When Analysts Get It Wrong: Expectation Bias, Market Inefficiency, and What It Means for Investors

Key Takeaways

• Machine learning predicts analyst forecast errors using public data.
• High expectation bias stocks underperform, delivering ~11% annualized alpha.
• Bias stems from underreaction to bad news, not mere optimism.
• Effect persists globally and in firms without analyst coverage.
• After costs, strategy still yields ~0.5% monthly net alpha.

Pulse Analysis

The efficient market hypothesis has long held that publicly available information is instantly reflected in prices, leaving little room for systematic outperformance. Gao, Ma and Yuan overturn this view by constructing an Expectation Bias (EB) measure that isolates the predictable component of analyst forecast errors. Using a rolling XGBoost model fed with 70 firm characteristics—past returns, earnings history, revision patterns and size metrics—they demonstrate that EB can be estimated solely from data that would have been known before each quarter’s forecasts, turning a behavioral flaw into a quantifiable signal.

Empirical results are striking. A decile‑sorted EB portfolio delivers a raw spread of –0.92% per month, roughly –11% annualized, and retains a five‑factor alpha of 0.88% per month after adjusting for standard risk factors. The underperformance of high‑bias stocks is linked to analysts’ failure to fully incorporate recent negative news, a pattern that intensifies in high‑turnover, overconfident markets and among firms with volatile earnings. Crucially, the anomaly survives in stocks without analyst coverage and replicates across developed markets—from the UK to Japan—indicating a universal cognitive constraint rather than a U.S.‑specific reporting quirk.

For practitioners, the EB signal offers a high‑turnover, alpha‑generating strategy, though transaction costs trim net returns to about 0.5% per month. The persistence of the effect after costs and its independence from known anomalies suggest a genuine market inefficiency, albeit one that may erode as awareness spreads. More broadly, the findings pressure the semi‑strong EMH, implying that markets can systematically misprice assets when investors collectively underreact to new information. Future work will likely explore cost‑efficient implementations, risk‑adjusted variants, and the interaction of EB with other behavioral factors, keeping the debate over market efficiency alive.