Home › Blog and Video Interviews › Page 5

Enhancing reversal strategies with artificial intelligence

Posted on by Jeff@AlphaInterface.com
Foreign exchange market chart

Foreign exchange market chart


Li, Hoi, Zhao, and Gopalkrishnan (2012), from Nanyang Technological University, Singapore, and the Deloitte Analytics Institute (Asia), Singapore, developed a very sophisticated algorithm for portfolio selection that they labelled the “Confidence Weighted Mean Reversion” (CWMR) strategy. The principle of mean reversion in portfolio selection is to invest in stocks that performed poorly during the previous period. The portfolio is long-only and, therefore, simply avoids stocks that have performed well in the previous period, rather than shorting them. CWMR is a machine-learning algorithm that responded to both the mean and the variance of the portfolio, and strove to always maintain a Gaussian, or normal, distribution.

The CWMR strategy was tested on eight extensive data sets, with daily data from 1962 to 2011, as well as intraday and weekly data. The data sets included stocks from NYSE, the Toronto Stock Exchange (TSE), as well as a collection of global stock indices (MSCI). Intraday data came from the DJIA 30 and NASDAQ 100 stocks.

Given the complexity of the data sets, the authors determined to test six versions of the CWMR strategy by comparing them to 13 other strategies. These included:

  • Buy and hold;
  • Select the best individual stock (or index) from the data set (a hindsight strategy);
  • Best constant rebalanced portfolio (a hindsight strategy);
  • And ten other algorithmic methods proposed by other researchers in previously published studies.

The experimental tests measured the cumulative wealth attained by each strategy at the end of the data period.

The result was described by the authors as “somehow beyond imagination.” The CWMR algorithms dominated the competition, in all instances but two. Typically, the cumulative wealth achieved by the CWMR strategy, in daily data series, was many multiples better than either of the two hindsight methods. Particularly notable were the intraday tests in which the trades issued by the CWMR strategies achieved success ratios exceeding 63%.

Further analysis indicated that these results could be sustained with transaction costs and that drawdowns were reasonable in comparison to other methods.

Li and Hoi (2012) submitted a paper with yet another machine-learning, mean-regression algorithm that they labeled the “online learning moving average reversion” (OLMAR) strategy. In a series of tests very similar to those reported above, they reported that OLMAR even outperformed the CWMR strategy.

Li, Zhao, Hoi, and Gopalkrishnan (2012) reported on yet another algorithm that they label the “passive aggressive mean reversion” (PAMR) strategy. This strategy attained a very high level of performance, comparable with those previously mentioned. Furthermore, the data set and MatLab source codes have been made available online at http://www.cais.ntu.edu.sg/~chhoi/PAMR/.

Trading strategy: This strategy requires a team with the resources and knowledge to implement these unique artificial intelligence applications. The research cited suggests that the potential payoff would justify the investment.

Posted in Book Two: Twenty-Four Trading Strategies Based on Scientific Findings About Technical Analysis Tagged with: , , , , , , , , , , , , , , , , , , , , , , , , , ,

Do reversal strategies work?

Posted on by Jeff@AlphaInterface.com

Chart2 Small


De Groot, Huij, and Zhou (2012), of Robeco Quantitative Strategies and Erasmus University, Rotterdam, The Netherlands, measured the profitability of reversal (or mean reversion) strategies applied to various market cap segments of the U.S. stock market. They examined stocks from 1990 through 2009.

Their reversal portfolios were constructed by daily sorting all available stocks into five equally sized portfolios based on their past-week (i.e., five trading day) returns. They assigned equal weights to the stocks in each quintile. The reversal strategy was to buy the twenty percent of stocks with the lowest returns over the past week, and to assume a short position in the twenty percent of stocks with the highest returns. The portfolios were rebalanced every day.

They found that, when their pool of stocks included the 1,500 largest U.S. stocks, the reversal portfolio gained approximately 0.62% per week, for an annualized profit of approximately 31% a year (before accounting for transaction costs). At the same time the reversal strategy had an extremely high portfolio turnover of 677 percent per week. They found that their average period for holding a stock was less than three days. Once trading costs were taken into account the profitability of the reversal strategy greatly diminished.

