Effect of tapering on accuracy of forecasts made with stable estimators of vector autoregressive processes

The effect of tapering in improving forecast accuracy in a vector autoregressive process is studied. The estimators used to produce the forecasts are stable estimators, i.e., the estimated process is stationary. A new correlation-type stable estimator for the first order vector autoregressive process is suggested for a better understanding the effect of tapering on stable estimators and to compare with the performance of the popular Yule-Walker estimator. Empirical investigations indicate that for multidimensional higher order processes tapering can significantly improve the forecasting performance of the Yule-Walker estimates. A feasible procedure for choosing the optimal degree of tapering in practice is suggested.