*
* Housing starts example, pp 111-115
*
cal(m) 1946:1
open data hstarts.dat
data(format=prn,org=columns) 1946:1 1994:11
*
graph(footer="Figure 5.4 Housing Starts, 1946.01-1994.11")
# hstarts
graph(footer="Figure 5.5 Housing Starts, 1990.01-1994.11")
# hstarts 1990:1 *
*
* The SEASONAL instruction defines a dummy for the last period of the year (here
* the month of December). The first lead of this series (that is, the series
* created by taking the data for the next time period into the future) is
* November, the second lead is October, etc. The lag field seasonal{0 to -11} (in
* RATS, the -lags are leads) covers all 12 dummies. If you want all but one
* dummy, use {0 to -10}, which will leave out January.
*
seasonal seasons
*
* We need to copy the coefficients into a data series so we can graph them later.
* This is done using the SET instruction which makes SFACTORS equal to the
* elements of %BETA from positions 1 to 12.
*
linreg hstarts * 1993:12 resids
# seasons{0 to -11}
set sfactors 1 12 = %beta(t)
*
@regactfit(footer="Residual Plot")
*
* This graphs the series of seasonal factors. The NODATES option is used to force
* the graph to be labeled with sequence numbers rather than dates. We've chosen
* to graph these as a bar rather than a line graph.
*
graph(footer="Figure 5.7 Estimated Seasonal Factors: Housing Starts",nodates,style=bargraph)
# sfactors
*
* Generate forecasts with their standard errors, and compute 95% confidence bands.
*
prj(stderrs=stderrs) hfore 1994:1 1994:11
set upperf = hfore + 1.96*stderrs
set lowerf = hfore - 1.96*stderrs
*
* Create a series which is 1's in the period that we want to shade.
*
set forezone = %year(t)==1994
graph(footer="Figure 5.8 Housing Starts: History and Forecast (1994:1-1994:11)",shading=forezone) 4
# hstarts 1990:1 1993:12
# hfore
# upperf / 3
# lowerf / 3
graph(footer="Figure 5.9 Housing Starts: History. Forecast and Realization (1994:1-1994:11)",shading=forezone) 4
# hstarts 1990:1 1994:12
# hfore
# upperf / 3
# lowerf / 3