTe II $800* US CFTe II $1,000* US *Additional Fees (CFTe II only): $250 US translation fee applies to non-English exams $100 US applies for non-IFTA proctored exam locations

IFTA.ORG PAGE 53 IFTA JOURNAL 2013 EDITION

Heikin-Ashi: A Better Technique to Trends in Noisy Markets By Dan Valcu, CFTe

Abstract series, opens always at the midpoint of the previous heikin- ashi body. This is the core of the noise filtering mechanism it One thousand traders, one thousand profiles. provides. Figure 2 shows the S&P 500 index both as a traditional candlestick representation and heikin-ashi candles. But one single element works like glue for all market Heikin-ashi makes trends immediately visible while much of participants and pushes prices up and down: The continuous the price noise is filtered out. quest to identify trends and position on the correct side of them. A big advantage of this technique is the low-cost entry for What “correct” means is a different story for everyone and traders. Even without chart reading rules, anyone can identify depends on the individual profile (time frame, capital and risk trends better by assuming that uptrends are defined by a management, perception of the external stimuli). sequence of white heikin-ashi candles while the downtrends This article introduces and discusses a lesser-known are associated with filled modified candles. The assumption Japanese trend technique, heikin-ashi, that from its beginnings is correct and is incorporated into the five rules that will be was used to filter out price noise in any time frame. Like many of discussed and exemplified later in this article. the Japanese charting techniques, heikin-ashi is a visual add-on A traditional candle price chart is a bi-color canvas with a to any trader’s existent decision tools. Since almost everything variety of candle sizes and patterns. The corresponding heikin- can be quantified, the technical indicators derived from the ashi chart is a better visual assessment of the price action and original heiki-ashi candle formulas are a step forward towards a uses only three candle types/patterns as shown in Figure 3. more complete and less risky trend technique. In the end, heikin-ashi becomes a tool that appeals to Figure 3: The building blocks of any heikin-ashi chart are white, filled, and -like candles. both sides of the trader’s brain: Right (heikin-ashi original visual charts) and Left (analytical, new heikin-ashi technical indicators). If the sequence Identification – Analysis – Decision – Execution is high on trader’s priority list then this simple trend technique is an excellent choice to reduce its duration and increase confidence in the decisions taken. Introduction Heikin-ashi is a simple price noise filtering technique based White Filled Doji-like on modified open, high, low, and close values. The formulas described in Figure 1 describe how to generate modified (heikin- The special element is the doji-like candle that consists ashi) prices and are the foundation of this technique. of a small body with both upper and lower wicks. Its main function is to alert about price reversals. The appearance of a Figure 1: These four formulas are used to define heikin-ashi open, first such candle on a chart after a trend leads to preparation high, low, and close values. for a reversal. Since life and trading are far from being binary worlds, several such short-body candles are typical for a price consolidation instead of reversal. Even in this case, risk management is required to protect capital. Heikin-Ashi Chart Reading Rules Reading a traditional Japanese candlestick chart is, in many cases, an art that depends on emotions and fluid translation rules. Currently, there are over one hundred documented Japanese candle patterns, each with a more or less subjective definition and translation. Trading with subjectivity and emotions is not what traders prefer to do. Stricter rules attached The most important value of the set is haOpen. Any heikin- to any technique are a must and heikin-ashi helps traders going ashi candle with the exemption of the first one of the data into this direction.

PAGE 54 IFTA.ORG IFTA JOURNAL 2013 EDITION

Figure 2: The monthly S&P 500 index is displayed as heikin-ashi (modified) candles in the top pane and Japanese candlesticks in the lower pane.

