Estimating corporate investment efficiency with bias correction: a semiparametric panel model approach

Empirical studies often use the residuals from ordinary least squares regression models to represent certain discretionary or unexpected components and then regress these residuals on potential determinants. However, this two-step approach has been criticized for leading to biased estimates, invalid inferences, and unreliable empirical results. This paper shows that the shortcomings of the two-step approach and alternative existing methodologies are retained and even more pronounced when analyzing inefficient corporate investment. To address these shortcomings, we propose a novel semiparametric model tailored for investment efficiency analysis. Our model effectively mitigates estimation bias caused by inappropriate model design or misspecified model structure, and accurately discerns overinvestment, underinvestment, and efficient investment along with their respective probabilities. Applying our model to a sample of Chinese listed firms reveals significant, previously obscured nonlinear impacts of Tobin’s q and sales on investment. Our results reveal pronounced tendencies towards overinvestment, contradictory to existing models which reveal opposite tendencies towards underinvestment. Our model is applicable to various types of efficiency analysis, where each firm may exhibit different performance outcomes with associated probabilities.