Estimating the autoregressive parameter: recursive mean adjustment and the initial condition

Abstract

The simulation results of Shin and So (2001) are revisited. It is shown that the properties of the initial condition of a time series have a substantial impact on the ability of recursive mean adjustment to reduce the negative bias associated with estimation of the autoregressive parameter. Interestingly, it is found that recursive mean adjustment can generate positive bias for a range of values of the initial condition.

Notes

In this paper unit root tests are considered in their ‘with intercept’ form, as the recursively mean-adjusted DF test relates to this specification.