Options
Assessing the price-earnings association in the age of machine learning
Date de parution
2019
Résumé
Most of the seminal papers mapping the relation between earnings and security prices predate the recent exponential developments in the field of machine learning. Our analysis is an example of how the new powerful non-linear estimation techniques and three dimensional visualization of data can provide the accounting researcher with new insights and/or help her document more forcefully patterns predicted by theoretical considerations.
We show that state-of-the-art linear models are problematic for hypothesis testing when fit to the non-linear relation between share prices and earnings. To bring remedy to the failure of linear modeling, we introduce a non-linear research design based on rigorous statistical considerations and accounting input and which consistently estimates prices’ relation to earnings. We validate the non-linear research design by verifying that the non-linear levels regression earnings-response coefficients (ERC) have the ’right’
size and yield economically justifiable risk-premium values. Consequently, the non-linearity of the price-earnings association provides a simple explanation for the small ERCs observed by prior research based on the linear model.
We show that state-of-the-art linear models are problematic for hypothesis testing when fit to the non-linear relation between share prices and earnings. To bring remedy to the failure of linear modeling, we introduce a non-linear research design based on rigorous statistical considerations and accounting input and which consistently estimates prices’ relation to earnings. We validate the non-linear research design by verifying that the non-linear levels regression earnings-response coefficients (ERC) have the ’right’
size and yield economically justifiable risk-premium values. Consequently, the non-linearity of the price-earnings association provides a simple explanation for the small ERCs observed by prior research based on the linear model.
Identifiants
Type de publication
working paper
Dossier(s) à télécharger
