Grid-Bootstrap Methods vs. Bayesian Analysis. Testing for Structural Breaks in the Conditional Variance of Nominal Interest Rate Spreads - Four Cases in Europe

I use numerical methods to test for the presence of one-time structural breaks in the conditional variance of nominal interest rate spreads in four European countries over a period of eleven years (Jan 1988 to Dec 1998). I start with an intuitive approach consisting of a sequence of breakpoint Chow tests performed at subsequent dates over a given subsample of the squared residuals of the autoregressions used to model the yield spreads. Results from this procedure are misleading and spurious to some extent because of the incorrect critical values produced, which make the interpretation of the test stastistics basically unreliable. I then switch to large Monte Carlo simulations and to a fixed-regressor grid-bootstrap method to derive the right critical values and refine the previous conclusions. Finally, I utilize classical Bayesian econometrics to estimate alternative models for the series of nominal spreads and to detect potential shifts in the innovation variances of the equations describing the data. Outcomes need some interpretation: in the cases of Germany and Spain a break might have occurred in 1990 and 1994 respectively, as derived from the grid bootstrap approach. Likewise, there is evidence of a shift in the case of France in 1996 according to the Bayesian techniques employed, which also validate the hypothesis of a break for Italian yield spreads in 1995.