Discussion Papers, Interdisciplinary Research Project 373: Quantification and Simulation of Economic Processes 2000,87
In the sequel of its seminal application in Davidson, Hendry, Srba and Yeo (1978) the single equation error correction model has been widely used in empirical practice. Providing a clear distinction between short- and long-run dynamics this model allows OLS-methods to be as efficient as (multivariate) full information maximum likelihood methods under a few assumptions on weak exogeneity and cointegration. We consider OLS-based tests on long-run relationships, weak exogeneity and short-run dynamics. For the latter issues it is known that common test-statistics are no longer pivotal if model errors exhibit conditional heteroskedasticity. We show that the wild bootstrap provides convenient critical values for the considered OLS-based statistics under both homoskedastic and conditionally heteroskedastic model errors. The wild bootstrap is easy to implement and turns out to improve considerably the empirical size of common test statistics compared to first order asymptotic approximations. We prove further that the wild bootstrap retains its validity for inference within a system of pooled equations exhibiting cross sectional correlation. Opposite to feasible GLS methods our approach does not require any parametric specification of cross sectional correlation, and copes with time varying patterns of contemporaneous error correlation.