Monitoring stationarity and cointegration

Abstract

We propose a monitoring procedure to detect a structural change from stationary to integrated behavior. When the procedure is applied to the residuals of a relationship between integrated series it thus monitors a structural change from a cointegrating relationship to a spurious relationship. The cointegration monitoring procedure is based on residuals from modified least squares estimation, using either Fully Modifi ed, Dynamic or Integrated Modified OLS. The procedure is inspired by Chu et al. (1996) in that it is based on parameter estimation on a pre-break \calibration" period only rather than being based on sequential estimation over the full sample. We investigate the asymptotic behavior of the procedures under the null, for ( fixed and local) alternatives and in case of parameter changes. We also study the finite sample performance via simulations. An application to credit default swap spreads illustrates the potential usefulness of the procedure.