What is Mean Reversion?

Mean reversion is an investment strategy based on the notion that all prices and returns will return to their mean averages. The average can be based on historical averages, current economic conditions and growth or an industry’s average return. Those who follow mean reversion believe they should hold out if a stock is slipping in value because its price will move back to average when the market takes a turn up.

Mean reversion trading involves selling stocks and securities that behave differently than their historical averages. On the other hand, investors will purchase stocks that show a trend compatible with history. Investors who see unexplainable differences in returns cannot automatically assume that the stocks are doomed to never reach average returns. It could simply be a sign that a company is not doing well and may be ready to fold.

A mean reversion model is based on percent returns and prices, as well as interest rates and price-earnings ratio. All of these things can be very difficult to predict and may be affected by many factors that might prevent the returns from ever hitting average. Interest rates, implied volatilities and stock market returns are more likely to hit mean reversion than exchange rates and stock prices. Reversion to mean is a stochastic process, meaning a random but continuous process in a time series. The process will go on as long as the investor leaves securities in the stock market, and it can also be documented by a model.

Any quantity can be charted with a mean reverting process. These charts are often referenced later, when stock holders are deciding how to play the market. Holding out for averages involves looking at past behavior of stocks, which is where the graphs come in. While looking at stocks’ behavior over time can definitely be inconclusive and even misleading, empirical observations and theoretical calculations can indicate whether reversion to mean is likely or not.

An example that might make mean reversion easier to understand is flipping a coin. Suppose somebody flips a coin multiple times, charting whether it lands on heads or tails. If the coin is only flipped five times, it could land on either heads or tails four or five of those times, and that would not be strange. Now suppose the person flips it fifty times; the odds are fewer that it will land on either heads or tails for the majority of those fifty flips. When the chart is complete, it will probably reflect a more or less equal representation of heads and tails landings.