Many investment strategies employ techniques that identify potentially attractive trades on the assumption that prices revert to a central trend, generally referred to as a ‘mean,’ although it may not actually have that mathematical pedigree. The claim for these techniques is that prices can be expected to pull back from the extremes to which supply or demand may push them, and that isolating a sufficiently stable central tendency allows traders to identify and exploit such reversions. For at least a year now, this assumption has been under strain in various markets.
Before proceeding further, however, there is a red herring to dispose of. In an August 2 Financial Times editorial, Richard Clarida and Mohamed El-Erian of Pimco muddy the waters by arguing from the extreme dispersion of current economic opinion – deflation vs. hyperinflation, recovery vs. recession – to the conclusion that investors will flee investment techniques that rely on some historical measure of value to which prices revert. Their reasoning is unsound. From a wide disparity of economic expectations it follows that there is a wide disparity of opinion on where value is to be found – but it does not follow that value itself is widely dispersed. Value is a matter of economic outcomes, not economic perceptions. Some forecasts will turn out to be correct, leading their adherents to value and attractive returns, while others will turn out to be wrong, suggesting value where in fact there is none (the bond market, perhaps?). Confusion over the economic outlook may cause investors to despair over their ability to identify sources of value, but it does not follow that there is no value to be found.
The issue regarding mean reversion does not relate to valuation but to technical analysis and algorithmic trading, particularly over the short time horizons employed by high frequency traders and statistical arbitrageurs. By definition, any procedure used to derive a central price tendency – whether it is a moving average, the arithmetic mean relative to which standard deviation is calculated, or what-have-you – must take time series data for its inputs. The longer the time series – the greater the number of observations used in the function or algorithm used to define the tendency – the less volatile the central value will be. That is, the longer the series, the less the next observation will cause the trend to shift, even if it diverges very sharply from the tendency. The effect of this “dilution” of recent observations can be compensated by various methods of weighting them more heavily than earlier ones, but it cannot be completely eliminated.
Since they are unavoidably lagging indicators, in periods of high volatility these central tendencies may not be of much value to traders. This is especially true when markets are volatile over several time horizons – day to day, week to week and month to month – and when their volatility is itself volatile, increasing and decreasing rapidly. In the charts below, standard deviation is calculated over ninety observations, which is about the shortest series that allows for statistical significance. It does not take much imagination to see why it would be difficult for mean reversion techniques to thrive in the U.S. equity market since April of this year. For much of the period those who rely on mean reversion were subject to frequent false signals or no meaningful signals at all.
All investment managers experience periods during which their discipline is not rewarded by their market’s characteristics. In some cases these periods can be quite protracted, inevitably causing them much soul-searching and not a little despondency. Managers without access to alternative trade identification procedures to fall back on may even be forced to close shop before conditions become more favorable to them. But conditions will eventually change, and if in the meantime the ranks of traders reliant on mean reversion trading signals have been thinned, the reward to mean reversion techniques will be that much greater.