Is inflation stationary?

Abstract

Ninety–three world–wide inflation series are tested for unit roots. Treating the data series' innovations as draws from a symmetric stable distribution, with possibly infinite variance, reduces the number that appear stationary.

Acknowledgements

Wojciech W. Charemza gratefully acknowledges financial support of the INTAS Programme No. 03-51-7174 Nonstationary multivariate and nonlinear econometric models: theory and applications, and Daniela Hristova gratefully acknowledges the financial support of the ACE Project Modelling and Forecasting Inflationary Processes and the School of Social Sciences, City University London, Post-Doctoral Research Fellowship.

Notes

 To simulate symmetric stable random variables the algorithm of Chambers et al. (1976), encoded in GAUSS by J. Huston McCulloch, is used.

 For series with less than 200 observations the maximum lag length is set to 24.

 Here and elsewhere, to estimate the index of stability

the method suggested by McCulloch (1986) is used.