Timing Structural Change : A Conditional Probabilistic Approach

We propose a strategy for assessing structural stability in time-series frameworks when potential change dates are unknown. Existing tests for structural stability have proven to be effective in detecting the presence of structural change, but procedures for identifying timing are highly inprecise. We present a likelihood-based procedure for assigning conditional probabilities to the occurrence of structural breaks at alternative dates. We find the procedure to be effective in improving the precision with which inferences regarding timing can be made. We illustrate parametric and non-parametric implementations of the procedure through a series of Monte Carlo experiments, and an assessment of the volatility reduction in the growth rate of U.S. GDP