The following five heikin-ashi rules are available to read and the beginning of May (R2 with R4, R3, R5). It was a stronger translate modified candle charts. downtrend and this is the reason why Rule 2 (R2) was attached to this segment. The short consolidation between the two Rule 1 (Trend) intermediary downtrends is translated with help of Rule 3 (R3) A sequence of white bodies with no lower shadows identifies which can also apply to the first two heikin-ashi candles in the an uptrend. A sequence of filled bodies with no upper shadow beginning of May. Rule 4 (R4) describes a trend slowdown when identifies a downtrend. the size of a body candle gets smaller. Finally, Rule 5 (R5) points to a doji-like reversal candle like the last one of the downtrend Rule 2 (Strong trend) or the first one in May. Taller the bodies, stronger the trend. Quantification of Heikin-Ashi Candles Rule 3 (Trend slowdown) The original Japanese technique is visual and consists only The trend gets weaker with the occurrence of smaller bodies of heikin-ashi candles. One additional and logical step to pursue and, possibly, with the emergence of both upper and lower was to make the technique more suitable to the analytical shadows. A body inside the previous one is a sig of a possible Western way of thinking, i.e., to quantify heikin-ashi candles trend slowdown. and generate one or more technical indicators. As a result, a new and very simple indicator, haDelta, was Rule 4 (Consolidation) born. It is simply the difference between haClose and haOpen. A series of smaller bodies with both upper and lower shadows Figure 5 shows the S&P-500 index in a daily time frame and (wicks). displayed with haDelta (black line) in the middle pane. We should remember at this point that heikin-ashi is Rule 5 (Trend reversal) not a mechanical technique but a component used with a A trend reversal is likely with the emergence of a small body discretionary trading system. with long upper and lower shadows or when sudden color The big advantage of using haDelta is its ability to generate change occurs. advance signals. Since haDelta is rough in many cases, we add its simple 3-bar average, SMA(haDelta,3) to achieve a smoother We will see now how the five rules work on the chart indicator and also to generate crossovers. displayed in Figure 4. The basic signal using this pair is the crossing of haDelta The chart was delimited into zones showing trends (Rules above its moving average (Long) or below it (Sell). Since haDelta 1 and 2). For each zone, we also indicated within parenthesis is in many cases noisy, traders can choose only the moving other rules that apply. It is worth noting that seldom a zone on a average to add or reduce positions when the average turns heikin-ashi chart is defined by a single rule. Since trends (R1 and positive or negative. In cases when the average hovers around R2) morph into consolidations (R3 and R4) or reverse (R5), any zero (end of February, end of March—beginning of April), the heikin-ashi chart is a blend of these five flavors. index signals a consolidation period. For simplicity, we look at the downtrend developed since

IFTA.ORG PAGE 55 IFTA JOURNAL 2013 EDITION

Heikin-Ashi vs. Japanese Candlestick recommendations in trading became available. Unfortunately, Patterns the purity of the original knowledge has been lost in subsequent translations, leading to many erroneous interpretations. The introduction of the Japanese candlesticks was a big A few formations are very clear such as the bullish and qualitative step towards a better understanding of the trading bearish engulfing patterns, evening and morning stars, and psychology. As a result, an extensive literature on Japanese have a higher probability for success. Many of the patterns are candlestick patterns, their definitions, interpretation, and either rare appearances on charts or require a translation in the

Figure 4: The German DAX index is also displayed as a heikin-ashi daily chart in the top pane. The five heikin-ashi rules (R1 through R5) help to identify at a glance trends, possible reversals, and consolidations.

Figure 5: The daily heikin-ashi S&P-500 chart is enhanced with the addition of haDelta and its short moving average SMA(haDelta,3) (middle pane).

PAGE 56 IFTA.ORG IFTA JOURNAL 2013 EDITION context they are generated. Subjectivity is a severe constraint between the two has its merits and is worth pursuing. despite many attempts to quantify and include them into Heikin-ashi is based on precise definitions and discretionary/mechanical trading systems. measurements so it is logical to try to use this technique either On one side we have the current Japanese candlestick theory. to confirm candlestick patterns or to remove them from trading. On the other side, heikin-ashi, a simple, clean, price filtering Figure 6 shows the S&P-500 index on a weekly chart. technique that focuses on trends and reversals. Since many For our discussion we look in particular at the patterns candle patterns are used for reversals, the idea of a cohabitation marked as P1, P2, and P3 in Figure 6. While P2 and P3 look easier

Figure 6: The heikin-ashi technique can be used both as a confirmation for traditional candle patterns or replacement for many. The trader has both options available.

