CMStatistics 2021: Start Registration
View Submission - CFE
Title: Bayesian reconciliation of the return predictability Authors:  Borys Koval - Vienna University of Economics and Business (Austria) [presenting]
Sylvia Fruehwirth-Schnatter - WU Vienna University of Economics and Business (Austria)
Leopold Soegner - Institute for Advanced Studies (Austria)
Abstract: A Vector Autoregression (VAR) for the returns, dividend growth and the dividend-price ratio is estimated, where the Bayesian Control Function approach is applied to account for biased coefficient estimates in the predictive regression. Motivated by financial literature we impose a stationarity condition on the auto-regressive dividend-price ratio process employing Bayesian priors. We develop two new priors, a Jeffreys prior and a prior based on the frequentist reduced-bias approach, and compare our Bayesian estimation routine to other approaches proposed in the literature (e.g., uniform and Reference prior) by means of an extensive simulation study. In terms of mean squared error (MSE), mean absolute error (MAE), and credible interval coverage, the proposed approach leads to superior performance relative to ordinary least squares estimation, a frequentist reduced-bias approach, and Bayesian estimation using priors proposed in the literature. We apply our methodology to financial data for the S\&P 500 and find strong evidence for return predictability after properly accounting for the correlation structure and imposing theory-motivated restrictions on the dividend-price ratio.