They then considered the results for the 500 and 100 largest U.S. stocks. Interestingly, the reversal strategies for the largest 500 and 100 stocks earned slightly higher returns than the reversal strategy for the largest 1,500 stocks. Moreover, it appeared that the impact of trading costs on the profitability of the strategy was much lower for their samples of large cap stocks.

To further reduce transaction costs, they employed a slightly more sophisticated approach. Rather than rebalancing their portfolio each day, they waited to close their long or short positions until those stocks moved into the fifty percent of winner, or loser, stocks ranked on past return. The stocks with the lowest or highest past-week return only then replaced those stocks. This “smart” approach held the same number of stocks in the portfolio as the standard approach. The holding period with the “smart” approach was flexible for each stock. They found that the effective holding period of a stock on average was approximately six days for this strategy.

The resulting reversal profits ranged between thirty and fifty basis points per week, after transaction costs. This computed to about 15% to 25% per year and was significant from both a statistical and an economical point of view.

A further enhancement of profits was obtained when the trading rule was modified again so that long positions were not closed until those stocks had moved above the seventieth percentile of profitable stocks for the past week. Short positions were not closed until those stocks had moved below the thirtieth percentile for the previous week.

Trading strategy: Select S&P 500 stocks. Use a formation or ranking period of five trading days. Buy those stocks in the lowest return quintile. Sell those stocks in the highest return quintile. Hold long until those stocks have moved above the seventieth percentile. Hold short positions until those stocks have moved below the thirtieth percentile.

Posted in Book Two: Twenty-Four Trading Strategies Based on Scientific Findings About Technical Analysis Tagged with: , , , , , , , , , , , , , , , , , , , , , , , ,

Log periodic signature associated with bubbles and crashes.

Posted on by Jeff@AlphaInterface.com


Johansen, Sornette, and Ledoit (1999) and Johansen, Ledoit, and Sornette (2000), all from UCLA at that time, argued that financial bubbles and crashes exhibited unique mathematical signatures known as log-periodic oscillations. This refers to a sequence of oscillations with progressively shorter cycles of a period decaying according to a geometrical series. The pattern is shown clearly in the chart below of the Credit Suisse First Boston Russia Index (ROSI). The crash there began in the second half of 1997.

Reprinted from Johansen, Sornette, and Ledoit (1999) with permission.

Reprinted from Johansen, Sornette, and Ledoit (1999) with permission.

Johansen, Sornette, and Ledoit (1999) documented 8 unrelated crashes from 1929 to 1998, on stock markets as diverse as the US, Hong-Kong or the Russian market and on currencies. In addition, they documented a significant bubble on Wall Street ending in 1962 as well as “anti-bubbles”on the Nikkei since 1990 and the Gold (after the 1980 bubble maximum). They also showed that the Russian bubble crashing in Aug. 1997 had close to identical power law and log-periodic behavior to the bubbles observed on Wall Street, the Hong-Kong stock market and on currencies. To their knowledge, no major financial crash preceded by an extended bubble had occurred in the prior 2 decades without exhibit a log-periodic signature.

The New York stock exchange index S&P500 from July 1985 to the end of 1987 corresponding to 557 trading days. The ◦ represent a constant return increase in terms of an exponential. Reprinted from Johassen, Ledoit, and Sornette (2000) with permission.

The New York stock exchange index S&P500 from July 1985 to the end of 1987 corresponding to 557 trading days. The ◦ represent a constant return increase in terms of an exponential. Reprinted from Johassen, Ledoit, and Sornette (2000) with permission.

Sornette,  Woodard, and Zhou (2009), at this time affiliated with the Swiss Federal Institute of Technology (i.e., ETH) in Zurich and the East China University of Science and Technology, Shanghai, analyzed oil prices (in USD and other currencies) and determined that, in 2008, they were rising in an unsustainable, “faster than exponential” manner. Their post-crash analysis, using the log periodic method of Johansen, Sornette, and Ledoit (1999), confirmed that the oil peak in July 2008 occurred within the expected 80% confidence interval predicted with data available in their pre-crash analysis.

