Dear All,
In the time series, if there is an intervention, the modelling of intervention, one usually uses step function at that time point, say (0 0 0 0 1 1 1 1 1 1),  then the response series will change the behavior after this intervention. However as I am doing drug evaluation project, when a big drug group off the market, GPs usually will notice it or be informed  about 3-6 months in advance. Then they probably would change their prescriptions before that drug off the market.  Hence I observed this behavior change before this intervention.  As there is no response after intervention, this type of drug is no longer in the market.
Say (0, 0, 1/3, 2/3, 3/3, 1 1 1 1 1), using ramp term like 1/3, 2/3, 3/3 sometimes could capture this.  Using 1/3, 2/3, 3/3, assumes the change is equally spaced, but it could be not as well.  If it is not, there is always one point would be detected as an outlier in this ramp period in the model, but it is not necessary is an outlier as this is just the response of GP to the information. Is there anyone knowing about the general modelling approach for such problem? I may not state quite clear here. If there is any confusion, please point it out.

Thanks and Regards