Regional price levels in Germany

Abstract

Cross-sectional evidence on price levels is scarce for all countries. However, several studies suggest that there might exist considerable differences in price levels within countries, which has obvious welfare implications. A sample of price levels in 50 German cities in 1993 is used to analyse the determinants of inter-city price level differentials. The most important explanatory variables for price level differentials are population size and density and the average wage level. Using this information, the price levels are predicted in all 440 German districts and aggregated to the state level. At the state level convergence of the price levels to a common mean is found, but at a very low speed. The estimated half-life is about 19 years.

Acknowledgements

An earlier version of the paper was presented in the brown-bag seminar of the University of Mannheim and at the ERSA 2003 conference in Jyväskylä. I am grateful to the participants for helpful comments. I owe special thanks to Heinz Holländer, Wolfgang Leininger, Bernd Fitzenberger, Uwe Blien, Thiess Büttner, Gergana Dimitrova, and Jürgen von Hagen. The sole responsibility for any errors is, of course, mine.

Notes

¹ Tests of New Economic Geography models have been somewhat unsatisfactory so far. A reason for this is the lack of regional price level data (see Hanson, 1998; Roos, 2001). The same is true for studies analysing the determinants of migration flows (see Decressin, 1994).

² Cecchetti et al. (2002) estimate a half-life of price level differentials between US cities of nine years. For US cities and individual commodities Parsley and Wei (1996) estimate half-lifes between four to five quarters (tradable goods) and fifteen quarters (services).

³ The density variable might also capture high land prices. In principle, land prices are available at the regional level but not for all regions and not for the year 1993.

⁴ However, compared with the variation of other economic variables, the coefficients of variation are rather low as the following table shows:

Table

⁵ The average growth rate of the Eastern states in 1993 was 21.2%.

⁶ The fitted trend lines have high values of R ². In general, they are larger than 0.9.

⁷ The number of overnight stays is available from 1995. Since this number is fluctuating strongly, there is no easy way of predicting the needed values. The classification of regions in tourist and non-tourist regions seems more robust against these errors.

⁸ An earlier version of the paper had instrumented the wage variable. The prediction result was very similar. The correlation between the two sets of predictions in 0.93.

⁹ RMSE_in and RMSE_out are not corrected for degrees of freedom, wheras RMSE is.

¹⁰ A data file with the estimated price levels is available from the author upon request.

¹¹ The values in parentheses are the White standard errors.