Florida State University Libraries

Transcription

1 Florida State University Libraries Electronic Theses, Treatises and Dissertations The Graduate School 2006 The Relationship Between Technical Analysis Generated s and the Fama and French Risk Factors as Applied to Individual Securities Debra I. Peterson Follow this and additional works at the FSU Digital Library. For more information, please contact lib-ir@fsu.edu

2 THE FLORIDA STATE UNIVERSITY COLLEGE OF BUSINESS THE RELATIONSHIP BETWEEN TECHNICAL ANALYSIS GENERATED RETURNS AND THE FAMA AND FRENCH RISK FACTORS AS APPLIED TO INDIVIDUAL SECURITIES By DEBRA I. PETERSON A Dissertation submitted to the Department of Finance In partial fulfillment of the Requirements for the degree of Doctor of Philosophy Degree Awarded: Summer Semester, 2006

3 The members of the Committee approved the dissertation of Debra I. Peterson defended on May 16, Stephen E. Celec Professor Directing Dissertation Allen W. Bathke Outside Committee Member Gary A. Benesh Committee Member Approved: James M. Nelson Committee Member Caryn K. Beck-Dudley Dean, College of Business The Office of Graduate Studies has verified and approved the above named committee members. ii

4 ACKNOWLEDGEMENTS I would like to express my deep appreciation to the numerous people who provided support and encouragement in addition to their professional assistance during the writing of this dissertation. First, to my husband, David Peterson, my love and my gratitude for enabling me to complete this task; next a most heart felt thank you to my dissertation chairman, Stephen Celec, who was with me and for me from the conception to the end of this dissertation; and to all my committee members, Gary Benesh, Al Bathke, and James Nelson for the time and effort they expended on my behalf. Lastly, a special thank you to Scheri Martin, Bill Christiansen, and Melissa Houston, each of whom helped me in their own unique way. iii

5 TABLE OF CONTENTS List of Tables... vi List of Charts... vii Abstract... viii 1. INTRODUCTION LITERATURE REVIEW AND TRADING RULE GUIDELINES DATA AND METHODOLOGY Data Moving Average Rules Filters Volume Rules Individual Stock Eligibility Grand Trading Strategies The Fama and French Three-Factor Model OBSERVATIONS AND FINDINGS Descriptive Statistics Twenty-Year Regression Estimates Table 2: Layout Grand Strategy Comparisons Trading Rule Constraint Comparisons Abnormal and Market-Adjusted Comparisons Five-Year Regression Estimates Tables 3 and 4: Layouts Table 3: Observations and Findings Table 4: Observations and Findings CONCLUSIONS REFERENCES BIOGRAPHICAL SKETCH iv

6 LIST OF TABLES Table 1: Descriptive Statistics Table 2: Twenty-Year Market-Adjusted s and Regression Coefficients Table 3: Five-Year Market-Adjusted s and Regression Coefficients Table 4: Five-Year Market-Adjusted s and Regression Coefficients Table 5: Five- and Twenty-Year Summary Values for the Fama and French Factors Table 6: Regressions of Excess Stock s on the Excess Market and the Mimicking s for the Size (SMB) and Book-to-Market Equity (HML) Factors: Jul 1963 to Dec v

7 LIST OF CHARTS Chart #1: Variable Descriptions Chart #2: Trading Rules and Trading Rule Abbreviations Chart #3: Table 1 Excerpt: Panels A2 and B Chart #4: Table 1 Excerpt: Panels C1 and F Chart #5: Market-Adjusted and Abnormal nificance for Grand Strategy Comparisons Chart #6: Market-Adjusted and Abnormal s nificance for Trading Rule Constraint Comparisons Chart #7: The Magnitudes of Abnormal and Market-Adjusted s by Beta Size Chart #8: Beta Magnitudes Chart #9: Market-Adjusted and Abnormal nificance from Select Trading Rules for Temporal Consistency Chart #10: Market-Adjusted and Abnormal nificance from Select Trading Rules for Temporal Consistency vi

8 ABSTRACT The purpose of this dissertation is to determine whether potential trading rule profits are unique to specific strategies or whether they are associated with factors already known to impact stock returns. In the course of examining this question, I explore: (1) whether successful technical trading strategies based on the financial literature can select individual firms that yield market-adjusted returns that differ from zero; (2) whether technical trading rules can yield abnormal returns after controlling for the three Fama and French factors of market return, size, and book-to-market equity; and (3) whether abnormal returns produced by technical trading rules are temporally consistent. I find that technical trading strategies can be devised that yield significant abnormal returns over the twenty-year period from 1984 through 2003 that are also temporally consistent over five-year time durations. In addition, these rules may also generate significant market-adjusted returns. These findings suggest that technical trading rule returns are not fully explainable by the Fama and French (1993) three-factor model. vii