Figure 7: Many of the popular candle patterns can be easily translated using haDelta and its short 3-bar average.

IFTA.ORG PAGE 57 IFTA JOURNAL 2013 EDITION to identify them as bullish engulfing patterns, P1 requires more As far as P2 is concerned, haDelta offers a bullish crossover flexibility and indulgence to match it with a known pattern. at the end of the two-bar pattern. This is an example of how It has some elements of an evening star but the uptrend is heikin-ashi can work in cohabitation with Japanese candle missing. Heikin-ashi does not pay attention to definitions, patterns. Finally, P3 (bullish engulfing) has been accompanied names, and interpretations of candle pattern. Once we see a by an haDelta bullish signal issued one bar earlier. potential candle pattern, we can use haDelta and its average as Several Japanese candlestick patterns and their heikin-ashi translators. translations are seen in Figure 7 in a daily time frame. In the case of P1, haDelta went below its average (bearish P1 and P2 are bullish harami patterns. Their heikin-ashi signal) on the second day of the three-bar formation, earlier that translations are done using haDelta and its short average the price rolled over. and their bullish characters coincide with haDelta positive

Figure 8: As with any visual technique and indicator, heikin-ashi in its dual format, improves the trend assessment when used in multiple time frames.

Figure 9: Double smoothing haDelta improves the timing and confidence especially when heikin-ashi technique is used in multiple time frames.

PAGE 58 IFTA.ORG IFTA JOURNAL 2013 EDITION crossovers at the end of the patterns. can be offset using the quantification of heikin-ashi candles: P3 looks like a bullish engulfing pattern but haDelta ignores haDelta and related moving averages. haDelta not only reduces this doubt sending a bullish signal after the first bar of the the lag between price and heikin-ashi chart reversals but it also uncertain pattern. Finally, P4 is a 3-bar pattern resembling, generates, in many cases, advance signals, worth investigating. but far from, a real morning doji star. This is not a problem for Heikin-ashi can be used in any time frame with an improved heikin-ashi that uses haDelta and its short average to confirm accuracy and confidence when traders use it in multiple time candles patterns. The character of 3-bar pattern turns bullish frames. after the third bar, when haDelta crosses above its average. Purists can use only heikin-ashi charts with haDelta. Other P3 and P4 point to another interesting feature of haDelta. traders may add technical indicators and other techniques to When the indicator reaches historical high or low values it confirm trading signals. Since no combination is perfect, risk is time to think about price reversals. In addition to these and capital management remains a must-have to ensure that situations, on the heikin-ashi charts we notice a body-inside- failures translate into small losses. body formation (Rule 3) which warns about a possible trend A very important aspect of using heikin-ashi is to confirm slowdown and reversal. Japanese candlestick patterns or even remove them from Heikin-ashi and especially its quantifiable indicator haDelta trading. Both options are available to each trader. prove very helpful to translate Japanese candle patterns and Heikin-ashi can be used with any financial instrument in any improve the confidence. time frame. Best results are achieved with instruments which, historically, display clear trends and display similar trends on Heikin-Ashi in Multiple Time Frames heikin-ashi charts in two or even three time frames. Heikin-ashi is a versatile technique and one can experiment with it in the context of existent strategies and techniques. The Bibliography use of modified candles and their quantification in multiple time NISON, S. Beyond candlesticks: New Japanses charting techniques revealed. Wiley Finance, 1994 frames is one way to achieve improved accuracy in terms of VALCU, D. Heikin-Ashi: How to Trade Without Japanese Candlestick trend assessment and, implicitly, better entry and exit timing. Patterns. Educofin, 2011. Figure 8 shows the S&P-500 index in three time frames: VALCU, D. Using the Heikin-Ashi Technique. Technical Analysis of Stocks & Commodities, February 2004. Daily, weekly, and monthly. It is well-known that higher perspectives remove more noise from the price actions. If we add heikin-ashi indicators, the timing of reversals look better. Software and Data We see that the short- and medium-term heikin-ashi charts Software used for this article was AmiBroker from AmiBroker.com, Lodz, Poland. www.amibroker.com look bullish because the current candles are white. The long- End-of-day data provided by Worden Brothers, Inc. (TC2007 software) term assessment is still bearish. Worden Brothers, Inc. Five Oaks Office Park, 4905 Pine Cone Drive, If we look at the more accurate haDelta charts (lower panes) Durham, NC 27707, USA. we see the same trend assessment. The only difference between the visual and quantifiable charts is the timing of the reversals, About the Author with haDelta offering earlier signals. It does not happen every Dan Valcu, CFTe is the author of the first book on this subject, time but when it does, the alert is worth taking. Heikin-Ashi: How to Trade Without Candlestick Patterns published Another way to use heikin-ashi in multiple time frames is in September 2011 (www.educofin.com). He is General Manager to enhance haDelta by double smoothing it. Figure 9 has in the of Educofin Ltd and serves on the Board of the International lower panes the pair {SMA(haDelta,3), SMA(SMA(haDelta,3),)}. Federation of Technical Analysis (IFTA). The loss of accuracy for reversals is not significant but the trend assessment is better. Conclusions Heikin-ashi is a charting technique designed to filter out price noise and a low-entry barrier for traders to assess trends, consolidations, and reversals. Although heikin-ashi charts look very attractive at first sight, traders should refrain from using this technique as a mechanical trading tool. Its best use and results derive from using it as a component of discretionary trading systems. Heiki-ashi’s foundation is built on modified open, high, low, and close values, three candle types and five rules to assess trends, reversals, and consolidations. Once these basic concepts are well understood, the advantages of using heikin-ashi become very clear. The visual aspect has a drawback due to the modified candles construction: A very short delay, usually one bar, as far as reversals on a heikin-ashi charts concern. The impact of this lag

