Smooth Transitions in Factor Augmented Models: Simulations and Applications in Macroeconomics

Sammanfattning: This paper is concerned with two important topics in macroeconomic research – structural instability and handling of large amount of data. The regime switching behavior of the data generating process is accounted for by using a smooth transition model. Furthermore, dimension reduction is achieved by using a factor augmented model with factors estimated by the method of principal components. A simulation study is conducted, investigating how inclusion of estimated factors affects the properties of the smooth transition estimators. The properties of a common test for linearity are also investigated as well as the predictive power of the model. It is found that the smooth transition parameters are consistently estimated at $T=500$ regardless of how the error terms of the factor model are specified. It is further shown that serial correlation in the error terms of the factor model severely reduces the power of the linearity test when $T<250$. The test is otherwise unaffected by the fact that estimated factors are being used. The paper further includes an empirical application, forecasting US industrial production, employment and CPI inflation using factor augmented smooth transition AR(1) models. In forecasting employment and CPI inflation, it is found that the models outperform the benchmark when an appropriate transition variable is used.