Estimation of Model Risk Caused by Endogeneity Using the Example of the Corporate Finance Regression Model

When constructing regression models that study the influence of various factors on return on equity (ROE), the issue of endogeneity can arise, leading to biased coefficient estimates. The aim of this work is to assess model risk based on models that account for the problem of endogeneity and those that do not. This paper provides definitions and sources of model risk, as well as methods for its assessment. The influence of various indicators on ROE is analyzed. The sources of endogeneity and the instrumental variable method as a way to address it are considered. To identify endogeneity, a correlation matrix of the residuals of the regression model with explanatory variables and the Hausman test were used. Tests for over-identification, "weak instruments," and under-identification were conducted to verify the quality of the selected instruments. The use of the instrumental variables method, along with correcting the model for heteroskedasticity and autocorrelation, allowed for the development of a final model that addresses the issue of endogeneity. The results of the study showed that the model accounting for endogeneity leads to a reduction in model risk and, consequently, to an increase in the accuracy of the obtained results compared to models that do not account for it. Thus, this work confirms the necessity of accounting for endogeneity to obtain more reliable estimates.