Bias Reduced Band Spectrum Least Squares in Fractional

Band spectrum regression procedure in a bivariate model of fractional nonstationary cointegration is proposed. Both variables and cointegrating error in the system are assumed to be fractionally integrated processes. The orders of integrations are unknown, but no need to be pre-estimated. The proposed estimator can reduce bias by modifying a frequency domain regression, and it is just a simple least squares and easy to use. Unlike other available estimation procedures, the estimator is free from any preliminary estimation of short memory components and fractional parameter. It is also expected to be less volatile and more reliable, which can be confirmed by finite sample performances. A limited version of asymptotic theory will be developed and some simulation results will also be provided.