Page 1 of 1

Serial Correlation Lagged Dependent Variable

Posted: Fri Oct 27, 2017 7:26 pm
by spiatkow

I have time-series data. My DV is nonstationary. To make it stationary, I created a first difference of my DV. It seems, however, that this differencing introduced serial correlation into the model (perhaps due to overdifferencing, as I read) because bgodfrey test indicated that there is a significant serial correlation. When I include a LAGGED DV, bgodfrey test reveals that serial correlation is no longer present. It is still present when I INCLUDE LAGGED DIFFERENCED DV.
My question is: should I include LAGGED DV or should I include LAGGED DIFFERENCED DV on the right side of equation (when my DV on left side of equation is differenced) to remove this serial correlation.

In other words , which equation is appropriate:
differenced Y = x1+x2 + lagged Y


differenced Y = x1+x2 +lagged and differenced Y

Is there any other way that I can solve these problems of overdifferencing?

Thank you in advance.