Empirical Asset Pricing: Economic Significance and Economic Model Evaluation

Abstract

Harvey (2017) and The American Statistical Association (2016) point out that business decisions should not be based only on whether the p-value of an empirical model passes a specific threshold and that statistical significance (p-value) cannot measure the size of an effect or the importance of a result. In other words, for economic problems economic significance is required and an economic model evaluation criterion is desirable.

This paper derives a criterion for economic significance of valuation differences between empirical models and shows empirically that nearly all empirical models applied in business valuation are dis-similar, i.e., result in economically significant valuation differences. Motivated by the degree of dis-similarity between empirical models, an economic model evaluation criterion is developed. It judges the implicit economic assumptions revealed by computing the dual program of empirical models with the help of compliance with the economic principle and fit to institutional circumstances. Based on this economic model evaluation criterion our paper elaborates that within the group of cross-sectional price models quantile regression proves to be the best model because it is able to offer a good approximation to the economic principle and mimics best the institutional circumstances, in particular, if the regression is run without a constant. On the other hand, statistically more advanced models like generalized least squares regression deteriorates the implied economic content of models.

Keywords: Economic significance, Lagrange duality, Economic model evaluation

JEL Classification: G12, G19