Application of ml bases methods in tfp analysis: a case of olley-pakes method

Abstract

The work presented in this paper contributes to the problem of selection bias while calculating the exit probability in the Total-Factor Productivity. Most conventional methods of estimating TFP, such as Olley and Pakes (1996) use a parametric form such as probit function to calculate the exit probability. I introduce a machine learning model with a model selection approach for calculating exit probability, where the functional form is data driven. To demonstrate the application, I used the LASSO model of machine learning. Amongst many other advantages of LASSO method, one main advantage because of which I chose this method is that, unlike of conventional methods which Olley-pakes used in their paper, the LASSO method will select the firms which are going to exit , ultimately removing the human factor from the method, hence removing the selection bias problem. The paper contributes by updating the method for estimating TFP.