Beyond the last surprise: Reviving PEAD with machine learning and historical earnings

Tomasz Kaczmarek, Adam Zaremba,
Beyond the last surprise: Reviving PEAD with machine learning and historical earnings,
Finance Research Letters,
Volume 86, Part E,
2025,
108751,
ISSN 1544-6123,
https://doi.org/10.1016/j.frl.2025.108751.
(https://www.sciencedirect.com/science/article/pii/S1544612325020057)
Abstract: How much information about future returns is hidden in a stock’s earnings surprises history? To answer this, we use elastic net models that extract information from multiple quarters of standardized unexpected earnings (SUE). We find that accounting for earnings patterns from the distant past substantially improves return forecasts: Sharpe ratios nearly double, and alphas remain significant even after controlling for one-quarter SUE and streak effects. While the traditional post-earnings-announcement drift (PEAD) anomaly has weakened, our approach effectively revives it. Gains are especially strong among large-cap stocks, where the latest surprises are quickly priced in, but older ones remain overlooked.
Keywords: Standardized unanticipated earnings; Machine learning; Post-earnings announcement drift; SUE; PEAD; Elastic net; Return predictability