Time series of observed prices in USD of NYMEX Light Sweet Crude and simple log-periodic fits for potential bubble and crash. The shaded areas represent the 80% confidence level. Reprinted from Sornette, Woodard, and Zhou (2009) with permission from Elsevier.

Time series of observed prices in USD of NYMEX Light Sweet Crude and simple log-periodic fits for potential bubble and crash. The shaded areas represent the 80% confidence level. Reprinted from Sornette, Woodard, and Zhou (2009) with permission from Elsevier.

The authors applied similar tools to the analysis of the Las Vegas real estate market from 2000 to 2008 (Zhou and Sornette, 2008).

 

Posted in Bubbles and Crashes Tagged with: , , , , , , , , , , , , , , , , , , , , , , ,

Do momentum strategies work for weekly stock returns?

Posted on by Jeff@AlphaInterface.com


Gutierrez and Kelley (2008), from the University of Oregon and the University of Arizona, hypothesized that weekly stock returns would show a momentum effect. They were aware of research from the 1990s that found a short-term reversal effect. In those studies, stocks with the lowest returns over the prior week or month outperformed stocks with the highest returns over the prior period. They expected that this relatively short-term reversal could also be compatible with long-term momentum.

They examined weekly U.S. stock prices from 1983 through 2003, for all listed stocks priced above five dollars. Each week, they ranked stocks based on their returns over the week. They formed a portfolio composed of a long position in the top decile of winning stocks and a short position in the bottom decile, i.e., the losers. They tracked the prices during the following 52 weeks.

They found that the subsequent momentum profits were strong enough to offset the initial reversal and to produce a significant momentum effect over the full year following portfolio formation. Thus, as the chart below shows, they were able to confirm the previously documented, short-term reversal and document a long-term momentum effect.

The vertical axis shows cumulative profits as a percentage. The horizontal axis shows time elapsed in weeks since the original portfolio formation. Reprinted from Gutierrez and Kelley (2008) with permission.

The vertical axis shows cumulative profits as a percentage. The horizontal axis shows time elapsed in weeks since the original portfolio formation. Reprinted from Gutierrez and Kelley (2008) with permission.

 Pan, Tang, and Xu (2013), from Beijing University, China, applied a similar momentum strategy to the Chinese stock market, with one unique and significant variation. They did not divide stocks into deciles, or quintiles, with an equal number of stocks in each. Instead, they created quintiles based on equal profit or loss intervals for the preceding week. This meant that there were a much smaller percentage of stocks in the outermost, positive and negative quintiles. They examined data from 1990 through 2009.

The overall result of their hedge portfolio, based upon a one-week ranking for formation period, is shown in the following chart. The initial momentum was followed by a brief reversal, and then by long lasting revived momentum. More than half of the profit was realized in the first three weeks. The authors applied the same method to other Asian markets. They found a similar weekly momentum effect in Hong Kong, Taiwan, Korea, Thailand, and Indonesia.

The vertical axis shows the annualized return percentage. The horizontal axis shows the number of weeks the portfolio is held. The green line represents the cumulative return, while the black line shows the weekly return. Reprinted from Pan, Tang, and Xu (2013) with permission from Elsevier (the black and green lines have been augmented for clarity).

The vertical axis shows the annualized return percentage. The horizontal axis shows the number of weeks the portfolio is held. The green line represents the cumulative return, while the black line shows the weekly return. Reprinted from Pan, Tang, and Xu (2013) with permission from Elsevier (the black and green lines have been augmented for clarity).

 Trading strategy: Use a one-week formation or ranking period. Skip two weeks between the end of the ranking period and the beginning of the holding period. Hold for two weeks or longer. Results will be different in different markets and is likely to be more pronounced in the more volatile emerging markets such as China.

