Forecasting with STAR model
Posted: Tue Jun 22, 2010 12:47 pm
This is an example of forecasting with a STAR model (specifically an LSTAR model). It is based upon the model from the Terasvirta JASA 1994 replication, but with a simpler model. Because a STAR model is non-linear, the forecast, in the sense of the mean of the process, can't be computed analytically. Instead, it is calculated by averaging across simulated values. The non-stochastic forecast is also an object of interest, called the "eventual forecast function". This shows the way that the process cycles in the absence of shocks.
Program: Data: Graph:
Program: Data: Graph: