*
* Example 7.3.2 from pp 238-241
*
open data ls2.tsm
data(format=free,org=columns) 1 150 lead sales
*
* Fit ARMA models to the first differences of the two series. Save the residuals
* and also the maximum likelihood estimates of the variances. (%SEESQ is
* adjusted for degrees of freedom).
*
boxjenk(diffs=1,ma=1,demean,maxl) lead / wlead
compute leadvar =%seesq*%ndf/%nobs
boxjenk(diffs=1,ar=1,ma=1,demean,maxl) sales / wsales
compute salesvar=%seesq*%ndf/%nobs
*
@crosscorr wsales wlead
*
* Compute an estimate for the coefficient on the third lag of the lead variable
*
compute bhat=sqrt(salesvar/leadvar)
disp "Estimated Coefficient on transfer model" bhat
*
* Compute the stationary noise from the two residuals processes.
* Identify and estimate a noise model for it
*
set nt = wsales-bhat*wlead{3}
@bjident nt
boxjenk(ma=1,demean,maxl) nt