*
* AR model from pp 41
*
calendar(m) 1926
*
* The VW data are simple returns
*
open data ch02_vw_m(26-97).txt
data(format=free,org=columns) 1926:1 1997:12 vwsmp
*
* The YuleLags procedure analyzes a set of AR(p) models using the Yule-Walker
* equations to estimate the residual variances. This is a very efficient way to
* analyze a whole series of such models.
*
@yulelags(max=10) vwsmp
*
linreg vwsmp
# constant vwsmp{1 2 3}
*
* The 1-%beta(2)-%beta(3)-%beta(4) divisor can be obtained (more flexibly), by
* using the %EQNLAGPOLY function to pull out the lag polynomial for the dependent
* variable of the AR, and then evaluating it at 1.0. By using this, you can
* change the number of lags in the AR without having to rewrite the code to do
* this calculation.
*
compute meanret=%beta(1)/%polyvalue(%eqnlagpoly(0,vwsmp),1.0)
disp "Mean Return" meanret
*
* The REGCORRS instruction is designed to do the diagnostic checking of a time
* series model by examining the residual autocorrelations. The DFC option is for
* correcting the degrees of freedom of the Q statistic - it should be equal to
* the number of estimated AR (or AR+MA) parameters.
*
@RegCorrs(dfc=3,number=10,report,title="AR(3) Model")
*
* Try an AR(5) model
*
linreg vwsmp
# constant vwsmp{1 to 5}
compute meanret=%beta(1)/%polyvalue(%eqnlagpoly(0,vwsmp),1.0)
disp "Mean Return" meanret
@RegCorrs(dfc=5,number=10,report,title="AR(5) Model")