Few propositions in economics are held with more fervour than the view that financial markets are “efficient”, or that future price changes are unpredictable. Another strongly held view is the random walk hypothesis, which state that stock prices evolve according to a random walk and thus cannot be predicted. If either of these hypothesis hold, then we can’t expect to consistently generate a return above the market average and may as well buy and hold a market index such as the JSE Top 40 ETF.
We obviously do not subscribe to either of those theories. In last week’s blog, I provided strong evidence that daily returns of the JSE Top 40 index do not follow a random bull curve distribution. In fact, daily returns follow a leptokurtic distribution, or a distribution that has a higher propensity to mean revert and display powerful trends than a random series would have us believe. Mean reversion exhibits performance characteristics that a very desirable to traders: high win rates, low drawdown’s, smooth equity curves and swift recoveries from drawdown. It’s these statistics that was the impetus for my research within this approach. But before I went all-in, I wanted to further validate the market’s tendency to mean revert.
The method I employed focused on correlation, in particular autocorrelation or serial correlation of the JSE Top 40 index daily returns. If the market moves in a purely random motion, we would expect the return today to have no bearing on the return tomorrow, or in statistical terms, we expect to see zero correlation between the returns today and tomorrow. However, what I found was strong negative serial correlation, in other words extreme price moves tend to be follow by moves in the opposite direction, or evidence of mean reversion.
I tested for serial correlation in the 5 day returns of the index. Basically, I tested for correlation between an extreme price move over the previous 5 day period (-3%, -5%, -7%) and the 5 day return immediately following that. If the market was truly random we would expect the correlation to be zero, BUT, what I found was strong negative serial correlation. Furthermore, the correlation is very robust: more extreme moves are associated with stronger negative correlations. In layman terms, the larger the 5 day sell-off, the higher the expected positive return in the following 5 days. Historically, when the market has fallen in excess of -7% we’ve seen an average return of 2.36% over the following 5 days!
Significant negative serial correlation, as we see in this study, implies that the market does indeed have a tendency to revert to the mean, and therefore appears to be somewhat predictable. We’ve developed proprietary mean reversion algorithms that have proven to be tremendously successful in historical simulations. Clients can gain access to our research and much more in QuantLab.
Next week I’ll discuss some behavioural finance theories that explains the possible cause of mean reversion in the financial markets.