Posted in Book Two: Twenty-Four Trading Strategies Based on Scientific Findings About Technical Analysis Tagged with: , , , , , , , , , , , , , , , , , , , , , , , , , ,

Have price jumps systematically followed analyst announcements?

Posted on by Jeff@AlphaInterface.com

price jump


Suzanne S. Lee (2012), from the Georgia Institute of Technology, investigated the predictability of intraday jump arrivals in U.S. stock markets. Using high frequency data from 1993 through 2008, for Dow Jones Industrial Average component stocks, she demonstrated that jumps have been predictable to some extent. Her sample included a total of 7,964 jump events. She found predictable causes for about a third of these jumps.

She employed a very complex mathematical formula for determining jump sizes and the probabilities of their occurrence (which varied according to the time of day). Jumps were determined by sampling prices every 15 minutes. Typical jumps in her study entailed price movements of 1.8% up or down. The average company experienced less than two such jumps per month.

Analyst recommendation announcements were one of the predictors that she discovered. Price jumps tended to occur within the first 30 minutes following analyst recommendations. This predictor accounted for 19% of all predictable jumps and about 6.5% of all jumps. Analyst announcements produced more jumps than any other firm-specific predictor in her study (probably because analyst announcements were, themselves, the most frequent). She described these jumps as idiosyncratic, meaning that their prediction was less precise than other jumps that she described as systematic. She also showed that jumps tend to cluster by size and that jumps tend to occur within three hours of the arrival of previous jumps of the same stock.

Trading strategy: For intraday trading, check news feeds for all analyst announcements. For upgrades buy. For downgrades, assume a short position. If there has been no stock jump, close positions after half an hour. If the stock price did jump in the anticipated direction, hold the position for three more hours or until the close.

Posted in Book One: Forty Trading Strategies Based On Scientific Findings About Analysts' Forecasts Tagged with: , , , , , , , , , , , , , , , , , , , , , , , , , ,

Does one price jump lead to another in the same direction?

Posted on by Jeff@AlphaInterface.com



Friesen, Weller, and Dunham (2009), from the University of Nebraska, the University of Iowa, and Creighton University, Omaha, Nebraska, theorized that confirmation bias (i.e., the human tendency to give added weight to information that confirms one’s own beliefs) was the psychological mechanism underlying the observed success of technical trading strategies. Therefore, they hypothesized that sequential price jumps for a particular stock would be positively autocorrelated. A price jump, in other words, tends to be followed by a subsequent jump in the same direction.

To test this hypothesis, they examined intraday (five minute bar) prices of S&P 100 stocks from 1999 through 2005. The average price jump in this study was 1.39% occurring with a five-minute interval. On any given day, the probability of a jump was about twelve percent, meaning that jumps occurred on average, once every eight trading days for a given stock. On days when such jumps occurred, the jump accounted for 90% of the daily price movement.

They found a low-magnitude, but statistically significant correlation between the direction of a price jump and that of subsequent jumps. They also found the correlation to be stronger for jumps that occurred on days of high price volatility. They further found the correlation to be stronger for jumps that were larger in magnitude.

This chart, based on data from Friesen, Weller, and Dunham (2009), shows a positive autocorrelation between jumps and subsequent jumps that increases in strength as the jump size increases. The absolute jump size is expressed as a fraction of the price. Thus a jump size of 0.025 is 2.5%. An auto correlation of 0.1 suggests that the probability of a subsequent jump in the same direction is 10% greater than chance, or about 55%.

This chart, based on data from Friesen, Weller, and Dunham (2009), shows a positive autocorrelation between jumps and subsequent jumps that increases in strength as the jump size increases. The absolute jump size is expressed as a fraction of the price. Thus a jump size of 0.025 is 2.5%. An auto correlation of 0.1 suggests that, roughly speaking, the probability of a subsequent jump in the same direction is 10% greater than chance, or about 55%.

Trading strategy: After a significant price jump, the odds are about six to nine percent greater than chance expectation that the next price jump will be in the same direction. The larger the jump, the greater the odds. To exploit this anomaly, wait for a pullback, and then take a position in the direction of the preceding jump. Hold the position for an average of eight days or until after the next jump.

