RDD forcing variable = year I am currently researching a gift-tax reform and have just found out that I will also get pre-reform data. I am wondering what people think about using the time of the reform as a forcing variable in an RDD setting. I have 2 years prior and 5 years after the reform and the idea is to show by how much the amount gifted to other people has changed. The obvious issue is that this could have been forseen. I am not in the lab right now but I assume that the amount prior was relatively flat and has since incrementally risen so there should be an effect...
 A: Yes, there is an issue if individuals can foresee the starting date of your reform, and adjust their behaviour. In this case, you would not be able to identify properly the impact of the reform itself, since your "pre" sample of untreated would somehow already contain treated people. You will have to think of whether in your case individuals can adjust, and how this affect the outcome variable of interest.
This is related to the so-called Ashenfelter dip, where Ashenfelter noted 
that earnings of participants in government training programmes declined in the period prior. See on this subject, among others:
-Heckman, Smith (2001) The Pre-programme Earnings Dip and the Determinants of Participation in a Social Programme. Implications for Simple Programme Evaluation Strategies. http://athens.src.uchicago.edu/jenni/dvmaster/FILES/ash_dip.pdf
A: I believe that you could, but the thing is that year could also be correlated with any number of other factors that could have changed at the same time. The usefulness of RDD is that the forcing variable is not correlated with other factors.
For example, in Angrist and Pischke's Mastering Metrics they provide the example of minimum legal age drinking laws. In this case you can use (age >= 21) as a forcing variable because not many other things change, and your sample all reaches age 21 at different times. If they all turned 21 at the same time, it wouldn't work as well because some other factors could have happened at that same time, which is what your year variable would represent.
Generally if you have a shock that is applied at a certain point, you can use a difference-in-differences approach instead, if you have a geographic area to which the shock was not applied to serve as your control.