IFTA.ORG PAGE 59 IFTA JOURNAL 2013 EDITION

Mastering Market Timing, Using the Works of L.M. Lowry and R.D. Wyckoff to Identify Key Market Turning Points By Richard A. Dickson and Tracy L. Knudsen Reviewed by Regina Meani, CFTe

During my time as editor of the Journal I had the pleasure and from strong to weak hands at tops. of including articles by Professor Hank Pruden on the Wyckoff Some useful indicators are brought into the mix including the Method and an article by Paul Desmond, the current president of Advance / Decline Line with Lowry’s tweak on this and the use of the Lowry Research Corporation. So it the 30-week moving average. seemed a natural follow on that we include A remarkable pointer to the Authors’ a review of a book which uses the works enthusiasm and confidence for their subject of both Wyckoff and Lowry; coming to is that they present a final chapter: Where my attention as the runner-up in The are we now? Technical Analyst: Best Book of the Year The authors’ macro approach should not 2012 Awards in the UK. put off the short term trader but rather they The authors’ work is exemplary in should realise that they need to be aware of taking us through the historical record the lessons learned from the examination of the classic major tops and bottoms of of major market tops and bottoms that is the 20th Century and into the 21st Century, meticulously provided, and is necessary using the combined methodology of for both the trader and investor alike. I Wyckoff’s laws of Supply and Demand, am sure that it is part of the mantra of Cause and Effect and Effort and Result most traders that the short-term outlook with the Lowry’s indicators for the has its beginnings in the longer term and determination of supply and demand; can be a major influence over the entity’s the Buying Power and Selling Pressure. movements both in the near and longer Through the Wyckoff focus on price term. This in-depth investigation is a and volume patterns and the Lowry practical start for beginners and a useful interpretations for market breadth, we reminder for the more experienced. are guided through the exploration of In the words of the authors’: the forces of supply and demand, with Although many years have passed since a primary reliance on price and volume. Wyckoff and Lowry developed their tools… We are given an explanation of how these Despite the changing world…human forces effect the development and stages emotions remain the same throughout the of bull and bear trends and the formation of the reversal of various stages of bull and bear markets. It is the consistency of these trends. human nature that causes major tops and bottoms to show little While each experience is unique the authors’ explanations of the change in their basic characteristics…1 way the market works in periods of distribution and accumulation provides us with a means of identifying when the supply of the 1. R A Dickson and T L Knudsen, Mastering Market Timing, Using the works of L.M. Lowry and R.D Wyckoff to Identify Key Market Turning stock or entity is moving from weak to strong hands at bottoms, Points, FT Press, New Jersey, 2012, p.194

