The Size and the Power of Unit Root Tests Against Fractional Alternatives ; A Monte Carlo Investigation

This paper investigates the size and power of a number of unit root tests, currently in use in both applied macro and financial economics, when the data generating process is fractionally integrated. The long persistence characteristic of fractionally integrated processes lowers the power of unit root tests, when compared with their power against stationary alternatives. The performance of the unit root tests is investigated extensively in a range of experiments, which permit the fractional process to have normal white noise errors, non-normal white noise errors, heteroscedastic errors and serially correlated errors. Power is not seriously affected by non-normality, but can be adversely affected by heteroscedastic and serially correlated errors.