Mean Reversion and Value Investing by Sui Chuan, Value Edge
We expound on mean reversion and value investing – its concept, issues and application to investing.
What is mean reversion?
“Regression to the mean” refers to an inverse correlation among roughly normally distributed observations that are made repeatedly over time. An extreme observation at one point in time (“outlier”) tends to be followed by a less extreme observation; extremes, in other words, revert or regress over time towards mean or average measurements.
Mean reversion in financial markets
Investors such as David Dreman and behavioural economists such as Richard Thaler and Werner De Bondt have uncovered strong evidence that regression to the mean, or something akin to it, occurs on financial markets. It occurs at both individual and aggregate levels, i.e., with respect to both individual securities and markets as a whole. Psychology plays a part in mean reversion. Investors have a predilection to linearly extrapolate the future from the past; yet, if the winners kept on winning and the losers kept on losing, then the economic and financial landscape would consist of a shrinking handful of companies with colossal market capitalization and virtually no small enterprises. This is also why value stocks tend to outperform growth stocks – an abnormally high performance is likely to be followed by a poorer performance and vice versa. Investors also tend to overreact to new information and reversal trends tend to develop after some delay.
Value investing and normalized earnings
In Security Analysis, Benjamin Graham alludes to the concept of mean reversion.
“Investment values can be related only to demonstrated performance so that neither unexpected increases nor even past results under conditions of abnormal business activity may be taken as a basis“
Graham advocated the use of normalized earnings covering a period between five to 10 years, over a single-year earnings ratio to adjust for the impact of the business cycle. Keith Anderson and Chris Brooks, from University of Reading and City University respectively, found evidence that substituting a long-term average of earnings in place of single-year earnings increased the spread in returns between value and glamour stocks by 6 percent per year.
Are the teachings of Graham congruent with the concept of mean reversion? While we recognise that there are some similarities, we believe there are vital differences which tend to be forgotten when investors cavalierly take mean reversion as a given.
It is a group phenomenon
Mean regression is a statistical, group phenomenon and the luxury of large numbers is not something that every portfolio possesses. Even though the outlier group’s average will move toward the population’s, you cannot tell which way an individual’s score will move based on the regression to the mean phenomenon. For those that do, sometimes regression proceeds at such a slow pace that an exogenous “shock” will disrupt or reverse it. Everything would seem exogenous to an ignorant investor; we assert mean reversion cannot be the sole validation of undervaluation.
Time period conundrum
Normalized earnings are often used as a benchmark for mean regression. From a pure statistical standpoint, a larger sample size (more historical years) is always preferred. Unfortunately, investing is a business decision and we know that that is not necessarily true – the performance of a company 10 years ago might not be an accurate benchmark of tomorrow’s normality. This represents the trade-off between using a longer and shorter time period when normalizing earnings. From a business perspective, the time period of normalization has absolutely no implication on the accuracy of the mean. Consider the following figures:
|Scenario A||Scenario B|
|3-year Average EPS||4||6|
|5-year Average EPS||6||4|
Assume that a stock is fairly valued based on current EPS, without any assessment of business conditions, one cannot draw a safe conclusion regarding the valuation of a company under either scenarios.
Having known the issues with mean reversion, our position is that a statistical quirk like mean reversion should never be the main thesis for buying into a stock. It should be backed up by competent understanding of the business and its history. The more knowledge you have, the more reliable your judgement of normalized earnings.
“Quantitative data are useful only to the extent that they are supported by a qualitative survey of the enterprise“
We note that Graham never referred to mean reversion as a basis of investing; instead, he spoke of normality and abnormality in earnings, implying it as concept of business rather than of statistics. Mean reversion can serve as a proxy from time to time and while we don’t suggest that investors start building complex models to forecast earnings, investors have to thread a fine line between using historical performance and obsessing over future earnings. In using mean reversion, always err on the side of caution and never forget the concept of margin of safety. How much does the business need to deteriorate from its current conditions before its current price becomes a fair value?