PAGE 60 IFTA.ORG IFTA JOURNAL 2013 EDITION

Author Profiles

Gary Antonacci Mohamed El Saiid, CFTe, MFTA Mr. Antoncacci has over 30 years Mohamed is currently an Executive experience as an investment Director and Head of the Technical professional focusing on under exploited Analysis department for HC Brokerage investment opportunities. He received (HCB), Cairo, Egypt. He started his career his MBA degree from the Harvard working for Momentum Wavers, Ltd., a Business School and managed a stock Middle East Technical Analysis firm options hedge fund during the 1970s. In (2001-2004). He joined HCB as an the 1980s, Mr. Antonacci became a highly associate/lead technical analyst successful commodity pool operator. (2004-2006). Later he joined Unifund, a During that time, he pioneered the use of modern portfolio Geneva-based global private fund (2006-2007) as a Chief theory principles and optimization practices in order to allocate Technical Strategist/Co-Fund Manager to the Middle East private and institutional investor funds to some of the world’s investments. Mohamed holds an MBA in Finance and is best traders. currently a Board Member, Technical Analysis instructor and During the past twenty years, Mr. Antonacci has Head of the R&D committee in the Egyptian Society for concentrated on researching, developing, and applying Technical Analysts (ESTA). asset allocation strategies that have their basis in academic research, such as price momentum. He is author of a number Kay Ying Timothy Fong, CFTe, MFTA of investment articles and books on portfolio management, Timothy is Director of Analytics at including the forthcoming book Optimal Momentum Investing. Canada’s federal banking regulator—the Mr. Antonacci serves as a consultant to private and institutional Office of Superintendent of Financial investors regarding asset allocation, portfolio optimization, and Institutions Canada. He has over 10 advanced momentum strategies. He is the 2011 second place years of experience in financial modeling and 2012 first place winner of the NAAIM Wagner Awards for and quantitative analysis for the Advancements in Active Investment Management. banking industry. He has been an active member of various groups for trading Stephan Belser, MSc, CFTe, MFTA book policy and implementation at the Stephan is Head of Portfolio Basel Committee of Banking Supervision. Management at Vermögen-Management Outside of the office, Tim enjoys designing and teaching BC GmbH, a private banking company in courses that develop future leaders in financial trading Germany. There, he works with high net and risk management. In 2008, he received the prestigious worth clients, in all areas of asset and Excellence in Teaching Award from the School of Continuing portfolio management. He is responsible Studies at the University of Toronto. He holds a Masters in for fundamental, technical and Mathematical Finance from the University of Toronto and a sentiment research for global markets number of other professional designations including FRM, and is author of their monthly market PRM, DMS, CAIA and CIM. commentary. He is Chairman of the Weekly Asset Allocation meeting and fund manager for individual and institutional clients. Bryan Lim Stephan is a lecturer for private banking and portfolio Bryan is currently a Masters student at management at the DHBW Villingen-Schwenningen. the University of Cambridge. He has a He holds an MSc in Banking and Financial Management keen interest in technical analysis and from the University of Liechtenstein and he received the the application of concepts from the Banking Award Liechtenstein for his master thesis in 2006. He physical sciences to the modelling of holds the CFTe and MFTA qualifications in technical analysis. financial data. He is currently working Stephan is a member of the Vereinigung Technischer Analysten on a research project involving the use of Deutschlands (VTAD). Bayesian particle filtering for the prediction of financial time series.

