Specification, Estimation, and Evaluation of Smooth Transition Autoregressive Models

Energy consumption, survey data and the prediction of industrial production in Italy: a comparison and combination of different models

“…This relationship has been found to be highly non-linear. In particular, we show that it can be well approximated by a smooth transition regression model aÁ la Terasvirta (1994Terasvirta ( , 1996. Although we obtained no gains in predictive power, the results suggest that some improvement may indeed be obtained by other speci®cations within that class of model or by varying coecient estimators (e.g.…”

“…In particular, we argue that the cubic trend can be seen as an approximation of the evolution over time of the coecient of energy consumption. 2 In fact, by modelling the relationship between electric power consumption and industrial production according to a smooth transition function, as suggested by Terasvirta (1994Terasvirta ( , 1996, we were able to ®nd a speci®cation that does not include trend terms. We also analyse the predictive power of models based on data from business surveys conducted by, respectively, Isco (the Italian national institute for business cycle analysis) and CsC, the research department of Con®ndustria (the Confederation of Italian industry).…”

Testing nonlinearity: Decision rules for selecting between logistic and exponential STAR models

“…2 By relying on a Taylor series approximation of the transition function (between states) around the scale parameter, 3 we introduce a new specification strategy to choose between logistic and exponential STAR models. This alternative strategy is simpler and more successful in selecting the correct ESTAR model while avoiding the pitfalls (detailed below) of the procedure proposed in Teräsvirta (1994). Additional results concern the power of the nonlinearity tests discussed in Teräsvirta (1994) for which we offer some comments and alternatives.…”

“…This alternative strategy is simpler and more successful in selecting the correct ESTAR model while avoiding the pitfalls (detailed below) of the procedure proposed in Teräsvirta (1994). Additional results concern the power of the nonlinearity tests discussed in Teräsvirta (1994) for which we offer some comments and alternatives. 4 All the claims will be supported by Monte Carlo evidence and by a well known empirical example based on Teräsvirta and Anderson (1992).…”

“…This paper introduces a simple decision rule for selecting between alternative Smooth Transition Autoregressive (STAR) models which improves the one suggested by Teräsvirta (1994). STAR models are a general class of state-dependent non-linear time-series models in which the transition between states is endogenously generated.…”