Technical Trading

What drives the profitability of algorithmic long-short statistical arbitrage trading (such as pairs trading) of liquid U.S. stocks? In their September 2015 paper entitled “Performance v. Turnover: A Story by 4,000 Alphas”, Zura Kakushadze and Igor Tulchinsky examine portfolio turnover and portfolio volatility as potential net return drivers for such trading. Their data source is 4,002 randomly selected portfolios (essentially synonymous with “alphas” in their lexicon) from a substantially larger survivorship bias-free pool of real trading accounts. Position holding periods for sampled portfolios range from 0.7 to 19 trading days. The authors exclude 366 portfolios with negative performance and then remove 347 portfolios as outliers for a residual sample of 3,289 portfolios. Using daily closing prices for holdings in these portfolios over an unspecified sample period, they find that: Keep Reading

Does simple technical analysis based on moving averages work on high-frequency spot gold and silver trading? In their August 2015 paper entitled “Does Technical Analysis Beat the Market? – Evidence from High Frequency Trading in Gold and Silver”, Andrew Urquhart, Jonathan Batten, Brian Lucey, Frank McGroarty and Maurice Peat examine the profitability of 5-minute moving average technical analysis in the gold and silver spot markets. They consider simple moving average (SMA), exponential moving average (EMA) and weighted moving average (WMA) crossing rules. These rules buy (sell) when a fast moving average crosses above (below) a slow moving average. They start with four commonly used parameter settings, all using a fast moving average of one interval paired with a slow moving average of 50, 100, 150 or 200 intervals [(1-50), (1-100), (1-150) or (1-200)]. They then test all combinations of a fast moving average ranging from 1 to 49 intervals and a slow moving average ranging from 50 to 500 intervals, generating a total of 66,297 distinct rules. To compensate for data snooping bias, they specify in-sample and out-of-sample subperiods and test whether the most successful in-sample rules work out-of-sample. They also use bootstrapping as an additional robustness test. Using 5-minute spot gold and silver prices during January 2008 through mid-September 2014, they find that: Keep Reading

Are expert technicians or fundamentalists better forecasters of short-term and intermediate-term asset returns? In the August 2015 version of their paper entitled “Talking Numbers: Technical versus Fundamental Recommendations”, Doron Avramov, Guy Kaplanski and Haim Levy assess the economic value of dual technical and fundamental recommendations presented simultaneously on “Talking Numbers”, a CNBC and Yahoo joint broadcast… “featuring fundamental and technical recommendations before and during the market open. Dual recommendations are made by highly experienced analysts representing prominent institutions.” Recommendations address both individual stocks and asset classes, including U.S. and foreign broad equity indexes, sector/industry equity indexes, bonds, commodities and exchange rates. Using 1,000 dual recommendations on 262 stocks and 620 dual recommendations on other assets, along with associated price data, during November 2011 through December 2014, they find that: Keep Reading

Stock return anomaly studies based on firm accounting variables generally employ annually reformed portfolios that are long (short) the tenth of stocks expected to perform well (poorly). Does adding monthly portfolio updates based on technical stock price trend measurements boost anomaly portfolio performance? In the June 2015 version of their paper entitled “Anomalies Enhanced: The Use of Higher Frequency Information”, Yufeng Han, Dayong Huang and Guofu Zhou test eight equal-weighted long-short portfolios that combine annual screening based on a predictive accounting variable with monthly screening based on a simple moving average (SMA)-based stock price trend rule. The eight accounting variables (screened in June based on prior December data) are: (1) book-to-market ratio; (2) gross profitability; (3) operating profitability; (4) asset growth; (5) investment growth; (6) net stock issuance; (7) accruals; and, (8) net operating assets. The price trend screen excludes from the long (short) side of the portfolio any stock for which 50-day SMA is less than (greater than) 200-day SMA at the end of the prior month. Using accounting and daily price data for a broad sample of U.S. stocks during July 1965 through December 2013, they find that: Keep Reading

What is the best stock pairs trading method? In their June 2015 paper entitled “The Profitability of Pairs Trading Strategies: Distance, Cointegration, and Copula Methods”, Hossein Rad, Rand Kwong Yew Low and Robert Faff compare performances of three pairs trading methods as applied to U.S. stocks.

1. Distance – Select the 20 stock pairs with the smallest sum of squared differences in initially normalized dividend-adjusted prices during a 12-month formation period. Then re-normalize prices of selected pairs and initiate equal long-short trades when prices diverge by at least two formation-period standard deviations during a subsequent six-month trading period. Close trades when prices converge or, if not, at the end of the trading period. Re-open trades if prices diverge again withing the trading period.
2. Cointegration – Sort stock pairs based on sum of squared differences in initially normalized dividend-adjusted prices during a 12-month formation period. Then determine which pairs are cointegrated (exhibit a reliable mean-reverting relationship) during the formation period, and select the 20 cointegrated pairs with the smallest sum of squared differences. Over the subsequent six-month trading period, trade pair divergences and convergences based on cointegration statistics, with long and short position sizes also determined by these statistics.
3. Copula – Select the 20 stock pairs with the smallest sum of squared differences in initially normalized dividend-adjusted prices during a 12-month formation period. Then construct best-fit copulas for each pair and use copula statistics to determine when pair prices diverge and converge during a subsequent six-month trading period, opening and closing equal long-short trades accordingly.