IFTA.ORG PAGE 61 IFTA JOURNAL 2013 EDITION

Shawn Lim, CFTe, MSTA Dan Valcu, CFTe Shawn is currently a student reading Dan is an independent trader and Economics at University College London. founder of the first company specialized He has a keen interest in the study of in technical analysis education and financial markets and holds a number of training in Romania. He also authored relevant professional qualifications. He four books about technical analysis and is a Certified Financial Technician and a strategies. He is and has been a Member of the Society of Technical contributor to various technical analysis Analysts (UK). In addition, he is a magazines (TASC, Traders’ Magazine) and certified Professional Risk Manager is credited with bringing Heikin-ashi (PRM) and he holds an Advanced Diploma in Data Systems charting to the western world in 2003–2004. His latest book and Analysis from the University of Oxford. Heikin-Ashi: How to Trade without Japanese Candlestick Patterns, a world premiere, is written for everybody who needs simple Regina Meani, CFTe techniques to highlight the trend, reduce the noise, and alert Regina covered world markets, as about possible reversals. In addition, this book is a guide for technical analyst and Associate Director easily translating candlestick patterns. Before joining the for Deutsche Bank, before freelancing. technical analysis field, Dan worked all over the world as an IT She is an author and has presented Consultant in banking & insurance. internationally and locally and lectured An active promoter of technical analysis, Dan serves on the for the Financial Services Institute of Board of the International Federation of Technical Analysis Australasia (FINSIA), Sydney University (IFTA) as VP for Europe and Director of Membership, holds the and the Australian Stock Exchange. She professional designation of a Certified Financial Technician is VP of the Australian Professional (CFTe), and is an Associate Member of the Society of Technical Technical Analysts (APTA) and immediate past Journal Director Analysts (UK). Dan holds a Master’s degree in Computer for IFTA. Regina carries the CFTe designation. She has regular Sciences from the Polytechnic Institute in Bucharest. He is columns in the financial press and appears in other media also the President and one of the founders of the Romanian forums. Her freelance work includes market analysis, webinars Association for Technical Analysis (AATROM). and larger seminars, advising and training investors and traders in Market Psychology, CFD and share trading and technical analysis. Regina is also a past director of the Australian Technical Analysts Association (ATAA) and has belonged to the IFTA Board of Directors Society of Technical Analysts, UK (STA) for over thirty years.

Ed Rowson, CFTe, MFTA President Directors at Large Ed is a Partner and the Trader at MENA Adam Sorab, CFTe, FSTA (STA) David Furcajg, CFTe, MFTA (AFATE) Capital, as well as being the Risk Vice-President—the Americas Akira Homma, CFA, CIIA, CFTe, FRM, CMA, Timothy Bradley (TSAASF) CMT (NTAA) Manager. Ed joined MENA Capital in April Regina Meani, CFTe (STA, ATAA) 2006. He has ten years institutional and Vice-President—Asia William Chin, MBA (CSTA) hedge fund experience focusing on Taichi Otaki (NTAA) implementing risk managed technical Vice-President—Middle East, Staff trading strategies. MENA Capital is a Africa Executive Director Mohamed Ashraf Mohfauz, CFTe, CETA Beth W. Palys, FASAE, CAE London-based investment management (ESTA) and advisory company with focus in the Vice President, Meetings Treasurer Grace L. Jan, CAE, CMP stock markets of the Middle East and North Africa. While Michael Steele (AAPTA) overseeing the risk management and trading duties of the funds, Senior Member Services Secretary Manager Saleh Nasser, CMT (ESTA) Ed investigated and tested the benefits of implementing a risk Linda Bernetich adjusted stop-loss strategy over the more widely used Education Director Senior Graphic Designer percentage drawdown stop-loss strategy while technical trading. (Academic & Syllabus), Journal Jon Benjamin His MFTA paper investigates whether there is empirical evidence Director Rolf Wetzer, Ph.D. (SAMT) Production Manager to support this as well as exploring methods to optimise the Penny Willocks return on capital and return on Risk that is achievable. Accreditation Director Roberto Vargas, CFTe (TSAASF) Accounting Dawn Rosenfeld Examination Director Gregor Bauer, Ph.D. (VTAD) Membership Director, Vice- President—Europe Dan Valcu, CFTe (AATROM) Conference Director Robert Grigg (ATAA)

