Bayesian Reconciliation of Return Predictability.

Area of Science:

• Financial econometrics
• Bayesian statistics
• Asset pricing

Background:

• Return predictability is a key question in financial economics.
• Existing methods like ordinary least squares (OLS) and reduced-bias estimators have limitations.
• Vector autoregressive (VAR) models are commonly used for analyzing financial time series.

Purpose of the Study:

• To develop and evaluate a new Bayesian approach for investigating return predictability.
• To compare the performance of the proposed Bayesian method against OLS and reduced-bias estimators.
• To assess return predictability using historical financial data and various prediction variables.

Main Methods:

• Development of a novel shrinkage prior for a key parameter in a bivariate VAR model.
• Comparison of the Bayesian approach with OLS and the Amihud and Hurvich (2004) reduced-bias estimator via simulation.
• Application of the methodology to historical CRSP value-weighted returns and dividend-price ratios, and alternative predictors (Welch & Goyal, 2008).

Main Results:

• Simulation studies indicate the Bayesian approach surpasses the reduced-bias estimator in controlling false positives and negatives.
• Empirical analysis using 1926-2004 data supports no return predictability; recent data (1953-2021) show weak predictability.
• Alternative prediction variables also yield weak evidence for return predictability, confirmed by out-of-sample forecasting.

Conclusions:

• The proposed Bayesian method offers a robust alternative for estimating return predictability.
• Evidence for return predictability is sensitive to the data period and prediction variables used.
• The findings suggest limited, albeit present, predictability in recent equity markets.