9 CHAPTER 1 INTRODUCTION Although academicians generally hold technical analysis in low esteem 1, the financial literature is replete with evidence of market anomalies casting doubt on whether the marketplace follows a random-walk, i.e., whether price changes over time are random and thus, not predictable. Some researchers have found evidence of predictability in stock returns while exploring other market phenomena such as momentum, reversals and market response to new information. 2 These findings have typically been rebutted by allegations of data-snooping, failure to risk-adjust returns, or failing to consider transaction costs or taxes. The debate over market efficiency persists and continues to be of interest to both academic researchers and investors. According to Fama (1991), evidence supporting return predictability over time is a major source of controversy in the market efficiency debate. 3 Technical analysts assume that stock returns are predictable. Even if granted this assumption, questions remain as to whether predictability can be used to generate abnormal economic profits and how this can be accomplished. If the stock market is predictable, it is logical to seek out patterns in historical prices and trading volumes since only current and past information is available with which to forecast. If an investor has superior skill in forecasting returns, he is likely to use this advantage in attempting to generate abnormal returns, i.e., returns greater than those justified by the risk undertaken. Technical analysis is an umbrella term that describes a myriad of strategies that endeavor to detect stock price patterns and trends. These strategies consist largely of applying mechanical trading rules to identify individual stocks, and market entry and exit points, with the goal of market inefficiency, trading rules occasionally appear in the financial literature. These rules, 1 See Brock, Lakonishok, and LeBaron (1992), p and Radcliffe, 1997, p For examples, see DeBondt and Thaler (1985,1987), Jegaheesh (1990), Lehmann (1990), Lo and MacKinlay (1990), Jegadeesh and Titman (1993) and Chan, Jegadeesh and Lakonishok (1996), and Kothari and Warner (1997). 3 See Fama (1991), p

10 however, are typically designed to investigate market efficiency hypotheses rather than to test the efficacy of the trading rules, themselves. These academic studies find contradictory evidence of trading rule value, a not surprising result given that the question of market efficiency remains open. What cannot be disputed, however, is that technical analysis is actively employed today by stock market advisors, brokerage firms, currency dealers and futures traders. For example, Taylor (1992) surveys the chief foreign exchange dealers in the London market and finds that technical analysis is widely used. Theoretical models posited in the literature allow for the possibility that technical analysis may provide investors with value. Treynor and Ferguson (1985) and Brown and Jennings (1989) propose theoretical models in which past prices contribute to profits by aiding in the assessment of information quality and in market entry decisions. They find that technical analysis, as applied in their models, can be used to generate profits. The literature also provides indirect empirical support for technical analysis. Evidence of market anomalies, predictability of returns, momentum and reversals, varying time-horizon impacts and investor under/over reaction all suggest the existence of patterns in historical information capable of exploitation by investors who are able to identify such patterns. 4 The beginnings of technical analysis date back to Charles Dow in the late 1800 s. The Dow Theory was popularized by Peter Hamilton who claimed to follow its tenets in his Wall Street Journal editorials that predicted U.S. bull and bear stock market trends. The Dow Theory is primarily a market timing strategy based on momentum. It is relevant to technical analysis in 4 Lo and MacKinley (1988) and Conrad and Kaul (1989) find that weekly returns are positively autocorrelated, particularly for the smallest capitalization stock portfolios. Kiem (1983) documents a pronounced January effect in which nearly half of the higher annual excess returns achieved by small stock portfolios over large stock portfolios are derived during the first five days of January. French (1980) and Gibbons and Hess (1981) find returns are greater in the first half of any month than in the second half. Harris (1986) finds that stock prices tend to increase substantially during the last fifteen minutes of trading regardless of the day of the week. Cutler, Poterba, and Summers (1991) findings generally support the position that returns are positively correlated over short durations and negatively correlated over long horizons. Loughran and Ritter (1997) document the long-term predictability of returns after seasoned equity offerings. Daniel, Hirshleifer, and Subrahmanyam (1998) show that overconfidence and excessively attributing success to ability while excessively discounting low ability can generate short horizon momentum followed by reversals at longer horizons. Hong and Stein (1999) also generate momentum profits followed by reversals when considering the interaction between two types of rational agents, one type that conditions on information about the future but ignores current and past prices, and another type that forms overly simple forecasts based solely on lagged information. Kyle and Wang (1997), Odean (1998), and Gervais and Odean (2001) consider models in which agents are overconfident about the quality of the information they trade on while DeLong, Schleifer, Summers, and Waldmann (1990a), (1990b) consider a model in which some investors make forecast errors that are correlated with their information sets. 2

