Abstract
When agents have rational expectations, anticipated changes in the structure of the economy have an immediate affect on their behaviour. In this article, we investigate the interplay between a linear rational expectation model with predictable structural changes and reduced-form evidence of structural breaks. In our study, we vary the length of time between the announcement and the implementation of an inflation target change. Using a model similar to Ireland (Citation2007) and the method presented in Bai and Perron (Citation1998) and Bai and Perron (Citation2003) to estimate unknown structural breaks, Monte Carlo simulation results suggest that reduced-form evidence of structural breaks are broadly in line with what is predicted by forward-looking rational expectation models; that is, as the transition period increases, break estimates gradually move farther from the policy announcement date.
Acknowledgement
We thank Mariano Kulish for detailed comments and suggestions.
Notes
1 See Schmidt-Hebbel and Tapia (Citation2002) for a discussion of the Chilean experience on the implementation of inflation targeting.
2 Another way to address our question could be to treat simulated data as a Markov switching process along the lines of Diebold et al. (Citation1994) and Filardo (Citation1994). In this context, candidate break points could be associated to the time-varying transition probabilities between the policy announcement and implementation period. Such an approach, however, entails making assumptions about the stochastic behaviour of the states (see Kim and Nelson, Citation1999). In this article, we take an agnostic view leaving these extensions for future research.
3 See, for example, Stock and Watson (Citation1996), Garcia and Perron (Citation1996) and Bai and Perron (Citation2003).
4 When using the Phillips curve, similar to the AR(1) case, we allow for breaks in both the intercept and the coefficient on lagged inflation. Allowing for shifts solely in the intercept (partial break) did not change our main findings. Results are available upon request.
5 Results for transition periods of 8 and 20 quarters are broadly similar and available upon request.
6 These results can also be interpreted as test size distortions not leading to misalignments between the total number of breaks suggested by statistical evidence and the actual number of regime shifts simulated from a structural model.
7 Results in are for a sample size of 200 data points and are robust for sample sizes of 100 and 150.