*
* Example 10.2.1 from pp 341-342
*
open data sbl2.tsm
calendar(m) 1975
data(format=free,org=columns) 1975:1 1984:12 step deaths
*
* The step function could have been created with
* set step = t>=1983:2
*
* You can do this with a LINREG instruction, but it will be more convenient here
* to just use BOXJENK, since it handles bookkeeping for the seasonal differencing.
*
boxjenk(sdiff=1,apply,reg) deaths / resids
# step
@regcorrs
*
* Find the best fitting MA model for the residuals
*
@bjautofit(qmax=13,pmax=0) resids
*
* This fits the MA(12) noise term with the intervention variable. This is a very
* difficult model to fit and the BOXJENK instruction is unable to estimate it by
* maximum likelihood, so we take the MAXL option out.
*
boxjenk(sdiff=1,ma=12,reg,applydifferences) deaths
# step
*
* This does an alternative noise term which just includes the seasonal MA,
* omitting MA(1),...,MA(11). This seems to provide an adequate model.
*
boxjenk(sdiff=1,sma=1,maxl,reg,applydifferences) deaths
# step