Posted in Book Two: Twenty-Four Trading Strategies Based on Scientific Findings About Technical Analysis Tagged with: , , , , , , , , , , , , , , , , , , , , , , ,

Do investors exhibit cyclic patterns of behavior?

Posted on by Jeff@AlphaInterface.com
Does seasonal affective disorder (SAD) influence the financial markets?

Does seasonal affective disorder (SAD) influence the financial markets?

Kamstra, Kramer and Levi (2003) – from the Federal Reserve Bank of Atlanta, the University of Toronto, Canada, and the University of British Columbia, Canada – examined returns from several large stock exchanges around the world at varying latitudes and on both northern and southern hemispheres. They found a Seasonal Affective Disorder (SAD) induced seasonal pattern in returns as depressed and risk-averse investors shunned risky assets in the fall and resumed their risky holdings in the winter, leading both to returns in the fall that were lower than average, and to returns following the longest night of the year that were higher than average.

Following up on this general finding Lo and Wu (2009), from the University of British Columbia, Canada, studied whether and to what extent stock analysts were affected by Seasonal Affective Disorder. They examined quarterly earnings forecasts from 1980 to 2006. Their data included over four million observations. They found that analysts were more pessimistic in the fall season, as indicated by their earnings forecasts as well as their forecast revisions. However, this difference, while statistically significant was relatively small in magnitude. Forecasts made in the fall were net positive for one quarter out – but net negative for two, three, and four quarters in the future.

Dolvin, Pyles, and Wu (2009) – from Butler University, Indianapolis, Indiana, the College of Charleston, South Carolina, and the State University of New York, Oneonta – extended this analysis. They examined the error in stock analyst annual earnings estimates relative to actual earnings from 1998 through 2004, specifically concentrating on the potential pessimistic bias that is associated with SAD.

Their results suggested that analysts were generally optimistic in their forecasts but significantly less so during fall and winter, low sunshine SAD months. They also found this pattern was most pronounced for analysts located in northern states, who should have been the ones most affected by the disorder. They concluded that while the effect of SAD appeared to be present, it seemed to overcome the, well-documented existing positive bent in earnings forecasts, thereby making estimates more accurate.

Trading strategy: Note that analysts, as well as investors generally, have a tendency to be less enthusiastic in the September and October, particularly in northern regions. This information, by itself, is insufficient for a specific trading strategy. However, an edge can be gained by looking at seasonal patterns of specific securities. This topic will be covered in a future Alpha Interface book.

Posted in Book One: Forty Trading Strategies Based On Scientific Findings About Analysts' Forecasts Tagged with: , , , , , , , , , , , , , , , , , , , , , , , ,

Are exponential moving averages superior to simple moving averages?

Posted on by Jeff@AlphaInterface.com

Camillo Lento (2010), of Lakehead University, Thunder Bay, Ontario, Canada, examined dual moving average crossover systems, comparing simple with exponential moving averages. The analysis was conducted on the exchange traded funds for the S&P 500 (SPY), NASDAQ (QQQ), and Dow Jones Industrial Average (DIA) from January 1999 to April 2009. Three different moving average crossover systems were tested: 1, 50; 10, 50; and 1, 200. Lento found that, in every instance, the exponentially smoothed moving averaged yielded improved results.

The chart above shows the annual profit of each moving average crossover system, as compared to a buy and hold strategy. Red is for SPY. Blue is for DIA. Green is for QQQ. The dark red, blue, and green bars represent the simple moving average (SMA) systems. The light red, blue, and green bars represent the exponential moving average (EMA) systems. Based on data from Lento (2010).

The chart above shows the annual profit of each moving average crossover system, as compared to a buy and hold strategy. Red is for SPY. Blue is for DIA. Green is for QQQ. The dark red, blue, and green bars represent the simple moving average (SMA) systems. The light red, blue, and green bars represent the exponential moving average (EMA) systems. Based on data from Lento (2010).