11 that it forecasts future stock prices based on historical prices. Alfred Cowles (1934) compares the returns obtained from Hamilton s market timing strategy to the benchmark of a fully invested stock portfolio in the Dow Jones railroad and industrial averages. Cowles construes his results as demonstrating that fixed-length sell strategies yield higher returns than Hamilton s timing strategy and thus concludes that the Dow Theory does not work. Brown, Goetzmann, and Kumar (1998) revisit Cowles findings. While they replicate his empirical results, they draw contrary conclusions based on 64 years of updated mathematical techniques. They find that the Dow Theory, as implemented by Hamilton, resulted in positive excess returns. Examples of technical analysis viability also appear in recent academic journals. For example, Lo, Mamaysky, and Wang (2000) (LMW, hereafter) develop algorithms permitting the objective identification of patterns that technical analysts profess to visually recognize in charts of historic stock prices. They find that certain technical patterns, when applied to many stocks over many time periods, do provide incremental information. Referring to this finding, they say this raises the possibility that technical analysis can add value to the investment process. In a discussion paper by Jegadeesh (2000), he notes that LMW subjectively define the technical patterns they attempt to identify. Furthermore, the smoothing technique applied to price data is also subjective. Jegadeesh points out that LMW do not examine the profitability of their technical trading rules which is the generally accepted method for evaluating a rule s usefulness. The question of whether trading rules can produce abnormal returns, therefore, remains open. The trading strategies employed by technical analysis have a long history of refinement. Alexander (1961) introduces filter techniques to price-determined market entry and exit point trading rules. A long (short) position is taken in the Dow Jones Industrial Average (DJIA) when the price rises (falls) by some pre-selected percentage (the filter) above (below) a previous price. Each subsequent day, the price is checked to determine whether the reference price should be changed or whether a position should be closed out. Generally, Alexander finds that all filter sizes provide greater profits than fixed-length sell strategies. Fama and Blume (1966) apply Alexander s filter strategies to the DJIA after adjusting for dividends and transaction costs. They find that a fixed-length sell strategy is superior to the filter strategies for all but two stocks. Sweeney (1988) reexamines the Fama and Blume (1966) results. He individually tests Fama and Blume s DJIA stocks to determine if returns are persistent in future periods. In other words, do past winners tend to remain winners? He finds that 15 of the 30 stocks appear to yield profits 3

12 using a 0.5 percent filter rule. Tests of 14 of the 15 winner stocks (one stock does not survive to be tested) yield profits with high significance for floor traders in an out-of-sample time frame even when commissions are considered. Sweeney suggests that perhaps trading rules should be applied to individual stocks rather than stock indexes. In terms of selecting successful trading rules, these studies suggest that filters should be applied to pricing rules and that pricing rules should be applied to individual stocks rather than market indices. Neftci (1991) devises formal algorithms to mimic technical trading rules that identify buy and sell signals. He tests a 150-day moving average rule for the DJIA and finds it has significant predictive power from 1911 to A moving average shows the mean value of some data series over a specified time period. Series can consist of a variety of data items such as opening or closing prices, high or low prices, volume, etc. The choice of a specified time period is potentially infinite. Additional moving average rules plus support/resistance (trading-range breakout) rules are scrutinized by Brock, Lakonishok and LeBaron (1992) (BLL, hereafter). Support/resistance rules are predicated on the supposition that stocks trade in ranges determined by commonly held beliefs about a stock s value. Over time, however, investors expectations change. New expectations result in higher or lower price levels determined by the law of supply and demand. When this happens, a stock s price may change rapidly (the breakout), often accompanied by an increase in transaction volume. A new trading range is then established. BLL find that each of the 26 rules they apply to the DJIA significantly outperforms a benchmark of holding cash and that profits are substantially enhanced by the addition of a one percent filter. Twenty moving average rules and six support/resistance rules comprise BLL s 26 trading strategies. When testing numerous trading rules on the same data, questions of data-snooping arise. Are profits really attributable to trading rules or would some rules be expected to outperform the market solely on the basis of chance? To address this problem, Sullivan, Timmermann, and White (1999) use White s (2000) Reality Check bootstrap methodology which corrects for datasnooping effects. They test nearly eight thousand trading rules and find that some rules tested by BLL continue to outperform the benchmark even after adjusting for data-snooping. When they apply the best performing trading rule to out-of-sample data, however, they find that the rule does not outperform the benchmark. The inconsistent results leave the profitability of moving average and support/resistance technical trading strategies an unresolved issue. 4

