Terasvirta JASA 1994

This is a replication of the examples from Terasvirta(1994). That paper describes the test for detecting possible "STAR" (Smooth Transition Autoregression) effects that is implemented in the @STARTest procedure. See Threshold Autoregressions for more on STAR and related models.

The empirical work in the paper uses two data files: the Canadian lynx data that is a standard in the non-linear time series literature, and quarterly data on German industrial production. There are three program files:

lynx.rpf

This picks a lag length for an autoregression, uses the @STARTest procedure to do a diagnostic test on a fixed regime autoregression vs an alternative of STAR or at least STAR-like behavior, estimates and analyzes the selected STAR model.

ger4ind.rpf

Largely the same as lynx.rpf applied to the German IP data.

lynxforecast.rpf

Shows methods of forecasting applied to (simplified versions of) the lynx data STAR model.

Note that the TARMODELS.RPF example is based upon a combination of lynx.rpf and lynxforecast.rpf. See the description of that example for step by step details of the calculations.