Evidence-based investing research
Value Investing Strategy (Strategy Overview)
Allocations for November 2026 (Final)
Cash TLT LQD SPY
Momentum Investing Strategy (Strategy Overview)
Allocations for November 2026 (Final)
1st ETF 2nd ETF 3rd ETF

Predictable Long-run Stock Market Returns?

Steve LeCompte | | Posted in: Fundamental Valuation, Technical Trading

Key Insight

S&P Composite Index exhibits mean reversion over 15-year horizons, with past 15-year returns explaining 42% of future return variation; mean reversion forecasting performs comparably to P/E10 models (1871-2011; descriptive analysis).

Are there exploitable long-term cycles in U.S. stock market returns? In the January 2013 update of his paper entitled “Secular Mean Reversion and Long-Run Predictability of the Stock Market”, Valeriy Zakamulin explores mean reversion of the S&P Composite Index over intervals ranging from two to 40 years. He then runs an out-of-sample horse race using inception-to-date data to compare three regression-based models for forecasting long-term stock market returns: (1) mean reversion over the dynamically optimal horizon; (2) the random walk (future mean return equals (evolving) historical mean return); and, (3) valuation based on Robert Shiller’s cyclically adjusted price-to-earnings ratio (P/E10). Using real (Consumer Price Index-adjusted) S&P Composite Index total annual returns and earnings over the period 1871 through 2011 (141 years), he finds that:

  • Based on in-sample autocorrelation analyses over the entire sample period and a 1926-2011 subperiod, the S&P Composite Index exhibits significant mean reversion with a full cycle of 30-34 years (peak-to-trough and trough-to-peak reversion of 15-17 years).
  • Over the entire sample period, S&P Composite Index return over the past 15 years explains 42% (R-squared statistic 0.42) of the variation in return over the next 15 years (see the first chart below).
  • Out-of-sample tests commencing in 1921 indicate that:
    • A stock market model based on mean reversion at the lagged optimal horizon outperforms a model based on the historical mean and performs comparably to a model based on P/E10 valuation (see the second chart below).
    • Mean reversion and P/E10 stock market forecasting models perform better during a recent 1959-1997 subperiod than for the overall 1921-1997 out-of-sample test period.

The following chart, taken from the paper, relates future 15-year real return for the S&P Composite Index to its past 15-year real return during 1871-2011 with overlapping intervals. The negative slope indicates mean reversion, with a high (low) past return relating to a low (high) future return. Past returns explain 42% of the variation in future returns. This analysis is in-sample; an investor operating in real time during the sample period would not have all these data. Also, the use of overlapping intervals overstates actual sample size (there are only 9.4 completely independent 15-year intervals in the sample).

15-year-stock-market-reversion

The next chart, also from the paper, compares actual returns to forecasted returns for the three-model horse race specified above. Based on correlations between forecasted and actual returns and mean squared prediction errors, mean reversion and P/E10 (Robert Shiller) models perform comparably, both significantly more accurate than the historical mean (random walk) model. 

reversion-mean-PE10-models-horse-race

In summary, evidence indicates significant mean reversion of U.S. stock market returns at a horizon (half-cycle) of about 15 years, with the forecasting power of mean reversion similar to the of P/E10.

Cautions regarding findings include:

  • The study does not explicitly address the economic value of exploiting mean reversion in U.S. stocks, whether from optimizing the long-term net return from U.S. stocks or from diversifying with asset classes that follow an uncorrelated long-term cycle. In other words, it does not test any trading systems.
  • Sample size is not large for this kind of analysis.

See also “Classic Research: Demography and the Stock Market”, “Return-based Analysis of Demographics as Stock Market Predictor” and “Demographic Headwind for U.S. Stock Market?” for potential explanations of a long-term stock market cycle.

“P/E10 and Future Stock Market Returns”, “Predictive Power of P/E10 Worldwide” and “Future Stock Market Returns and P/E10” are also relevant.

Frequently Asked Questions

What is mean reversion in stock market returns?

Mean reversion in stock market returns is the tendency for periods of high returns to be followed by low returns, and low returns by high returns, over a given horizon. In Valeriy Zakamulin's study of the S&P Composite Index (1871-2011), mean reversion appears as a negative relationship between past and future returns, with a full cycle of 30-34 years and half-cycles of 15-17 years.

Does past 15-year stock market performance predict future 15-year returns?

Past 15-year returns of the S&P Composite Index explain 42% of the variation in the next 15-year return (R-squared 0.42) over 1871-2011, with high past returns tending to precede low future returns. This result is in-sample and uses overlapping intervals, so it overstates what an investor could have known in real time. Out-of-sample tests starting in 1921 show a mean reversion model beats a historical-mean model and performs comparably to a P/E10 model.

How is long-term stock market mean reversion tested against P/E10 and a random walk?

Zakamulin's test uses real (CPI-adjusted) annual total returns and earnings for the S&P Composite Index from 1871 to 2011, first measuring in-sample autocorrelation over horizons of two to 40 years. He then runs an out-of-sample horse race using inception-to-date data, starting in 1921, comparing three regression models: mean reversion at the dynamically optimal horizon, a random walk (evolving historical mean return), and Shiller's P/E10 valuation. Models are judged by correlations between forecasted and actual returns and by mean squared prediction errors.

How does a mean reversion forecasting model compare to the Shiller P/E10 model for predicting stock returns?

A mean reversion model at the lagged optimal horizon performs comparably to a Shiller P/E10 valuation model in forecasting long-term S&P Composite Index returns, and both are significantly more accurate than the random walk (historical mean) model. Both mean reversion and P/E10 models performed better in the 1959-1997 subperiod than over the full 1921-1997 out-of-sample period.

What are the limitations of using long-term mean reversion to forecast stock market returns?

The main limitations of the 15-year mean reversion finding are that the 42% explanatory power is in-sample, and overlapping intervals overstate the sample size, since 1871-2011 contains only about 9.4 completely independent 15-year intervals. The study also does not test trading systems or the economic value of exploiting mean reversion, such as optimizing long-term net returns or diversifying with asset classes on uncorrelated cycles. The sample is not large for this type of analysis.

Guru Grades

1,206 predictions by 42 market gurus since 2022: 41% right. Always saying stocks will go higher: 62%.

As of the Nov 18, 2026 close

  1. 1 Ed Yardeni A 71%
  2. Always bullish B 62%
  3. 2 Jim Paulsen B 61%
  4. 3 Sam Stovall B 58%
  5. 4 Adam Turnquist B 58%
  6. 5 Ryan Detrick B 56%

Full leaderboard ›