Wild Bootstrap Tests for Autocorrelation in Vector Autoregressive Models

Standard asymptotic and residual-based bootstrap tests for error autocorrela- tion are unreliable in the presence of conditional heteroskedasticity. In this article we propose wild bootstrap tests for autocorrelation in vector autoregressive mod- els when the errors are conditionally heteroskedastic. In particular, we investigate the properties of Lagrange multiplier tests. Monte Carlo simulations show that the wild bootstrap tests have satisfactory size properties in models with con- stant conditional correlation generalised autoregressive conditional heteroskedas- tic (CCC-GARCH) errors, whereas the standard asymptotic and residual-based bootstrap tests are oversized. The tests are applied to credit default swap prices and Euribor interest rates