Overreaction

Overreaction

Overreaction is the systematic tendency to update beliefs too much in the face of new information — the parametric mirror image of conservatism. In the Rabin (1998) template
$\frac{\pi (A\mid S)}{\pi (B\mid S)}={\left(\frac{p(S\mid A)}{p(S\mid B)}\right)}^c{\left(\frac{p(A)}{p(B)}\right)}^d,$
overreaction is the case $c>1$ — the investor places more than full Bayesian weight on the likelihood ratio of the new signal. Bayes’ rule corresponds to $c=d=1$; the overreacting agent moves her posterior further than she should. Empirical estimates of $c$ in overreaction-prone settings sit in the range $1.2$–$2.0$.

In behavioral asset pricing, overreaction is the leading microfoundation for long-run reversal. After a long string of same-direction news, an investor with $c>1$ has built a posterior that is too extreme; the next shock is therefore likely to reverse the prior judgment, generating the negative serial correlation in returns at horizons of three to five years documented by De Bondt & Thaler (1985). Overreaction is the cognitive content of Regime $2$ in the Barberis-Shleifer-Vishny (1998) model: after a streak of consistent earnings news, the representative investor switches from the conservative regime to a trend-following regime that over-weights recent evidence and produces the eventual overshoot. Together with conservatism, overreaction generates the joint pattern of $1$–$12$ month momentum and $13$–$36$ month reversal that any complete first-generation behavioral model must explain.

Intuition

Conservatism asks: what if the investor multiplies the likelihood ratio by a power less than one, dampening the update? Overreaction asks the symmetric question: what if she multiplies by a power greater than one, amplifying it? A signal that should move the posterior from $0.5$ to $0.7$ instead moves it to $0.85$. The direction of belief change is correct; the magnitude is exaggerated.

The bias shows up most cleanly after long streaks. A single noisy signal rarely produces enough subjective certainty for overreaction to bite — there is too little to extrapolate from. But once the same direction of news arrives several quarters in a row, the overreacting investor reads the streak as a strong indicator of a persistent shift, builds a posterior that overshoots the true rate, and then is forced to reverse when fundamentals fail to validate her conviction. The experimental evidence (Massey & Wu, $2005$; Bloomfield & Hales, $2002$) confirms exactly this pattern: subjects under-react to isolated signals and over-react to extended streaks. The same pattern shows up in the stock market as continuation followed by reversal.

Properties

(Parametric Form) Overreaction is $c>1$ in the Rabin template.

In the Rabin (1998) deviation from Bayes’ rule, $c$ governs the weight on the new signal’s likelihood ratio. Bayes is $c=1$. Overreaction is $c>1$, with empirically estimated values typically between $1.2$ and $2.0$ depending on the experimental setting and the length of the signal streak. The parameter $d$ on the prior is left at $1$, meaning the overreacting investor still uses the prior correctly — she just over-weights new evidence relative to it. This makes overreaction a strict mirror of conservatism along the signal-weight axis.

(Mechanism for Long-Run Reversal) Excess updating produces eventual overshoot.

If every investor in the market over-updates after a streak of consistent news, the cumulative price move overshoots the new fundamental value. The next surprise is therefore disproportionately likely to be in the opposite direction, generating the negative serial correlation in returns at horizons of three to five years that De Bondt & Thaler ($1985$) document. This is the cognitive content behind long-run reversal in the cross-section of stocks (past losers outperform past winners over $3$–$5$ year horizons) and in the time series of the aggregate market (high P/E ratios predict low subsequent returns).

(BSV Regime 2) Overreaction emerges after a streak.

In the BSV (1998) model, the representative investor switches between two regimes. Regime $1$ is conservative: each shock is treated as transient and absorbed only partially. Regime $2$ is trending: after several same-direction shocks, the agent updates toward the hypothesis that earnings are on a persistent path and over-weights subsequent confirming signals. The switch is endogenous — long streaks raise the posterior weight on the trending regime. The result is short-horizon under-reaction (Regime $1$) followed by long-horizon over-reaction (Regime $2$), which produces the observed continuation-then-reversal pattern from a single representative agent who oscillates between two biases.

(Asymmetry with Conservatism) The two operate on different signal lengths.

