Technical AnalysisMarket Efficiency

Do Moving Averages Predict Stock Returns? 90 Years of Dow Jones Evidence Says Yes

Summary by Robert Gorak · Published June 18, 2026 · Last reviewed June 18, 2026

William Brock and Josef Lakonishok and Blake LeBaron·1992·Journal of Finance
Sample: 25,806 daily observationsData: Dow Jones Industrial Average (DJIA)Period: 1897-1986

Technical trading rules use patterns in past prices, such as moving averages and trading-range breakouts, to generate buy and sell signals for predicting future returns. Brock, Lakonishok, and LeBaron's (1992) "Simple Technical Trading Rules and the Stochastic Properties of Stock Returns" tested these rules on Dow Jones Industrial Average data. The full sample contained 25,806 daily observations from 1897 to 1986, and buy signals earned an average daily return of 0.042 percent. Sell signals earned -0.025 percent daily, a pattern not consistent with the random walk, AR(1), GARCH-M, or Exponential GARCH null models.

What the Study Found

Buy signals from the variable-length moving-average rule produced an average daily return of 0.042 percent, about 12 percent annualized, over the full 1897-1986 sample. Sell signals produced an average daily return of -0.025 percent, about -7 percent annualized. Daily standard deviation was 1.34 percent during sell periods versus 0.89 percent during buy periods. Across all 10 variable-length moving-average rules tested, the average buy-sell spread on the full sample was 0.00067. The fixed-length moving-average rule produced a 10-day buy-sell spread of 0.77 percent with no band and 1.09 percent with a one percent band.

Methodology

The study uses daily closing prices of the Dow Jones Industrial Average, a continuous series available since 1897. The full sample contains 25,806 daily observations from the first trading day of 1897 to the last trading day of 1986. Returns following buy and sell signals from variable-length moving average, fixed-length moving average, and trading range break-out rules were tested separately across four nonoverlapping subperiods. Statistical significance was assessed using bootstrap simulations with 500 replications under four null models: the random walk, AR(1), GARCH-M, and Exponential GARCH.

Key Statistics

Metric Finding Context
VMA buy vs. sell daily return, full sample 0.042% vs. -0.025% 1897-1986, variable-length moving average rule
VMA buy vs. sell daily standard deviation 0.89% vs. 1.34% 1897-1986, variable-length moving average rule
Average VMA buy-sell spread (10 rules) 0.00067 1897-1986 full sample, Table II
FMA buy-sell spread, no band vs. 1% band 0.77% vs. 1.09% 10-day holding return, full sample 1897-1986
TRB average buy-sell return 0.86% 10-day holding return, full sample 1897-1986, six rules
AR(1) null model r_t = b + ρr_{t-1} + ε_t Used in bootstrap simulations against trading rule returns
GARCH-M null model r_t = a + γh_t + bε_{t-1} + ε_t Used in bootstrap simulations against trading rule returns

Why This Matters

The persistent gap between returns following buy signals and sell signals challenges the view that past prices carry no information about future returns. Because the pattern held up against four different statistical models of return behavior, it is not easily dismissed as an artifact of one model choice. For risk-based theories of asset pricing, a profitable rule that requires no additional measurable risk is difficult to reconcile with standard models. For practitioners, the results offered early rigorous statistical evidence that simple, widely used technical rules captured a real pattern in historical price data.

Frequently Asked Questions

12 percent was the approximate annualized return during buy-signal periods of the variable-length moving average rule. The average daily return was 0.042 percent for buy signals and -0.025 percent for sell signals, about -7 percent annualized. Brock, Lakonishok, and LeBaron (1992) found this using Dow Jones Industrial Average data from 1897 to 1986.

0.00053 was the buy-minus-sell daily return spread for the (1,150,0) variable-length moving average rule on the 1897-1986 Dow Jones sample, with a t-statistic of 3.78784. Brock, Lakonishok, and LeBaron (1992) rejected the random walk, AR(1), GARCH-M, and Exponential GARCH null models using 500-replication bootstrap simulations.

0.86 percent was the average 10-day buy-sell return for the trading range break-out rule across six rule variations. Brock, Lakonishok, and LeBaron (1992) tested this on Dow Jones Industrial Average data from 1897 to 1986. Buy-period returns averaged 0.55 percent over the same 10-day holding window, exceeding the unconditional 10-day return of 0.17 percent.

10 moving average rules were tested on Dow Jones data from 1897 to 1986 by Brock, Lakonishok, and LeBaron (1992). Buy signals earned an average daily return of 0.042 percent, compared with -0.025 percent for sell signals. The rule signals buy when a short-period price average rises above a long-period average, and sell when it falls below.

Source

William Brock and Josef Lakonishok and Blake LeBaron (1992). Simple Technical Trading Rules and the Stochastic Properties of Stock Returns. Journal of Finance.

Read the full paper