Difference-in-difference - Grangertest

[Lorenzo Chiesa]

Good afternoon,
I am performing a DiD with panel data and using xtreg (with fixed effects).
I found a statistically significant ATET both using the above-mentioned functions and with the xtdidregress method.

After doing using xtdidregress, I tested the parallel trends assumption for the pre-treatment period and rejected the null hypothesis of "no parallel trends".
Should I verify the presence of possible anticipatory effects with estate grangertest or this second test is needed only if there is evidence of "no parallel trends" for the pre-treatment period?? The manual provided by Stata is not very clear on this point.

Thank you in advance

[George Ford]

I think the null of ptrends is equal trends, not "no parallel trends" (you want to reject the latter but not the former).

How bad does the graph look?

granger is more like a pseudo-treatment test (is there an effect prior to the treatment). It is worth doing, but not for a ptrends test.

Mastering Metrix suggests including a unit specific linear trend when parallel trends does not hold. Not sure where the literature is on that option today.