13 The financial literature contains numerous tests of momentum and contrarian (reversal) theories. 5 Evidence involving these hypotheses is inexorably linked to technical analysis since both presuppose price predictability. Although variations abound, methodologies to test these strategies generally revolve around the buying or selling of stocks chosen from a larger set of stocks to be included in a portfolio. Stocks are selected based on their returns over a recent time period. The performance of the subset portfolio is then compared to the average performance of the universe of stocks from which they were originally chosen. Momentum strategies have typically been found to be profitable over short horizons consisting of months or weeks while contrarian strategies appear to be profitable for long horizons of three to five years. Conrad and Kaul (1998) provide a comprehensive review of the momentum and contrarian literature. While the substantiation of stock return predictability indirectly supports that technical traders can potentially earn abnormal profits, the inability of momentum or contrarian strategies to be profitable does not necessarily indicate that trading rules lack value. A foundation of technical analysis is the recognition that stock prices move in cycles. Trading strategies, successful or not, attempt to determine when reversals (up or down) will occur so that the investor may benefit by being in (out) of the market at the appropriate times. Technical analysis, therefore, encompasses both momentum and contrarian strategies, but these strategies represent only one component of technical analysis. Technical analysis commonly considers the volume of stock trades to be a clue in determining when price movements are normal fluctuations about an equilibrium price and when they represent a broader movement. Blume, Easley, and O Hara (1994) develop a theoretical model in which volume provides knowledge about the quality of a trader s information. They conclude that traders do best when they observe both price and volume. Cooper (1999) finds that including volume in decision making substantially improves the predictability of returns. Lee and Swaminathan (2000) also find a link between trading volume and future returns. Specifically, they note that, price momentum eventually reverses and that the timing of this reversal is predictable based on past trading volume. While they do not claim that their findings will yield abnormal returns, they do state that conditioning on past volume can potentially have economic significance. These 5 See, for example, Jegadeesh and Titman (1993, 2001, 2002), Rouwenhorst (1998), Chui, Titman, and Wei (2000), Asness (1997), Lee and Swaminathan (2000), Hong, Lim and Stein (2000), and Chang, Jegadeesh, and Lakonishok (1996). 5

14 periods. 6 A question remains as to whether potential trading rule profits are unique to specific studies suggest that considering trading volume when selecting individual stocks can potentially increase a technical strategy s profitability. Further support for the profitability of filters, moving averages and volume rules comes from Pruitt and White (1988) who by combining these three techniques, devise an overall successful trading strategy. They empirically test individual stocks selected due to a combination of their strength relative to the S&P 500, their trading volume, and their upward momentum. Once the established criteria in these three areas are met, stocks must still satisfy a final price filter requirement to be purchased. Stocks are sold when they signal a loss of upward momentum or when their price rises above another filter. Pruitt and White find that their multicomponent trading strategy outperforms a fixed-length sell strategy even after adjustments for risk and transaction costs. Although these findings offer strong support for technical analysis, the authors acknowledge making exceptions to the mechanical trading rule, albeit rare. Furthermore, their trading rule is applied to only nine years of data. The literature suggests that trading rules found to work well in one time frame do not necessarily work well in other strategies or whether they are associated with time-series factors commonly known to effect stock returns. The most widely used factors are those of Fama and French (1993): market return, size, and book-to-market equity. These factors are commonly employed in tests of market efficiency, mutual fund performance and other areas. In Fama and French (1996), they advocate that their 1993, three-factor pricing model captures many linked average-return anomalies. For example, firm characteristics such as earnings/price, book-to-market equity, cash flow/price, historical sales growth, and prior returns are known to be associated with average stock returns, but Fama and French (1993, 1996) find that their effects are subsumed by the three-factor Fama and French model. Diether, Malloy, and Scherbina (2002) employ the three-factor model to test hypotheses concerning the behavior of portfolios with different dispersion levels of analyst forecasts. Jegadeesh and Titman (2001) use sensitivity to the three factors as a means of characterizing momentum portfolio returns. They also determine the extent to which various portfolios load on the different factors. It is important to examine whether profits from technical 6 For example, see Brown, Goetzmann, Kumar (1998), Brock, Lakonishok, and LeBaron (1992), and Sullivan, Timmerman, and White (1999). 6

15 analysis are explained by their sensitivity to market returns, size, and book-to-market factors because if they are, any abnormal trading rule returns may not be indicative of an unknown, unique market inefficiency. If they are not fully explained by the factors, however, technical analysis could potentially incorporate an additional factor(s) that affect market returns. To date, the relationship between technical trading returns and factors known to impact the market has not been examined. This dissertation reexamines some of the technical trading rules appearing in the financial literature that exhibit evidence of profitability and/or the ability to forecast trends in future stock returns. It tests whether these strategies can select individual firms and identify firm-specific market entry and exit points that result in returns greater than the market return. It further tests whether trading rule returns are related to the stock characteristics of beta, size, and book-tomarket equity, known to impact stock returns, and if abnormal returns are present after controlling for these factors. It is essential to note that even if trading rule returns do not numerically exceed those of the market portfolio, it is still possible that the rules select stocks with below market risk (a beta less than one), large capitalization, or growth firms known to generally underperform the market. In this event, after controlling for the three Fama and French factors, abnormal returns may appear, i.e., a portfolio created by following technical rules could provide higher returns than would otherwise be anticipated for stocks expected to have below market returns. If trading rules select firms with specific market risk, size and/or book-to-market equity characteristics, trading rule returns, whether positive or negative, may be explained. Under these circumstances, it would suggest that the rules do not contribute to the incremental knowledge of the investor. If, however, superior returns cannot be fully attributed to these three factors, it is feasible that technical rules provide the investor with new information beyond that contained in the Fama and French (1993) asset pricing model. To summarize, my contribution to the financial literature is to provide empirically based information about technical trading rules beyond that currently known. Given the financial literature on momentum, market return anomalies, investor psychology, and persistence of returns (see Chapter 2 for details), I anticipate that some trading rules will yield returns exceeding those obtained by a stock portfolio that proxies for the market. The primary purpose of this dissertation is to provide insight as to whether these good trading rules, on average produce abnormal returns after accounting for the three Fama and French (1993) factors. If 7