Although all the exponential moving average crossover systems outperformed the buy and hold strategies, the only strategies to independently achieve statistical significance were the (1, 50) and (10, 50) EMA strategies with the NASDAQ QQQ exchange traded fund.

Trading strategy: Use exponential moving averages, rather than simple moving averages, in all applicable situations. Also notice that the strategy was most successful with the more volatile QQQ exchange traded fund. 

Posted in Book Two: Twenty-Four Trading Strategies Based on Scientific Findings About Technical Analysis Tagged with: , , , , , , , , , , , , , , , , , , , , , , , , ,

Do momentum and trend-following strategies work in futures markets?

Posted on by Jeff@AlphaInterface.com

Shen, Szakmary, and Sharma (2007) – from Alabama A&M University, the University of Richmond in Virginia, and the University of Southern Illinois in Carbondale – conducted basic momentum tests on commodity futures at the end of each calendar month, for over forty years from July 1959 to December 2003. The study included coffee, corn, cocoa, crude oil, cotton, feeder cattle, gold, copper, heating oil, orange juice, gasoline, wheat, lumber, live cattle, lean hogs, natural gas, oats, palladium, pork bellies, platinum, rough rice, soybeans, sugar, silver, and soybean meal.

Their results showed that momentum strategies with formation and holding periods of up to nine months earned significant, positive, abnormal returns. When they divided the total holding period returns by the length of the holding period, then the intrinsic profitability was highest at the shortest horizons, e.g., one to two months. The strategies were quite profitable, earning excess returns of about one percent per month or slightly more (without leverage). They also noted that, in contrast to findings in equity markets, their results showed no firm evidence of long-horizon contrarian profits.

Cumulative returns to a 2-month formation-period momentum strategy by post-formation month. The graph provides raw returns to a 2-month formation-period momentum strategy in which the indicated long and short positions in nearby commodity futures contracts are maintained for 30 months after portfolio formation. Reprinted from Shen, Szakmary, and Sharma (2007) with permission.

Cumulative returns to a 2-month formation-period momentum strategy by post-formation month. The graph provides raw returns to a 2-month formation-period momentum strategy in which the indicated long and short positions in nearby commodity futures contracts are maintained for 30 months after portfolio formation. Reprinted from Shen, Szakmary, and Sharma (2007) with permission.

Szakmary, Shen, and Sharma (2010) conducted a near replication of the above study, using even more data, and focusing on trend-following strategies. They found that all values for the dual moving average crossover and channel strategies that they implemented yielded positive, mean excess returns after transactions costs in at least 22 of the 28 markets. When they pooled their results across markets, they showed that all the trading rules earned significant positive returns that prevailed over most sub-periods of the data as well. These results were robust regarding the set of commodities, distributional assumptions, data-mining adjustments, and transactions costs.

Ayora and Torro (2010), from the University of Valencia, Spain, examined momentum in the financial futures markets. They examined eighteen different futures contracts from January 1002 to December 2007. The list included seven currency futures (Euro/USD, Yen/USD, Canadian dollar/USD, British pound/USD, Australian dollar/USD, Swiss franc/USD, and U.S. dollar index); six stock index futures (CAC 40, FTSE 100, Nikkei 225, S&P 500, Swiss market index, Dow Jones EURO STOXX 50); and five fixed income futures (Euro bund ten-year, Canadian government bond ten-year, U.S. Treasury note two-year, U.S. Treasury note five-year, and U.S. Treasury note ten-year).

Considering profitability in the formation period, they then ranked the futures into one of three portfolios (P1, P2, and P3), where P1 contained the six futures with highest past returns (winners); and P3 contained the six futures with lowest past returns (losers). The remaining six futures were included in the P2 portfolio. Ayora and Torro then computed portfolio returns for each holding period. Finally, the momentum portfolio return was calculated: P1 – P3. That is, the momentum portfolio contained long positions in the past winning contracts and short positions in the past losing contracts.

They employed five different formation and holding periods of one, three, six, twelve, and 24 months. Their basic result was statistically significant for the six and twelve month formation and holding periods of equal length. The annual returns were 6.488% and 7.847% (without leverage) respectively.