PAGE 62 IFTA.ORG Master of Financial Technical Analysis (MFTA) Program

Examinations IFTA’s Master of Financial Technical Analysis In order to complete the MFTA and receive your (MFTA) represents the highest professional Diploma, you must write a research paper of no achievement in the technical analysis community, less than three thousand, and no more than fi ve worldwide. Achieving this level of certifi cation thousand, words. Charts, Figures and Tables may requires you to submit an original body of be presented in addition. research in the discipline of international technical analysis, which should be of practical Your paper must meet the following criteria: application. • It must be original • It must develop a reasoned and logical The MFTA is open to individuals who have argument and lead to a sound conclusion, attained the Certifi ed Financial Technician (CFTe) supported by the tests, studies and analysis designation or its equivalent, e.g. the Certifi ed contained in the paper ESTA Technical Analyst Program (CETA) from the • The subject matter should be of practical Egyptian Society of Technical Analysts (ESTA) application For those IFTA colleagues who do not possess • It should add to the body of knowledge in the the formal qualifi cations outlined above, but discipline of international technical analysis who have other certifi cations and/or many years experience working as a technical analyst, Timelines & Schedules the Accreditation Committee has developed There are two MFTA sessions per year, with the an “alternate path” by which candidates, following deadlines: with substantial academic or practical work in technical analysis, can bypass the requirements Session 1 for the CFTe and prequalify for the MFTA. “Alternative Path” application deadline The alternate path is open to individuals who February 28 have a certifi cation, such as: Application, outline and fees deadline May 2 • Certifi ed Market Technician (CMT) or a Society Paper submission deadline of Technical Analysts (STA) Diploma, plus three October 15 years experience as a technical analyst; or • a fi nancial certifi cation such as Certifi ed Session 2 Financial Analyst (CFA), Certifi ed Public “Alternative Path” application deadline Accountant (CPA), or Masters of Business July 31 Administration (MBA), plus fi ve years Application, outline and fees deadline experience as a technical analyst; or October 2 • a minimum of eight years experience as a Paper submission deadline technical analyst. March 15 (of the following year) A Candidate who meets the foregoing criteria may apply for the “alternate path”. If approved, they can register for the MFTA and submit their To Register research abstract. On approval, the candidate Please visit our website at http://www.ifta.org/ will be invited to submit a paper. certifi cations/master-of-fi nancial-technical- analysis-mfta-program/ for further details and to register. Cost $900 US (IFTA Member Colleagues); $1,100 US (Non-Members) PICTURE PERFECT Perfect your trade strategies with charting on Bloomberg. CHART is your visual gateway to 20 million securities, fundamentals, economic releases & more. All this, integrated into an intuitive charting platform with technical alerts, market moving events, custom studies, backtesting and impressive visualizations to boot.

Explore CHART on the BLOOMBERG For more information about Bloomberg, contact a Bloomberg Sales representative ® PROFESSIONAL service. or visit www.bloomberg.com.

New York +1 212 318 2000 London +44 20 7330 7500 Tokyo +81 3 3201 8900 São Paulo +55 11 3048 4500 Dubai +971 4 373 9400

©2011 Bloomberg Finance L.P. All rights reserved. 44756261 0711