Bayesian Integration of Event Signals with Equity Factor Models

Summary

This report describes combining event-driven signals with a multi-factor stock selection model for Chinese equities. It first recommends estimating event-related abnormal returns with cross-sectional regressions that control for industry and market-cap effects, then using rank tests to identify events with meaningful alpha. Rather than adding event alpha directly to factor alpha, it treats factor-model residual return forecasts as a prior and event forecasts as observations, following a Black–Litterman-style Bayesian approach. The posterior mean weights each forecast according to its predictive uncertainty.

The report evaluates a monthly CSI 500 enhancement strategy alongside four negative events and two positive events. It finds that negative events had more robust residual-return forecasting power and could improve the factor model, while positive-event alpha often appeared before announcement and could weaken the combined forecasts. The evidence is specific to the tested Chinese market setting and event sample; model failure and extreme market conditions remain risks, and sparse event coverage limited the monthly strategy’s enhancement.

Key ideas

• Control for industry and market capitalization when estimating event-related abnormal returns.
• Combine event and factor forecasts as uncertainty-weighted predictions rather than simply adding their alphas.
• Use factor-model residual return forecasts as a prior and event forecasts as new observations in a Bayesian update.
• In the tested Chinese equity setting, negative events improved forecasts more reliably than positive events.
• A larger event library is not automatically better; event signals need robust alpha and predictive power.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.