Testing Parameter Stability: A Wild Bootstrap Approach

Unknown-breakpoint tests for possible structural change have become standard in recent years, with the most popular being the so-called Sup-F tests, whose asymptotic distribution was derived by Andrews (1993). We highlight two problems that lead to poor performance when testing for structural breaks in dynamic time series models using the Andrews critical values: High persistence of explanatory variables and heteroskedasticity. We propose a so-called ``wild bootstrap'' approach to generating critical values for the Sup-F statistic and report that this approach performs well across a wide variety of possible data generating processes, including those with large coefficients on lagged dependent variables and heteroskedasticity.