Stationary properties of the real interest rate and the per-capita consumption growth rate: empirical evidence for theoretical arguments

Abstract

Many economic theories connecting the real interest rate and the per-capita consumption growth rate require that both rates evolve together over time. This article investigates whether these rates present similar stationary behaviour for the seven most industrialized countries over the 1957–2005 period. The analysis relies on the unit root tests developed by Elliott et al. (1996) and Lopez (2006) to look for stationary or regime-wise stationary behaviour, respectively. Furthermore, the final break selection uses Bai and Perron's (2003) method. The results show, for all the countries considered, that both rates are either stationary or regime-wise stationary with the same number of breaks and, mostly, with corresponding dates. The results hold whether the rates are calculated annually or quarterly.

Acknowledgements

Lopez gratefully acknowledges the financial support of the Charles Phelps Taft Research Center at the University of Cincinnati. The authors would like to thank Gary Ferrier, Harland Wm. Whitmore Jr and the participants of the 2005 Missouri Economic Conference, for helpful comments and discussions.

Notes

¹ Rose (1988) concludes that there is no evidence of stationarity for the real interest rate of 18 OECD countries. These results initiated a profound controversy in the literature due to the implications that a nonstationary real interest rate has for some major economic and financial theories. In essence, every model that deals with intertemporal decisions (consumption, investment, savings) uses a constant (stationary) ex ante real interest rate as the foundation for the theory and/or the results regarding optimal choices of individuals, firms and policy makers.

² Regime-wise stationarity means that the series are overall nonstationary because of the presence of time breaks, but it is not I(1).

³ Ferson and Merrick (1987) and Rapach and Weber (2004) provide a deeper discussion regarding the reasons why these specific features should be observed in the data in order for the theory to hold.

⁴ Lopez (2006) compares the finite sample properties of these tests and of Saikkonen and Lutkepohl's (2002), and concludes that the tests used here have a higher finite sample power.

⁵ It should be noted that our data for Germany includes the reunification period. Therefore, our results with respect to this country should be taken with caution.