ABSTRACT
The probability of rejecting a true null hypothesis when specification error is present depends, in part, on the number of variables included in Z1 • In general. our results indicate, regardless of the type of rnisspeeifieation, that the use of p = :~ in the construction of Z1 as in (15) yields the most powerful RESET lest. However, this does not always hold, such as in model :3 where the HESET test with only the term j·12 included in the auxiliary regression yields higher power for relatively large positive misspecification (y > 0.3) and large negative misspecification (y < -0.7).
