Specification of Logistic Smooth Transition Auto-Regressive Model: An Alternative Representation of the Transition Function

Abstract

Problem statement: In this study we proposed an alternative represent ation of Smooth Transition Auto-Regressive (STAR) model. Logistic Smooth Transition Auto-Regressive (LSTAR) model, Exponential Smooth Transition Auto-Regressive (ESTAR) models and Absolute Logistic Smooth Transition Auto-Regressive (ALSTAR) were being variously sighted and used in empirical research to model nonlinearity adjustments in the b ehavior of macro-economic variables. Approach: We showed that forecasts produced using this alternati ve model outperform most of these popular forecasting models. Results: We used LSTAR model as the benchmark. LSTAR model described asymmetrical non-linear adjustment process, while t he ESTAR model described symmetrical non- linear adjustment process. ALSTAR was introduced in the literature to modify the undesired property of asymmetry of LSTAR. The forecasting performances of these models were investigated. Conclusion: The alternative model we proposed in this study, t he Error Logistic Smooth Transition Regression (ELSTR) was shown to posses a better estimation of model parameters and performs better than those earlier ones. We demonstrated our claim with life data.