@TVARSET sets up a VAR SYSTEM for time-varying coefficients estimation. It is based upon the standard Minnesota prior for the mean with the variances for time variation derived proportionately from those. An example of its use is provided in TVARSET.RPF.

 

@TVARSET( options )  (no parameters)

# list of dependent variables

 

Options

MODEL=model to define [must be used]

 

LAGS =number of lags for each dependent variable.

 

The following are either identical, or very similar, to the corresponding options on SPECIFY.

 

TIGHTNESS=parameter of overall tightness [0.20]

The overall tightness of the prior.

 

OTHER=relative weight on lags of other variables[1.0]

A value other than 1 gives an prior which downweights the other variables relative to the dependent variable.

 

MEAN=mean of first own lag (same across equations) [1.0]

MVECTOR=VECTOR of means of first own lags [unused]

Use one of these if you want the means of the prior distributions on the first own lag to be something other than 1.0. The MEAN option sets them all to the specified first lag mean value. The MVECTOR option sets the mean for the first own lag in equation j to entry j of the VECTOR of first lag means.

 

LTYPE=[HARMONIC]/GEOMETRIC

DECAY=parameter of lag decay [no decay]

These two options control the how the standard deviation changes with increasing lags. (Note that it's LTYPE rather than LAGTYPE, since the option LAGS is already in use). With \(d\)=parameter of lag decay, SPECIFY uses the formulas:

 

\(g(l){\rm{ }} = {l^{ - d}}\) for LTYPE=HARMONIC and

\(g(l){\rm{ }} = {d^{l - 1}}\)           for LTYPE=GEOMETRIC

 

By default there is no decay of standard deviations with increasing lags. Notice that for HARMONIC, \(d=0\) is the default and larger values produce a tighter prior, while for GEOMETRIC, \(d=1\) is the default and smaller values produce a tighter prior.

 

These are options which are specific to time-varying parameters models. 

 

SEFACTOR=fraction of OLS autoregression variance used for variance of the observation equations [.9].

TVREL=fraction of the initial covariance matrix used for time-varying matrix [0.00000001]

The .9 for the SEFACTOR allows for the fact that the variance of a (fixed parameter) OLS autoregression is likely to over-estimate the equation variances in a time-varying parameter model. While the TVREL default seems remarkably small, experience has shown that substantially larger values work poorly, mainly because the number of coefficients in the typical VAR makes it too easy for the model to "explain" individual data points by shifting coefficients if the time-variation isn't sufficiently stiff.

 

CONTIGHT=standard deviation on the constant term (relative to the equation standard error) [.0001]

Variables Defined

Copyright © 2026 Thomas A. Doan