16 trading rules do not generate abnormal returns, but tend to choose stocks with certain Fama and French risk characteristics, it suggests that trading rules merely proxy for factors already known to impact stock returns. In either case, a greater understanding of how technical trading rules function is contributed to the body of financial knowledge. Given that technical strategies are commonly employed by individual and professional investors, this is a worthy endeavor. The remainder of this dissertation is organized as follows. Chapter 2 contains a review of the relevant literature and the development of assumptions used to select trading rules and strategies. Chapter 3 presents the data and methodology. Chapter 4 contains the empirical results which are interpreted and summarized in Chapter 5. Chapter 5 also discusses analytical limitations. 8

17 CHAPTER 2 LITERATURE REVIEW AND TRADING RULE GUIDELINES Before the relationship between trading rules and the three factors of Fama and French can be explored, we must devise trading rules likely to produce abnormal stock returns. There is no attempt on the part of this dissertation to find an optimal trading rule since the academic literature clearly demonstrates that optimal strategies vary over different time periods. Instead, this dissertation researches the financial literature to suggest strategies likely to produce superior returns for further testing. The evidence suggests that merging multiple strategies into trading rules is likely to improve the probability of earning returns in excess of the market. Regardless of profit outcome, however, it is valuable to establish whether trading rule returns are related to risk factors already known to impact stock returns or whether they provide new information to the investor. The answer to this question may help explain the persistence of technical trading rules. Brock, Lakonishok and LeBaron (1992) (BLL, hereafter) test two categories of well known technical trading rules: moving averages and support/resistance rules, the latter of which they refer to as trading-range breakout rules. The DJIA from 1897 through 1986 comprises their data set. BLL acknowledge that this index is composed of large, well-known firms. They begin with a general definition of the moving average rules they later test. According to the moving average rule, buy and sell signals are generated by two moving averages of the level of the index a long-period average and a short-period average. Per this strategy, a buy (sell) signal is produced when the short moving average crosses the long moving average from below (above) so that its value is higher (lower) than that of the long moving average. Once a buy (sell) signal is received, BLL follow one of two general strategies. The first, called the fixed-length moving average (FMA), records index returns for the ten days following a crossover. The second, called a variable moving average (VMA), records index returns until another crossover signal is received. BLL examine five moving average rules: 1-50, 1-150, 5-150, 1-200, and where 9

18 the first figure is the number of days in the short moving average and the second figure is the number of days in the long moving average. BLL double the trading rules tested by applying a 1 percent filter to each of the moving average rules producing ten FMA and ten VMA rules. The filter is a band around the crossover price. To initiate a buy (sell) signal, the short moving average must rise (fall) 1 percent higher (lower) than its crossover value. If the short moving average fails to rise (fall) by this amount, no signal is produced. For the full sample period, BLL find that six of the ten VMA rules generate significant economic profits. Buy returns for all rules are positive and each rule generates a higher daily average return than the unconditional average return. All VMA sell strategies have negative returns and their average one-day return is significantly different than the unconditional average daily return. Additionally, in all but one instance, buy signals with a one percent filter yield higher returns than the same rule lacking a filter. For the one exception, the return is consistent regardless of whether a filter is applied. With sell signals, adding a one percent filter results in lower returns than when no filters are used suggesting that trading rules that identify good stock portfolio entry points do not necessarily identify good exit points. For the fixed moving average strategies (FMA), buy (sell) signals provide positive (negative) returns in all cases. Seven of the ten rules provide statistically significant profits. Buy signals for all rules have 10-day average returns significantly greater than the unconditional 10- day average return. Once again, buy signals with a 1 percent filter generally result in higher 10- day returns than when no filter is applied. All sell signals with filters have lower 10-day returns than when no filter is present. The second category of trading rules, support/resistance rules, generate buy (sell) signals when the index price surpasses (declines below) a pre-determined local maximum (minimum) price. To employ this strategy, BLL develop six trading rules. For the first three rules, local maximums and minimums are computed over the previous 50, 150 or 200 days. The next three rules add a 1 percent filter to the first three rules, thus, a buy (sell) signal is generated when the index price exceeds (falls below) a maximum price plus 1 percent (minimum price less 1 percent). Stocks are held for a 10-day period following buy or sell signals and returns are calculated over that horizon. All six rules generate statistically significant profits. All six buy rules have positive returns while all six sell rules have negative returns. Three of the six buy rules produce average 10

