Long-term reversal is the tendency for stocks that performed poorly over the past 3-5 years to outperform in subsequent periods, and vice versa. It operates on a longer horizon than (and in the opposite direction to) momentum, which works over 3-12 months.
Discovery
DeBondt and Thaler (1985) provided the foundational evidence in one of the first papers in behavioral finance:
- Data: CRSP monthly returns, all NYSE common stocks, January 1926 to December 1982
- Method: rank stocks by cumulative market-adjusted returns over a 3-year formation period, form portfolios of extreme winners (top 35) and losers (bottom 35), track returns for the subsequent 36 months. Repeated across 16 non-overlapping formation periods.
- Key result: loser portfolios outperform winners by 24.6% in cumulative abnormal returns over 36 months (t=2.20). Losers gain ~19.6% over the market; winners underperform by ~5.0%.
- 5-year horizon: with 5-year formation and test periods, the difference reaches 31.9% (t=3.28).
Winner-loser asymmetry
The reversal effect is asymmetric: losers outperform far more than winners underperform. Additionally, losers have lower CAPM betas (1.026) than winners (1.369), meaning losers outperform despite being less risky by standard measures. This strengthens the case that the anomaly represents genuine mispricing rather than risk compensation.
The overreaction hypothesis
DeBondt and Thaler ground their prediction in Kahneman and Tversky’s representativeness heuristic: investors overweight recent information and underweight base rates, violating Bayes’ rule. After a long string of bad news, investors become excessively pessimistic about a firm’s prospects, pushing its price too low. The subsequent correction produces positive abnormal returns for losers.
January effect
Most of the loser portfolio’s excess returns concentrate in January: 8.1%, 5.6%, and 4.0% in the Januaries at months 1, 13, and 25 after formation. This overlap with the January effect complicates interpretation but does not fully explain the reversal, as it persists (at reduced magnitude) excluding January.
Relationship to momentum
Long-term reversal and momentum operate at different frequencies and in opposite directions:
| Effect | Horizon | Direction | Explanation |
|---|---|---|---|
| Short-term reversal | < 1 month | Contrarian | Liquidity, bid-ask bounce |
| Momentum | 3-12 months | Continuation | Underreaction to information |
| Long-term reversal | 3-5 years | Contrarian | Overreaction, mean reversion |
Jegadeesh and Titman (1993) documented the transition between these regimes: momentum profits partially reverse over months 13-36, with the 6-month strategy losing more than half its first-year gains over the subsequent 24 months. This coexistence of medium-term momentum and longer-term reversal requires nuanced models of investor behavior.
Connection to other anomalies
DeBondt and Thaler argue that the P/E effect and small-firm effect may be manifestations of overreaction: firms with poor past performance (losers) tend to have low prices, high E/P ratios, and small market caps. In their framing, the small-firm anomaly is really a “losing firm” effect.
Construction in MAC3
In the MAC3 model, long-term reversal uses the same weighted log-return framework as momentum but with a different window:
- Months 1-12: zero weight (excludes the momentum signal)
- Month 13: linearly increasing weights
- Months 14-47: constant weights
- Month 48: linearly declining weights to zero
Introduced in MAC3 (not present in MAC2). IPOs are handled identically to momentum: assigned the ESTU cap-weighted mean until sufficient history accumulates.
In academic factor models
Long-term reversal is not a factor in the standard FF3, Carhart, or FF5 models. Some argue it is subsumed by book-to-market: past losers mechanically have lower market prices and thus higher BE/ME. It is included in MAC3 as a risk factor to improve volatility forecasting.