Momentum Regimes and Mean Reversion: A Framework

In 1993, Narasimhan Jegadeesh and Sheridan Titman published a paper that reshaped how we think about stock returns. They showed that buying stocks with strong recent performance and selling stocks with weak recent performance generated roughly 12% in annual excess returns over a 3-to-12 month horizon (Jegadeesh and Titman, 1993). The finding held across time periods and geographies. At the time, no risk-based explanation fully accounted for the premium.

The momentum effect has since survived three decades of out-of-sample testing. Asness, Moskowitz, and Pedersen (2013) documented its persistence across eight asset classes and dozens of countries. Fama and French, whose three-factor model initially excluded momentum, eventually acknowledged its empirical strength.

The return profile, however, is choppier than that suggested by the long-run averages.

Daniel and Moskowitz (2016) documented that momentum strategies experience fewer but severe crashes. These tend to occur during sharp market reversals, when last period's underperformers suddenly outperform last period's outperformers by a wide margin. The most extreme episode was observed in spring 2009. As markets rebounded from the financial crisis lows, the momentum factor suffered a drawdown exceeding 70% in a matter of weeks. many systematic investors were unprepared for such a rapid shift in market risk factors.

These crashes remain interesting because of their timing. They tend to cluster at inflection points, when the market shifts from one regime to another. During the late stages of a bear market, the number of stocks with negative momentum increases substantially. As soon as the recovery begins, those stocks snap back. Momentum, by construction, is short exactly those names. The factor itself doesn't necessarily underperform during a crash because it is behaving exactly as its construction dictates. The investor who did not anticipate the regime shift, and therefore did not adjust the portfolio, is the one caught offguard.

There are signals that may help. For example, return dispersion, the cross-sectional spread between the best and worst performing stocks, tends to widen during momentum-friendly periods and tightens before reversals (Stivers and Sun, 2010). Realized volatility diverging from the level implied by options markets has, in some studies, preceded momentum breakdowns. Market breadth offers another statistic. When a rally is broad, momentum tends to persist. When it is narrow, limiting to a handful of names, the conditions for a mean-reversion appear more frequently. Majority of these indicators are individually insufficient and even unstable across sample periods to serve as standalone market timing indicator. Together too, they may tilt the odds in favor of the investors, but they still do not resolve the underlying challenge and therefore are largely unreliable.

The academic evidence on momentum is widely accepted by both academics and practitioners. The factor is present, it is persistent, and it appears in most markets that have been studied. Research also indicates that the return stream is lumpy and characterized by fatter left tails. Investors who have captured the long-run momentum premium are those who reduced their exposure during unfavorable regime conditions rather than those who held constant exposure through transitions. Whether this is skill, or luck depends on the sample period to a large extent.