On Sun, 14 Jul 2019, Giuseppe Vittucci wrote:
The FD specification should not include the constant if the model in
levels
does not include time dummies.
(or a time trend, presumably)
Given
Y_t = a + b X_t
by first-differencing you get:
Y_t - Y_t-1 = b ( X_t - X_t-1)
Alright, that makes sense. So it looks like we should go with the
way xtserial handles this. Any objections?
Allin