How well do trend-following rules work when applied to the S&P 500 Index? In the March 2012 version of their paper entitled “Breaking into the Blackbox: Trend Following, Stop Losses, and the Frequency of Trading: The Case of the S&P 500”, Steve Thomas, James Seaton, Andrew Clare and Peter Smith evaluate a variety of simple daily moving average (SMA, 10 to 450 days), moving average crossover (25/50 to 150/350 days) and channel breakout (10-day to 450-day highs) trading rules as applied to the S&P 500 Index. They further investigate: (1) how measurement frequency affects rule performance; (3) effectiveness of combining the rules with stop-losses; and, (3) whether fundamental valuation metrics outperform the rules. They assume an index-cash switching cost of 0.2%. Using daily S&P 500 Index levels and monthly total returns from January 1952 through June 2011, daily S&P 500 Index total returns from July 1988 through June 2011 and contemporaneous Treasury bill yields as the return on cash, they find that:
- Except for very short-term rules, trend following based on either daily or end-of-month calculations generally outperforms buy-and-hold by a wide margin. Specifically, during July 1988 through June 2011:
- A buy-and-hold approach produces a Sharpe ratio of 0.31 and an annualized total return of 9.5%.
- For daily (end-of-month) measurements applied to SMAs, the 400-day (450-day) rule generates the highest net Sharpe ratio of 0.54 (0.59), with annualized net return 10.5% (11.2%).
- For daily (end-of-month) measurements applied to moving average crossovers, the 150/300 (100/250) rule generates the highest net Sharpe ratio of 0.56 (0.58), with annualized net return 10.9% (11.1%). There is little discrimination among rules longer than 50/200.
- For daily (end-of-month) measurements applied to channel breakouts, the 250-day (250-day) rule generates the highest net Sharpe ratio of 0.59 (0.62), with annualized net return 11.2% (11.6%).
- In general, rules based on monthly calculations outperform those based on daily calculations due to reduced trading. For example:
- For the July 1988 through June 2011 sample, net Sharpe ratios for SMA rules range from 0.06 to 0.59 (-0.79 to 0.54) for monthly (daily) calculations.
- For the January 1952 through June 2011 sample, the best end-of-month calculation rule (12 months) is at least as good as the best daily calculation rule (Sharpe ratio 0.58 versus 0.57).
- Popular stop-loss rules do not add value to trend following. In other words, a change of trend itself is the most effective stop-loss.
- Whipsawing is not a problem for longer measurement intervals.
- The simplest trend following rules are about as good as the most complex rules.
- During January 1952 through June 2011, a 10-month SMA rule outperforms fundamental valuation trading rules based on dividend yield, earnings yield, the Fed Model, the relative yield on bonds and equities and Shiller’s cyclically adjusted price-earnings ratio.
In summary, evidence indicates that long-interval trend-following rules outperform both buy-and-hold and commonly used valuation metrics when used to time the S&P 500 Index , with monthly calculations superior to daily calculations and additional stop-loss rules of no value.
Cautions regarding findings include:
- The July 1988 through June 2011 sample is short in terms of number of independent measurement intervals (for example, just 13 450-day intervals) and number of bull-bear stock market cycles.
- Using an index to assess trading rules ignores the costs (trading frictions and management fees) of creating and maintaining a tradable asset that tracks the index.
- The cost of switching between cash and an S&P 500 Index proxy would vary considerably over the sample period since 1952, and would likely have been much higher than 0.2% during some subperiods (see “Trading Frictions Over the Long Run”). Higher trading frictions generally favor monthly over daily calculations and longer over shorter measurement intervals (and buy-and-hold generally).
- Testing a large number of rules on the same data introduces data snooping bias (luck), such that the results for the best rules overstate expectations for future returns.
- The study does not address tax implications of trading.
- Findings for other asset classes may be different (see “SMA Signal Effectiveness Across ETFs” and “Use ‘Standard’ SMAs to Identify Gold Market Regimes?”).
See also “10-month Versus 40-week Versus 200-day SMA”, “Is There a Best SMA Calculation Interval for Long-term Crossing Signals?”, “Simple Tests of an Asymmetric SMA Strategy” and “Pure Versus Buffered SMA Crossing Signals”.
For counterpoint, see “Technical Trend-following: Fighting the Last War?”.
See also “Do Stop Losses Work?” and “Using Trailing Stop Losses to Reduce Risk”.