A pure conservative agent ($c<1$ throughout) generates continuation forever and no reversal. A pure overreacting agent ($c>1$ throughout) generates instantaneous overshoot and reversal but no continuation. Neither, alone, matches the empirical pattern. The data require both: under-reaction at horizons of one to twelve months and over-reaction at horizons beyond. The reconciliation, due to BSV and Griffin-Tversky ($1992$), is that the sign of the deviation depends on the length and consistency of the signal stream — short, isolated signals trigger conservatism; long, consistent streaks trigger overreaction. The two are not mutually exclusive cognitive states but a single agent’s response to different information environments.

(Empirical Hook) De Bondt-Thaler losers outperform winners over $3$–$5$ years.

De Bondt & Thaler (1985) rank stocks on cumulative return over the past three years, form decile portfolios, and track them over the following three years. The loser portfolio outperforms the winner portfolio by approximately $25\%$ over the three-year holding period, with the result robust to risk adjustment. The interpretation is that the original price moves were excessive relative to fundamentals — investors had over-reacted to the streak of bad news on the losers and the streak of good news on the winners, and the subsequent reversal corrects the overshoot. The De Bondt-Thaler result is the canonical empirical hook for overreaction in asset pricing.

(Tension with the Joint Hypothesis Problem) Reversal is hard to identify cleanly.

Long-run reversal is harder to interpret than momentum because the time horizon is long enough that risk-based explanations (time-varying expected returns, distress risk on losers, liquidity premia) become plausible. Fama ($1998$) argues that the De Bondt-Thaler result weakens substantially after correcting for size, book-to-market, and other risk factors. The behavioral response is that even after such corrections a residual reversal remains, and that the pattern is replicated internationally and across asset classes — but the cleanness of the test is lower than for short-horizon momentum. The interpretation of long-run reversal as belief overreaction is therefore the leading but not the unique reading.

Examples

Numerical update under $c=1.5$

An investor starts with prior odds $1:1$ that a firm is in a “high earnings” state versus a “low earnings” state. She observes a single positive earnings surprise with likelihood ratio $\frac{p(S\mid \text{high})}{p(S\mid \text{low})}=2$. The Bayesian posterior odds are $1\cdot 2=2:1$, so $\mathrm{Pr}(\text{high}\mid S)\approx 0.67$. The overreacting investor with $c=1.5$ computes posterior odds $1\cdot 2^{1.5}\approx 2.83:1$, so $\mathrm{Pr}(\text{high}\mid S)\approx 0.74$. The direction is the same; the magnitude is exaggerated by roughly seven percentage points. After three consecutive positive surprises, the Bayesian posterior is $2^3=8:1$ ($89\%$), while the overreacting investor’s posterior is $2^{4.5}\approx 22.6:1$ ($96\%$). The gap widens with the streak — exactly the pattern that produces eventual reversal when the streak ends.

De Bondt-Thaler (1985) winner-loser portfolios

Each year, rank all NYSE stocks by their cumulative return over the prior $36$ months. Form a “winner” portfolio of the top $35$ stocks and a “loser” portfolio of the bottom $35$ stocks. Hold each for the next $36$ months and measure cumulative excess returns. Pooling across portfolios formed in $1933$ through $1980$, the loser portfolio outperforms the market by approximately $19.6\%$ and the winner portfolio underperforms by approximately $5.0\%$ over the three-year holding period — a winner-minus-loser spread of about $-24.6\%$. The pattern is monotonic across the deciles: the worse the prior performance, the better the future performance. The behavioral reading is that prices over-reacted to the long streaks of good and bad news that built the original winner and loser categorizations, and the subsequent three years correct the overshoot. The result has been replicated outside the U.S. (Rouwenhorst, $1998$) and remains the canonical empirical hook for belief overreaction in the long-run cross-section.

See Also

• 8.6. Belief-Based Biases — parent hub.
• 8.6.1. Conservatism — sibling: the parametric mirror image ($c<1$ instead of $c>1$). Together they generate continuation and reversal.
• 8.6.5. Representativeness Heuristic — the cognitive root of overreaction after streaks (the gambler’s-fallacy/hot-hand axis).
• 8.6.7. Extrapolative Beliefs — the second-generation framework that subsumes overreaction inside a single recency-weighted-average mechanism.
• 8.9.1. Barberis-Shleifer-Vishny Model — the first-generation asset-pricing model where overreaction is Regime $2$.
• 8.8. Momentum — the short-horizon empirical pattern that conservatism explains; the long-horizon flip is overreaction.
• 8.6.4. Confirmation Bias — the parametric mirror image on the prior axis ($d>1$).