Bai and Perron's and spectral density methods for structural change detection in the US inflation process

Abstract

This paper addresses the issue of estimating the number of breaks and their locations in the monthly US inflation series using two different approaches to testing for structural changes. The first approach considers Bai and Perron's selection procedure based on a sequence of tests. This approach focuses on the instability problem in time. The second method uses a test similar to the one based on Kolmogorov–Smirnov statistics applied to the evolutionary spectrum. The results obtained are similar and economically significant.

Acknowledgements

We would like to thank Eric Girardin, Claude Deniau and Costin Protopopescu for their useful comments.

Notes

¹  is then interpreted as the minimal number of observations in each segment. From Bai and Perron (2003a), if the tests are not required and the estimation is the sole concern, then the minimal number of observations in each segment can be set to any value greater than q.

² This condition implies that E ( X _t ) = 0.

³ We make the same choice as Artis et al. (1992).

⁴ Proof: See, Ahamada and Boutahar (2002).

⁵ For more details, see Ben Aïssa and Boutahar (2003).

⁶ We adopt the same modelling strategy as Ben Aïssa and Jouini (2003) since it provides significant results and allows one to capture well the break points. Using the monthly US inflation series covering the period 1956:1–2002:9 and based on some selection procedures using the information criteria, these authors find significant results showing in particular that the evolution curve of inflation in the USA was flattened during the last twenty years since it is noted that this reduction in extent of inflation is stable and durable.

⁷ Figures in parentheses are the asymptotic critical values at the 5% significance level taken from Bai and Perron (2003b).

⁸ These confidence intervals are obtained using the asymptotic distribution given in Bai and Perron (1998).

⁹ Using 5% trimming does not allow one to select the date 2000:3 and another neighbour date as a change point.