19 10-day returns significantly different than the 10-day unconditional average return, but only one sell rule is significantly different than the 10-day unconditional average return. BLL also perform a number of random walk simulations using bootstrap methodology. Their findings support their previous results, i.e., trading rules generate higher returns than anticipated by a random walk process. They further use bootstrap methodology to test their actual returns against simulated comparison series produced by four different null models. They find that the buy and sell signals result in returns that are higher or lower, respectively, than are likely to be produced by the null models. Furthermore, these returns are not explainable by autocorrelation in the returns, nor by market risk. BLL find that buy signals occur most frequently during periods of lower market volatility while sell signals seem to be triggered during periods of higher market volatility. Overall, they find strong support for the technical strategies [moving averages and support/resistance rules] that [they] explored. 7 The bootstrap process performed by BLL is intended to counter the presumed criticism that technical trading returns are a function of data-snooping rather than strategy value. At the time of their study, BLL were unable to, compute a comprehensive test across all rules. Such a test would have to take into account the dependencies between results for different rules 8. Sullivan, Timmermann and White (1999, hereafter STW) use White s Reality Check to determine whether superior performance is due to technical trading rules or to chance. Following BLL s methodology, STW expand the two general technical strategy categories of BLL to five groups: filter rules, moving averages, support and resistance, channel breakouts, and on-balance volume averages (OBV). According to STW, a channel is formed when the high price of a stock over some specified number of prior days is within a predetermined percent of the low price over the same time span. The channel excludes the current price. A buy (sell) signal is generated when the stock s closing price breaks out of the established channel by closing higher (lower) than the channel. OBV is a running total of transaction volume that indicates the level of interest in a stock. When a stock s closing price is above (below) that of the previous day s close, the entire amount of daily volume is added (subtracted) to (from) the cumulative volume total. STW test 7,846 specific trading rules on the DJIA by varying the parameters of their five 7 See Brock, Lakonishok, and LeBaron (1992), p See Brock, Lakonishok, and LeBaron (1992), p

20 technical strategy categories. This is far more than BLL s 26 rules. STW divide 90 of their 100 data-years into four subperiods that replicate the time frames of BLL. They also form an out-ofsample data set with ten years of data not available to BLL. This data set is used to determine that data-snooping is not responsible for BLL s results. STW find that while the number of long and short trades is approximately equal, the percentage of profitable long trades is substantially higher than the percentage of short trades, and the average profit related to short trades is half as large as that related to the long trades. Alexander (1961 and 1964) concludes that his mechanical filter strategy applied to the Dow Jones Industrial Average (DJIA) results in returns larger than those obtained using a buyand-hold policy. Fama and Blume (1966), not advocates of technical trading, fault the filter techniques of Alexander (1961 and 1964) for not including brokerage fees and for improper dividend adjustment. Alexander s methodology requires purchasing the DJIA on buy signal receipts and shorting the DJIA on sell signal receipts. Investors partaking in short sales generally reimburse stock lenders for dividends paid prior to a transaction s completion. Dividends, therefore, reduce short sale returns. FB examines the individual returns of the thirty Dow Jones stocks and finds that the short positions initiated by the filter rule are usually disastrous for the investor. 9 All but one stock has negative average returns when the filter strategy is applied to short sales. Long positions fare better. The findings of both FB and STW suggest my first guideline: Guideline 1: Trading strategies that exit the market on sell signals will have higher returns than strategies that short the market on sell signals. From their large number of technical strategies, STW find that the five-day moving average rule without a filter provides the highest average return from 1897 through In addition, the trading rule that yields the highest mean return in each of the four in-sample periods outperforms the benchmark strategy of being in cash even after data-snooping adjustments are made. The best performing rule, however, varies with the time period. Regardless of the rule category, short-term strategies of 2 to 5 days result in the highest average annualized returns. Over the five subperiods and the full sample period, no channel breakout rule was ever the best performer. Cooper (1999) applies filter rules to eight trading strategies. 9 Fama and Blume, 1966, p Two strategies select 12

