Contrarian Investing

Scope. This article describes the contrarian position, the evidence for long-horizon reversal, and the two very different explanations that fit it equally well. It recommends nothing, supplies no rule, formation period or holding period, and characterises no security. No price, multiple or valuation appears.

It is not the same thing as value investing, though the two overlap. Value compares a price to an estimate of what a business is worth. Contrarian compares a security's standing in other people's opinion to some notion of what it deserves. A cheap and widely admired company is a value candidate and not a contrarian one; an expensive and detested one is the reverse.

The evidence for reversal

The founding study appeared in 1985. De Bondt and Thaler sorted US stocks by performance over a prior period and tracked what happened afterwards, finding that prior losers subsequently outperformed prior winners by roughly 24.6 percentage points cumulatively over the 36 months after formation, with a t-statistic of 2.20 — losers beating the market by about 19.6% and winners lagging it by about 5.0%.

Three details in the original result deserve as much attention as the headline. The effect was asymmetric, much larger for losers than winners. Most of the excess return arrived in January, which ties the finding to a seasonal pattern rather than to overreaction alone. And the loser portfolios had lower measured betas than the winners — 1.026 against 1.369 — meaning that on a conventional risk model they outperformed while being less risky, which is harder to explain away than a simple return difference.

And the effect has not held up in the way the original suggested. Later work applying comparable filters to a later period found prior losers outperforming prior winners by around 0.02% over three years — a difference indistinguishable from nothing. A more recent replication using data from 2000 onward reported that both winner and loser portfolios earned strong forward three-year returns with a minimal differential between them. Two readings are available and this portal does not choose between them: that the effect was real and has been competed away as it became known, or that the original result reflected the particular period, the January seasonality and the specific construction. Both readings are consistent with everything published, which is an uncomfortable but accurate place to leave it.

The problem with the horizons

Reversal at three to five years and continuation at three to twelve months are both documented, and they point in opposite directions. Momentum says recent winners keep winning; contrarian says extended winners eventually lose.

They are reconcilable only as horizon-specific phenomena — underreaction first, overreaction later, correction eventually — which is exactly the behavioural story their proponents tell. But the reconciliation should be viewed with some suspicion, because a framework in which prices continue over one window and reverse over another can accommodate almost any observation by adjusting which window is being discussed. The horizons were chosen after the data were seen, in both literatures.

Why reversion happens even when nothing is wrong

Selecting the extremes of any noisy measure guarantees subsequent reversion, with no overreaction, no psychology, and no mistake by anyone. This is regression to the mean, and it is worth computing rather than asserting.

Suppose an observed outcome is part durable signal and part noise. Ranking on the observed value and taking the top decile selects securities whose observed extremity is partly real and partly luck — and only the real part persists.

What the approach demands

The position is uncomfortable by construction. A contrarian holds what a great many informed people have decided to avoid, and must do so while continuing to be wrong in the eyes of everyone visible.

And the central difficulty is that the crowd is frequently right. Companies that have fallen a long way have often fallen for reasons, and the reasons sometimes continue. Contrarianism supplies no method for separating a security that is unloved and salvageable from one that is unloved and finished — that separation requires the analysis the other approaches in this pillar are about, and disliking the consensus is not a substitute for it.