Do moving average rules work for timing stocks over the long run? In his January 2013 paper entitled “The Rise and Fall of Technical Trading Rule Success”, Nicholas Taylor examines the performance of moving average trading rules as applied to components of the Dow Jones Industrial Average (DJIA) over the long run. He considers 10,800 variants of a general moving average trading rule: buy (sell) when the short-interval moving average price crosses above (below) the long-interval moving average price, with moving average measurement intervals ranging from 1 to 250 trading days. Rule variants include signal refinements that specify: a range of the ratio of short-interval to long-interval moving average prices; the number of days a signal must persist before taking action; and, the number of days for ignoring all new signals after executing a trade. He defines the return for a specific rule as the equally weighted average for applying it to all DJIA stocks. He tests both static rules and dynamically optimal sets of rules, with the latter comprised of the best rule each month from four distinct ways of measuring lagged net performance. He estimates trading frictions based on bid-ask spreads. He compares monthly performance of moving average rules to a monthly buy-and-hold benchmark based on raw return statistics and on alphas from factor (market, size and book-to-market, momentum) models of stock returns. Using daily prices of the 30 then-current DJIA stocks during October 1928 through December 2011 (82 stocks over the sample period), he finds that:
- Over the full sample period, with shorting allowed (no stock borrowing costs), the best static (in-sample) moving average rule and the monthly optimal dynamic rules (out-of-sample) generate higher average net returns with lower volatilities than the buy-and-hold benchmark. The dynamic rule with exponential smoothing of lagged performance beats both other dynamic rules and the best static rule. However:
- While the best moving average rules beat buy-and-hold for a 1934-1986 subsample, they do not for a 1987-2011 subsample.
- Much of the profitability of the best moving average rules comes from shorting (no borrowing costs).
- Similarly, the alphas of moving average rules with shorting (no borrowing costs) are significantly positive over the full sample period. However:
- Significantly positive alphas concentrate during the mid-1960s to the mid-1980s, peaking during the early 1970s, and it is plausible that shorting costs exceed alphas during this time.
- Long-only alphas are close to zero and do not vary over time.
- A possible interpretation is that moving average rules work during some periods, with their growing popularity subsequently extinguishing their effectiveness.
In summary, the effectiveness of stock trading rules based on moving averages varies over time, peaking in the early 1970s and disappearing after the mid-1980s, and further depends on short selling at low cost.
Cautions regarding findings include:
- The method of estimating trading frictions may be optimistic for some subperiods and some investors.
- The study ignores dividends paid by the specified stocks, to the advantage of returns from shorting. Short positions would have to pay the dividends (in addition to any stock borrowing costs)
- As noted in the paper, lack of information about the feasibility and costs of shorting over the sample period confounds estimation of net profitability of technical trading with shorting.
- The best static (in-sample) moving average trading rule incorporates data snooping bias. A trader could not have known that this rule is best while trading within the sample period.
- Technical trading may work better on less liquid stocks (see “Does Technical Trading Work for Certain Kinds of Stocks?” and “Technical Trading Thoroughly Tested”).
See also “True Out-of-Sample Test of “Best” Technical Trading Rules” and “Technical Analysis Tested on Long-run DJIA Data” for closely related studies.
