*
* Example from section 6/5/2, pp 294-297
*
open data bankdata.dat
calendar(m) 1988:2
data(format=free,org=columns) 1988:2 1993:1 eom aaa govt34
*
set deom = eom{-1}-eom
set d34  = govt34{-1}-govt34
*
seasonal seasons
*
set trend  = t
set trend2 = t**2
set trend3 = t**3
*
* It turns out that this stepwise regression will end up picking the same model
* used. (The .35 significance levels are roughly the ones that indicate whether
* the addition or deletion of a variable will raise the R-bar squared).
*
stwise(method=stepwise,slstay=0.35,slenter=0.35) deom * 53
# constant aaa govt34 d34 seasons{-11 to -1} trend
*
* The simple static forecasts for a regression can be computed using the
* instruction PRJ. The STDERRS option allows you to also compute the standard
* errors of forecast.
*
prj(stderrs=stderrs) forecast 54 59
*
* There seems to be an error in the forecast calculations in the text, since the
* calculation on p. 296 uses .028 rather than .080 for the value of D(3-4).
*
print(picture="*.###") 54 59 deom forecast stderrs
*
graph(footer="Figure 6-9 Predictions for bank data") 2
# deom
# forecast
*
set lower = forecast+%invnormal(.05)*stderrs
set upper = forecast+%invnormal(.95)*stderrs
graph(footer="Figure 6-9 Predictions for bank data",overlay=fan,ovcount=2,ovsame) 4
# deom
# forecast
# lower
# upper