They iterate each method monthly, so each always involves six overlapping portfolios. They assume round trip broker fees start at 0.7% in 1962 and gradually decline to 0.09% in recent years. They estimate impact of trading on price as 0.3% during 1962-1988 and 0.2% since. They assume zero cost of shorting. They calculate returns based on both employed capital (funding only actual trades) and committed capital (funding 20 concurrent positions per portfolio, with no return on cash). Monthly return for each method is the equally weighted average for the six overlapping portfolios. Using daily dividend-adjusted prices for a broad sample of relatively liquid U.S. common stocks during 1962 through 2014, they find that: Keep Reading

What is the best scheme over the long run for identifying U.S. stock market trends? In the May 2015 version of his paper entitled “Market Timing With a Robust Moving Average”, Valeriy Zakamulin isolates the most robust moving average weighting scheme for a U.S. stock market index based on monthly data. He tests 300 weighting schemes. For all schemes, test portfolios are in stocks (a risk-free asset) when the last index price is above (below) the moving average. His principal performance metric is the Sharpe ratio. He defines robust as: (1) being insensitive to outliers; and, (2) generating consistent performance across all observed market environments. He specifies the range of observed market environments as 30 subperiods, each 10 years in length (with 5-year overlaps). He assumes that there is no optimal trend measurement look-back interval and therefore considers 15 intervals (4 to 18 months). He therefore generates 450 ranks by Sharpe ratio for each of the 300 weighting schemes and defines the most robust as the one with the highest median rank. Using monthly estimates of the Standard and Poor’s Composite Total Return Index and the risk-free rate during January 1860 through December 2014, he finds that: Keep Reading

What is the best way to do asset price trend analysis? Two recent papers address this question. In the May 2015 version of their paper entitled “Which Trend is Your Friend?”, Ari Levine and Lasse Pedersen compare time series (intrinsic or absolute) momentum, moving average (fast and slow) crossovers and other trend indicators to determine the best way to identify a price trend. In the May 2015 version of their paper entitled “Uncovering Trend Rules”, Paul Beekhuizen and Winfried Hallerbach describe how to determine the underlying historical weighting schemes (a combination of continuation and reversion) of price moving averages and combinations of price moving averages. Using both theoretical analyses and examples, these papers conclude that: Keep Reading

Which moving average rules and measurement (lookback) intervals work best? In the March 2015 version of his paper entitled “Market Timing with Moving Averages: Anatomy and Performance of Trading Rules” Valeriy Zakamulin compares market timing rules based on different kinds of moving averages, including simple momentum. He first compares the mathematics of these rules to identify similarities and differences. He then conducts very long run out-of-sample tests of a few trading rules with distinct weighting schemes to measure their market timing effectiveness. He tries both an expanding window (inception-to-date) and rolling windows to discover optimal lookback intervals. He uses Sharpe ratio as his principal performance metric. He estimates one-way trading friction as a constant 0.25%. Using monthly returns for the S&P Composite Index and for the risk-free asset during January 1860 through December 2009, he finds that: Keep Reading

Is stock pairs trading particularly successful under predictable conditions? In their December 2014 paper entitled “On the Determinants of Pairs Trading Profitability”, Heiko Jacobs and Martin Weber present a large-scale analysis of pairs trading, evaluating the effects on profitability of the type of news driving pair divergence, the level of available investor attention and obstacle to exploitation (limits of arbitrage). Their pairs trading approach (see the first chart below as an example) employs daily stock price data to:

1. Calculate each month normalized total return trajectories of stocks over the past 12 months.
2. Measure differences in trajectories for all possible stock pairs.
3. Select the 100 pairs with minimum differences and re-normalize their prices.
4. Whenever over the next six months a pair diverges by more than two standard deviations (per the above 12-month interval), buy the underpriced stock and sell the overpriced stock after a one-day delay.
5. Close the positions upon price convergence within the next month with a one-day delay. If prices do not converge, close the positions after one month. A pair may trade several times during the six-month trading period.

Using stock return data from 34 countries during 2000 through 2013 (excluding small and illiquid firms) and a sample of U.S. stocks with greater than median capitalizations during 1962 through 2008 with contemporaneous news, investor attention and cost of trading proxies, they find that: Keep Reading

Is the difference between upside and downside asset participation ratios relative to a benchmark a useful metric for evaluating asset investment performance? In his June 2014 paper entitled “On the Holy Grail of ‘Upside Participation and Downside Protection'”, Edward Qian defines and investigates the performance implications of the Participation Ratio Difference (PRD) as a measure of combined upside participation and downside protection. He defines the upside (downside) participation ratio of an asset as the ratio of expected excess return for the asset to the expected excess return of its benchmark when benchmark returns are positive (negative). “Excess” means in excess of the return on cash (such that cash has zero participation rates). He defines PRD as the simple difference between positive participation ratio (P+) and negative participation ratio (P-). He then investigates the relationship between asset PRDs and one-factor (market) alphas. He then checks PRDs for the S&P 500 sectors (with the S&P 500 Index as the benchmark) and PRDs for Russell style indexes (with the Russell 3000 Index as the benchmark). Using monthly returns of the S&P 500 index and its ten sectors during October 1989 through April 2014 and monthly returns of Russell broad and value-growth style indexes during January 1979 through April 2014, he finds that: Keep Reading