Robust stochastic frontier using Cauchy distribution for noise component to measure efficiency of rice farming in East Java

Abstract

Abstract Several studies have been conducted to estimate efficiency of rice farming in Indonesia. Those studies generally apply standard stochastic frontier (SF) model which assume that noise component has normal distribution. However, the normality assumption of noise is inadequate if the outliers are present. This paper employs Cauchy distribution for noise component to produce a more robust SF model. There are two variants of models that are employed: Cauchy-Half Normal and Cauchy-Exponential models. These models are applied to measure technical efficiency of rice farming in East Java. The findings presented in this study are based primarily on data from Cost Structure of Paddy Cultivation Household Survey 2017 (SOUT2017-SPD) that is compiled by Statistics Indonesia. Output and input variables are linked using transcendental logarithmic production function, while the parameters are estimated using maximum simulated likelihood. The results showed that Cauchy-Half Normal and Cauchy-Exponential models effectively reduce the effect of outliers. The performance of model is improved after the outliers are handled. Cauchy-Half Normal and Cauchy-Exponential models revise the technical efficiency scores of several rice farming units.