In PP test, the truncation parameter is the truncation parameter in estimating the so-called long run variance (HAC). The reason for using this estimator is that where the innovations of the unit root process are correlated, the limiting distribution of the DF statistic is not nuisance parameter free and depends on the long run variance (Phillips (1987), PP (1988)). Thus you need the estimator to correct for the nuisance parameter. This phenomenon is similar to that of ADF test where lagged differences are introduced to eliminate the effect of serial correlation.
Theoretically speaking, under general assumptions, the truncation parameter p should go to infinity as sample size go to infinity. There are data-drived methods to select p (Newey and West (1987) or Andrews (1991)). However, it seem that in the software, the p value are fixed before doing unit root test. Hope my explanation may help.
|