21 individual winner (loser) stocks based on the size of their one-week lagged returns. The third and fourth strategies choose winner (loser) stocks according to both their one- and two-week lagged returns. To be included in winner (loser) portfolios, lagged returns for each lagged week must be within a specified filter range. The smallest winner filter ranges from 0 to less than 2 percent while the largest loser filter varies between less than 0 percent to greater than -2 percent. Each of the five successive filters rises incrementally by two percent until the largest winning portfolio filter is reached which requires a stock s lagged returns to be greater than ten percent. The most extreme losing portfolio filter selects stocks with lagged returns that have fallen more than ten percent. s are equally-weighted. Cooper s next four strategies combine lagged volume and lagged returns. Trading volume is measured as the lagged one-week percentage change in an individual security s volume divided by the number of its outstanding shares. Filters applied to low volume stocks range from a -75 percent or smaller volume change to a less than zero percent volume change in 15 percent increments. For high volume stocks, filters vary from between a zero percent volume change to a 250 percent or greater volume change in 50 percent increments. All portfolios are held for one week and then sold. Cooper examines the relationship between future price reversals and the size of filters. He finds that for large capitalization NYSE and AMEX stocks, both winner and loser portfolios exhibit increased price reversals as the absolute value of filter sizes rise. Cooper s investment strategy is to go long on the winner portfolio comprised of stocks expected to exhibit momentum generated, high future returns; hence, price reversals are undesirable. The loser portfolio, anticipated to drop in future value, is shorted. If loser portfolio returns reverse, investors profits will decrease; thus, here too, reversals are unwanted. Since both winner and loser portfolios have less reversals when the absolute size of the filter is small, smaller filters result in higher investor profit. According to Cooper, differences exist between the statistical significance and size of reversals for winner and loser portfolios. Winner portfolio reversals are generally not statistically significant, but half of the loser portfolio reversals are statistically significant. Again, the evidence supports the likelihood that going long on the winner portfolio is a better strategy than shorting the loser portfolio. In addition to examining the consistency of returns over filter sizes, Cooper also investigates the relationship between filter size and return consistency over various time 13

22 horizons. When using the smallest absolute filter size, both the winner and loser portfolios tend to have increasingly positive returns as the holding period moves from one week to four weeks to thirteen weeks and finally to fifty-two weeks. This finding provides evidence that winner portfolios maintain consistently upward short-term momentum while losing portfolios tend to have unprofitable short-term reversals that make shorting a losing portfolio an unattractive strategy. This conclusion is consistent with the financial literature suggesting that taking long positions on buy signals is likely to be more profitable than shorting on sell signals. It also supports Guideline 1. Cooper s results regarding filter rules lead directly to Guideline 2 since trading strategies that take long positions in stocks want positive momentum to continue. Guideline 2: Small filters will be more profitable with long position strategies than large filters. Cooper also investigates the possibility that lagged transaction volume used in conjunction with lagged returns improves the likelihood of forecasting future returns. He finds that conditioning on high volume in week t-1 tends to result in weaker return reversals for both winner and loser portfolios. Blume, Easley and O Hara (1994, hereafter, BEO) create a mathematical model in which changes in trading volume supply knowledge about the quality of traders information beyond that obtainable from price information alone. BEO s conclusions are consistent with those of Cooper, i.e., large price changes, both positive and negative, are inclined to be accompanied by increased volume. In their model, all traders who employ technical analysis benefit. While incorporating price into trading rules is helpful, traders who consider both price and volume do best. BEO s theory that large changes in transactional volume confirm the reliability of information is a plausible explanation for how technical analysis can generate profits in apparently efficient markets. It is also consistent with Cooper s findings that including lagged transaction volume improves the profitability of both winner and loser portfolios. Lee and Swaminathan (2000, hereafter LS) examine the usefulness of trading volume in predicting cross-sectional returns for various price momentum portfolios. Trading volume is defined as the mean daily turnover expressed as a percent during portfolio creation. Turnover is defined as the ratio of traded shares to shares outstanding. Monthly returns are computed for winner and loser portfolios held for various time lengths. Stocks are divided into deciles based on monthly returns, then separately sorted into three portfolios based on trading volume. LS 14

23 exclude Nasdaq stocks from selection because they believe that the double counting of dealer trades would result in methodological inconsistencies when comparing the trading volume of NASDAQ stocks to the trading volume of New York and American Stock Exchange listed stocks. LS are primarily interested in intermediate and long-term horizon returns while the mechanical rules of technical analysis are generally geared to short-term returns. Still, the findings of LS provide some useful insights with which to frame technical trading rules. LS find that past trading volume predicts both the magnitude and the persistence of future price momentum. Furthermore, they find that the majority of excess returns achieved by their volumebased strategies are attributable to changes in trading volume and that lower (higher) trading volume is associated with worse (better) current operating performance. The fact that a market statistic [volume] widely used in technical analysis can provide information about relative under- or over- valuation is surprising and is difficult to reconcile with existing theoretical work. 10 Guideline 3 is derived from the combined findings of LS, BEO and Cooper. Guideline 3: Utilizing strategies that select stocks having experienced previous high trading volume will improve profitability. Braden Glett comments in his book on technical analysis, Stock Market Stratagem: Loss Control and Management Enhancement (2003), In a sense, all investing is momentum investing. Who buys any investment unless there are hopes for a long upward trend to begin within a reasonable time? 11 Glett, a technician, goes on to suggest that investors seek stocks that appear to be moving upward. While mentioning other ways to determine individual stock trends, Glett concentrates on stocks making 52-week highs. This recommendation is consistent with momentum investment strategies and is supported in the academic literature by the work of George and Hwang (2004, hereafter GH). GH compare the momentum strategies of Jegadeesh and Titman (1993, hereafter JT) and Moskowitz and Glinblatt (1999, hereafter MG) with their own 52-week momentum strategy. The three strategies form winner and loser portfolios from stocks ranked by different historical performance criteria. JT sort stocks based on their own individual returns over a pre-determined number of past months. Using ten percent breakpoints, they form equally-weighted portfolios of winners and losers. Winner stocks are bought long and loser stocks are sold short. Zero-cost 10 See Lee and Swaminathan (2000), p See Glett, Stock Market Stratagem: Loss Control and management Enhancement (2003), p

