Assessing the Price–Earnings Relation via Machine Learning
77 Pages Posted: 14 Apr 2021
Date Written: October 10, 2019
The relation between stock prices and accounting earnings has been a central theme in accounting research for more than half a century. By almost exclusively emphasizing a lin- ear, parametric approach, the literature has been unable to convincingly overcome a number of modeling issues, including the effects of non-linearity and lack of fit.
We show that a non-linear, non-parametric approach based on recent advances in statistics and machine learning successfully addresses these modeling issues. Our methodology is validated in three ways: 1) residuals meet the orthogonality property, i.e., the estimated relation fits, 2) the firm-specific dependence of price on earnings agrees with theoretical predictions in the literature, and 3) our empirical earnings response coefficients yield reasonable cost of capital estimates.
Keywords: earnings, price–earnings relation, earnings response coefficient, price level regression, machine learning, non-linear association, non-parametric regression
JEL Classification: G10, G30, M41
Suggested Citation: Suggested Citation