Are there any seasonal, technical or fundamental strategies that reliably time the U.S. stock market as proxied by the S&P 500 Total Return Index? In the February 2018 version of his paper entitled “Investing In The S&P 500 Index: Can Anything Beat the Buy-And-Hold Strategy?”, Hubert Dichtl compares excess returns (relative to the U.S. Treasury bill [T-bill] yield) and Sharpe ratios for investment strategies that time the S&P 500 Index monthly based on each of:
- 4,096 seasonality strategies.
- 24 technical strategies (10 slow-fast moving average crossover rules; 8 intrinsic [time series or absolute] momentum rules; and, 6 on-balance volume rules).
- 18 fundamental variable strategies based on a rolling 180-month regression, with 1950-1965 used to generate initial predictions.
In all cases, when not in stocks, the strategies hold T-bills as a proxy for cash. His main out-of-sample test period is 1966-2014, with emphasis on a “crisis” subsample of 2000-2014. He includes extended tests on seasonality and some technical strategies using 1931-2014. He assumes constant stock index-cash switching frictions of 0.25%. He addresses data snooping bias from testing multiple strategies on the same sample by applying Hansen’s test for superior predictive ability. Using monthly S&P 500 Index levels/total returns and U.S. Treasury bill yields since 1931 and values of fundamental variables since January 1950, all through December 2014, he finds that:
- During 1966-2014, the average monthly total return (standard deviation of total returns) for the S&P 500 Total Return Index is 0.79% (4.39%), compared to 0.36% (4.44%) during 2000-2014 and 0.80% (5.42%) during 1931-2014. Monthly Sharpe ratios are 0.08 for 1966-2014, 0.05 for 2000-2014 and 0.09 for 1931-2014.
- During the full 1966-2014 out-of-sample test period:
- The best seasonality strategy based on average net monthly excess return (0.84%) and net monthly Sharpe ratio (0.13) is the traditional Halloween strategy, in the stock index (cash) during November through April (May through October). However, outperformance is not statistically significant for any seasonality strategies after correction for data snooping bias.
- The best technical strategies mostly beat the best seasonality and fundamental models. The best strategy based on average net monthly excess return (0.90%) and net monthly Sharpe ratio (0.15) is crossover of 2-month and 12-month moving averages of on-balance volume. Outperformance of this strategy is modestly statistically significant on a risk-adjusted basis after correction for data snooping bias.
- The best fundamental strategy based on average net monthly return (0.85%) and net monthly Sharpe ratio (0.12) is the simple average forecast for 14 individual variables. However, outperformance is not statistically significant for any fundamental models after correction for data snooping bias.
- During 2000-2014 “crisis” subsample:
- The best seasonality strategies differ from the traditional Halloween strategy, instead holding the stock index in March, April, October, November and December and cash in January, February, June, and September. However, outperformance is not statistically significant for any seasonality models after correction for data snooping bias.
- The best technical strategies all beat the best seasonality and fundamental models. The best strategy based on average net monthly return (0.74%) and net monthly Sharpe ratio (0.23) is crossover of 1-month and 9-month price moving averages. Outperformance of this strategy is statistically significant on a risk-adjusted basis after correction for data snooping bias. Significance also holds versus a 50% stock index and 50% cash benchmark portfolio. Price moving average strategies positively affect skewness and dramatically reduce maximum drawdown.
- The best fundamental strategy based on average net monthly return (0.85%) and net monthly Sharpe ratio (0.12) is the multivariate regression forecast involving all 14 individual variables. However, outperformance is not statistically significant for any fundamental models after correction for data snooping bias.
- During the extended 1931-2014 test period for seasonality and those technical strategies not using volume:
- The best seasonality strategy based on average net monthly return (0.86%) and net monthly Sharpe ratio (0.11) holds the stock index during all months except September. However, outperformance is not statistically significant for any seasonality models after correction for data snooping bias.
- The best technical strategy based on average net monthly return (0.81%) and net monthly Sharpe ratio (0.15) is crossover of 2-month and 12-month price moving averages. Outperformance of this strategy is modestly statistically significant on a risk-adjusted basis after correction for data snooping bias.
In summary, evidence supports belief that some U.S. stock market timing strategies that exploit investor underreaction and overreaction via technical indicators reliably beat buy-and-hold and are largely superior to seasonality and fundamental timing approaches.
Cautions regarding findings include:
- The assumed constant level of stock index-cash switching frictions is very low for early parts of the sample period and high for the recent part (see “Trading Frictions Over the Long Run”). Accounting for variation in realistic frictions over time may affect findings.
- As noted in the paper, restricting seasonality strategies to a few published strategies induces statistical significance for some of them during 2000-2014. However, this restriction does not (cannot) account for snooping bias inherited from research involving their discoveries. This caution applies also to the already restrictive selection of technical and fundamental strategies.
- The study does not address the effects of market timing on taxes.
