[Zurück]

T. Dangl, M. Halling, O. Randl:
"Equity Return Prediction: Are Coefficients Time Varying?";
Vortrag: NHH Conference 2006, Skarsnuten, Norwegen; 14.03.2006 - 17.03.2006.

Most papers in equity return prediction rely on the assumption that coefficients are constant over time. We want to question this assumption and evaluate it empirically. Therefore, we develop an econometric framework enabling us to compare models that assume static coefficients and models that allow for time variation in coefficients. We find strong empirical support for the validity of models allowing for time varying regression coefficients and document that the uncertainty about the true level of time-variation in coefficients is of equal magnitude than the uncertainty about the true choice of predictive variables. Furthermore, we find that best-performing static models change more often over time and contain a larger number of predictive variables than best performers among dynamic models. Finally, we address the question of overall predictability. We find in-sample predictability, are unable to unambiguously document out-of-sample predictability, and show that only predictive models with time varying coefficients consistently outperform the non-predictability benchmark model.

Equity return prediction, Bayesian econometrics, empirical asset pricing.

http://www.skinance.com/Papers/2006/DanHalRan.pdf