Trading strategy: Momentum and trend following strategies are the most fundamental to be found in technical analysis. While they are not perfect, they appear to have merit for virtually all markets and in most timeframes. They are particularly appropriate for long-term investors. The research cited above does not provide consistent data regarding formation and holding period lengths. Therefore, traders interested in this approach will want to conduct further tests of their own.

Posted in Book Two: Twenty-Four Trading Strategies Based on Scientific Findings About Technical Analysis Tagged with: , , , , , , , , , , , , , , , , , , , , , , , , , ,

Does momentum-trading work with monthly charts, part 2

Posted on by Jeff@AlphaInterface.com

Wei and Yang (2012), from the University of Toronto, Canada, examined momentum in a large sample of U.S. stocks from 1964 to 2009. At the beginning of each month, stocks were sorted quintiles based on their realized past returns. Equally weighted portfolios were formed and held for up to 12 months. The overall results show significant and consistent momentum, as is evidenced in the following chart:

The average monthly return percentage is based on the combined profits of both long and short trades. Based on data from Wei and Yang (2012).

The average monthly return percentage is based on the combined profits of both long and short trades. Based on data from Wei and Yang (2012).

Interestingly, although the momentum effect was statistically significant, the researchers also discovered a subpopulation of stocks for which there was a significant reversal effect: large-cap stocks of low volatility. For small stocks, no reversals were observed. For both large and small cap stocks, the momentum effects were stronger when volatilities were higher.

The following chart show large cap stocks separated into quintiles according to volatility. The formation or ranking period is for one month. This finding suggests that, if you wish to benefit from a volatility-based portfolio, you would be wise to eliminate large cap stocks of low volatility.

Based on data from Wei and Yang (2012).

Based on data from Wei and Yang (2012).

While the study of Gutierrez and Kelley (2008) [presented elsewhere in Book Two] suggests that there is a time-sequence relationship between reversal and momentum, Wei and Yang’s findings point to the simultaneous occurrence of these opposite tendencies in different populations of large cap stocks. These distinctly different findings are not necessarily inconsistent with each other. However, the relationship between reversal and momentum is an area of inquiry that clearly merits further research.

Trading strategy: Focus on Australian and Canadian stocks or on other markets where volatility is relatively high. Use a six-month formation or ranking period. Skip one month between the end of the ranking period and the beginning of the holding period (to mitigate potential reversal effects). Hold long and short positions for one to three months. 

Posted in Book Two: Twenty-Four Trading Strategies Based on Scientific Findings About Technical Analysis Tagged with: , , , , , , , , , , , , , , , , , , , , ,

Book Three: Trading With The News

Learn about a news-based trading system that yielded a back-tested, average annualized, compounded return from 2000 to 2011 of 58.6%.

“Only once you’ve done your homework will you be able to understand how the stock market works and learn to distinguish between news and noise.” Maria Bartiromo, Use The News

Book Two: Technical Analysis

Learn about the "trend recalling" algorithm that yielded researchers a simulated annual return of greater than 400% in multiple tests.

“The scientific method is the only rational way to extract useful knowledge from market data and the only rational approach for determining which technical analysis methods have predictive power.”
David Aronson, Evidence Based Technical Analysis

Book One: Analysts’ Forecasts

Learn the strategy, based on analysts' revised forecasts, that yielded researchers an average of 1.13% - 2.19% profit per trade, for trades lasting one to two days?

Learn how certain analysts' recommendations, following brokerage hosted investment conferences, yielded profits of over 3% during a two-day holding period?

Learn how researchers found an average profitability of 1.78% for two-hour trades following an earnings announcement?

"This set of tools can help both ordinary and professional investors alike to re-think and re-vitalize their stock picking, timing and methods. A young, aspiring Warren Buffet could put this book to good use."
James P. Driscoll, PhD, investor

Statistically Sound Machine Learning for Algorithmic Trading of Financial Instruments by David Aronson (software included)

Evidence-Based Technical Analysis by David Aronson

Archive of Earlier Posts