Note that the model you have just fit is intended only to illustrate the specification of interventions. It is not intended as an example of good forecasting practice.
The use of continuing (step and ramp type) interventions as predictors has some limitations that you should consider. If you model a change in trend with a simple ramp intervention, then the trend in the data before the date of the intervention has no influence on the forecasts. Likewise, when you use a step intervention, the average level of the series before the intervention has no influence on the forecasts.
Only the final trend and level at the end of the series are extrapolated into the forecast period. If a linear trend is the only pattern of interest, then instead of specifying step or ramp interventions, it would be simpler to adjust the period of fit so that the model ignores the data before the final trend or level change.
Step and ramp interventions are valuable when there are other patterns in the data—such as seasonality, autocorrelated errors, and error variance—that are stable across the changes in level or trend. Step and ramp interventions enable you to fit seasonal and error autocorrelation patterns to the whole series while fitting the trend only to the latter part of the series.
Point interventions are a useful tool for dealing with outliers in the data. A point intervention will fit the series value at the specified date exactly, and it has the effect of removing that point from the analysis. When you specify an effect time window, a point intervention will exactly fit as many additional points as the number of lags specified.