Robustness of Two-Part Fractional Regression Models in Modelling Fractional Outcomes (Published)
The bounded nature of the fractional dependent variables, for instance in corporate finance leverage ratio clustering with a substantial number of observations at unit interval raises some important issues in estimation and inference. Ordinary Least Square (OLS) regression with Gaussian distributional assumption has been the main choice to model fractional outcomes in many business problems. Nevertheless, it is conceptually flawed to assume Gaussian distribution for a response variable in the interval [0,1]. Tobit model which is a Single-component method for modelling proportional outcome also share properties with OLS. Two-part Fractional regression models have been shown as the most natural way of modelling bounded, proportional response variables. Beta regression method has been used to achieve the objective in this paper.
Keywords: Beta regression, Fractional outcomes, Gaussian distribution, Ordinary Least Square, Tobit model