SUR estimation of multiple time-series models with heteroscedasticity and serial correlation of unknown form

Abstract Ordinary least squares (OLS) estimation with non-parametric estimation of the coefficient's covariance matrix is a widely used procedure when the pattern of correlations of the errors is unknown. With multiple time series the seemingly unrelated regressions (SUR) estimator is a natural alternative to OLS. Simulation results show that the SUR estimator can be substantially more efficient than OLS. A non-parametric covariance matrix estimator is still required to deal with remaining heteroscedasticity and serial correlation. Further refinements are possible when there is more specific prior information on the conditional autocovariances.