Synthetic Control Method Placebo Test

[Aymen Al Lawati]

Hi, I am currently using the synth command for my dissertation. The results came out exactly the way I wanted. The in-time placebo test worked. However, when I tried conducting the In space placebo test, I used the code provided in synth help. When I tried using the same command I used earlier (just changed the treated unit in the code), the SCM was not able to follow the trajectory in the pre-treatment period. I don't know what to do. Can someone please help? Should I change the code?

[Dimitriy V. Masterov]

There is no guarantee that a convex combination of others could approximate every unit in your data. For example, you cannot have a weighted average of 1 and 2 add up to 10, with weights between 0 and 1 adding up to one. I think you should try to limit the in-space placebo exercise to units that have a pre-treatment RMSE that is not too dissimilar from your treated unit.

[Carlo Lazzaro]

Aymen:
welcome to this forum.
The same may also occurr when you reduce the number of your control units from N-1 to 1 mimicking the approach whose results are reported in Table 3, page 50'7 of Comparative Politics and the Synthetic Control Method - Abadie - 2015 - American Journal of Political Science - Wiley Online Library

[Aymen Al Lawati]

This might sound like a simple question, however I don't know if the RMSPE that STATA provides is for the pre or the post treatment period? Also, from what I understood from page 502 of the tobacco study, it is fine if the SCM is not able to follow the trajectory of the countries in the donor pool in the pre-treatment period. I wish I can give more details about my topic but I am not allowed to share it due to the university's policy.

[Carlo Lazzaro]

Aymen:
quoting page 496 of the same article (rightmost column):

Following Abadie and Gardeazabal (2003), in the empirical section of this article we choose V among positive
definite and diagonal matrices such that the mean squared prediction error of the outcome variable is minimized for the preintervention periods (see Abadie and Gardeazabal 2003, appendixB, for details).

[Aymen Al Lawati]

Thank you for your help Carlo! Wish you the best!