24 portfolios, winner minus loser returns, are calculated for specified time periods. The best performing portfolio is one held for three months that contains stocks selected for their returns over the past 12 months. MG use a technique similar to that of JT. They also form selffinancing portfolios based on 6-month lagged returns except that they value-weight the returns of those stocks in the highest 30 percent and the lowest 30 percent rankings to calculate winner and loser portfolio returns. Next, MG create industry portfolios by value-weighting stocks within an industry as determined by their two-digit Standard Industrial Classification (SIC) codes. Next, they rank the industry portfolios by their previous six-month returns and purchase equal amounts of the three highest returning industries while short selling the three industry portfolios with the lowest returns. All positions are maintained for six months. Like JT and MG, GH form winner and loser portfolios determined by ranked individual stock returns over the past 6-months. GH, however, order stocks by the magnitude of the ratio of a stock s price on its last trading day of the prior month to its highest price during the 12-month period ending the same day. This 52-week strategy produces equally-weighted winner and loser portfolios consisting of the 30 percent of stocks with the largest ratio and the 30 percent of stocks with the smallest ratio, respectively. As with JT and MG, the winner portfolio is bought long while the loser portfolio is sold short. s for both portfolios are computed for six months. When compared to the JT and MG strategies, the 52-week high strategy of GH earns the largest and statistically most significant risk-adjusted returns. GH conclude that the proximity of the current price to the 52-week high price is superior to measures of past price performance in forecasting future returns. They also find that long-term reversals do not occur for stocks selected by the 52-week high strategy. This finding leads to a fourth guideline. Guideline 4: Strategies that select individual stocks close to or at their 52-week high price will tend to provide higher future returns than strategies that apply trading rules to portfolios comprised of stocks selected solely on the basis of high prior returns. Pruitt and White (1988, hereafter PW), combine moving averages with measures of cumulative volume and relative strength into a technical trading system they call CRISMA. They find that after adjustments for trade timing problems, risk, and transaction costs as high as 1% per stock trade, their system beats the market for substantial time periods. A buy signal is received when a stock complies with three conditions. First, the 50-day moving average must 16

25 cross the 200-day moving average from below at a time when the slope of the 200-day moving average is either zero or positive. This insures that the stock s price is trending upward over past time intervals. Secondly, the relative strength (RS) line must have a slope equal to or greater than zero for all points over the previous four weeks. A RS line is generally obtained by dividing the price of one item by another. In this instance, the numerator would be the stock price and the denominator would be some measure of the market such as the S&P 500. This criterion ensures the stock s performance has been minimally equal to that of the market measure for the last four weeks. Lastly, the cumulative volume graph must have a positive slope from its starting to its ending point over the preceding four weeks. This last condition is rooted in the assertion that increasing stock prices are related to rising transaction volume. The point at which a stock s 50-day moving average crosses its 200-day moving average from below establishes the stock s base price. A 10% buy filter is applied so that a stock is purchased when its price is 110% of its base price. A sell signal is generated when the stock s price falls below its 200-day moving average or it increases 120% above its base price. The first instance is believed to forecast a downward change in the stock s price and the second instance is presumed to imply an over-bought condition. PW use the Scholes and Williams (1977) market model, the ordinary least squares market model, the market-adjusted returns model and the mean-adjusted returns model to estimate expected returns. 12 The period of one through 200 days prior to the date that each stock is purchased is used to estimate model parameters. They then compute the excess stock return over a future holding period by subtracting the cumulative expected returns from the cumulative actual returns over the period. PW find that the CRISMA system outperforms a fixed-length sell strategy. Thus far, this review has examined academic studies justifying the development of technical trading rules that encompass measures of past stock price and volume while adjusting for noise fluctuations through the application of filters. Although the financial literature generally probes aspects of stock fundamentals and investor psychology rather than mechanical trading rules, many studies focus on understanding factors that could potentially impact market returns. The effort often begins with a theory from which a mathematical model is developed. Empirical data is then used to test model specifications. 12 See Brown and Warner (1985) for a discussion of the Scholes and Williams (1977) market model, the ordinary least squares market model, the market-adjusted returns model and the mean-adjusted returns model. 17