Handling Outliers in The Stochastic Frontier Model Using Cauchy and Rayleigh Distributions to Measure Technical Efficiency of Rice Farming Bussiness

Abstract

Technical Efficiency (TE) is one of the essential indicators used to evaluate the development of the agricultural sector. Generally, the statistical model used to measure TE is a stochastic frontier model with the noise being normally distributed and the inefficiency being half-normally distributed. The problem is that the model is not robust when outlier observations occur. This study proposed a stochastic production frontier model with a fat-tailed distribution to overcome outlier observations. This study used two stochastic models with fat-tailed distribution used in this study: Chaucy-half normal and normal-Rayleigh stochastic models. The translog production function was selected as a connecting function between the input and output. These two models were applied to estimate the technical efficiency of rice farming in Central Kalimantan. The results showed that the proposed model could reduce or eliminate outliers in the remaining inefficiencies. In addition, the range of technical efficiency values had also narrowed. Thus, the Chaucy-half normal and normal-Rayleigh stochastic models can handle outliers.