Intra-national Purchasing Power Parity and Balassa–Samuelson Effects in Italy

Abstract

Considering a sample of 71 Italian metropolitan areas, this paper goes beyond the assumption that a unique core inflationary process exists in a macroeconomy. It shows that local long-run inflation rates can display remarkable variability. On the one hand they are negatively correlated with productivity growth; on the other, the less competitive the local retail sector, the higher is long-run inflation.

Parité de pouvoir d'achat intranational et effets Balassa–Samuelson en Italie

Résumé La présente communication, qui examine un échantillon de 71 zones métropolitaines en Italie, va au-delà de l'hypothèse de l'existence, dans une macroéconomie, d'un processus unique d'inflation de base. Elle démontre que les taux d'inflation locaux de longue durée font preuve parfois d'une variabilité remarquable: d'un côté, ils sont en corrélation négative avec l'expansion de la productivité, de l'autre moins le secteur local du commerce au détail est compétitif, plus l'inflation à long terme est élevée.

Paridad de poder adquisitivo intranacional y efectos Balassa–Samuelson en Italia

Extracto Este trabajo, que considera una muestra de 71 áreas metropolitanas italianas, va más allá de suponer que dentro de una macroeconomía existe un único proceso inflacionario básico. Muestra que los índices locales de inflación a largo plazo pueden exhibir una variabilidad significativa. Por una parte, se correlacionan negativamente con el crecimiento de la productividad; por otra parte, cuanto menos competitivo es el sector minorista local, más alta es la inflación a largo plazo.

Keywords:

• Purchasing power parity
• long-run inflation
• Balassa–Samuelson model
• Kaldor–Verdoorn model

JEL CLASSIFICATION::

• R1
• E31
• F49

Acknowledgements

The author is grateful, for their helpful comments, to an anonymous referee, Guido Ascari, Stefano Schiavo, Francesco Bogliacino, Roberto Basile, and all attendees at the seminars held during the conferences on ‘Small Open Economies in a Globalized World II’ in Waterloo, Canada (13 June 2008) and AISRE 2008 in Bari, Italy (25 June 2008). The usual disclaimer applies.

Notes

1. For reviews see Froot & Rogoff (1995), Rogoff (1996) and Taylor & Taylor (2004).

2. On pilot studies see, for instance, Van Teijlingen & Hundley (2001).

3. See Rogoff (1996), p. 663), where the hypothesis that government spending may have an effect on PPP is discussed.

4. Intranational price convergence has recently been the topic of a number of papers, such as Fan & Wei (2006) for China, Ceglowski (2003) for Canada, Dan & Bhattacharya (2008) for India. Morshed (2007), instead, focused on Bangladeshi and Pakistani cities, seeking to determine whether state borders have an impact on price convergence.

5. One of the most active researchers in the field is David Papel; see for instance the papers cited in Banerjee et al. (2005).

6. Where the long-run is identified by a low elasticity of substitution of labour inputs in the tradable and non-tradable sectors.

7. It is possible to consider the model as also a heterogeneous panel one. In this case, the seasonal dummies will also account for the possible effects of national common factors, though factor loadings have been restricted to be constant across time and allowed to vary across different spatial units. Common factors in regional inflation dynamics have been investigated by Beck et al. (2009) and they were not found to reduce the variability of idiosyncratic parameters.

8. We used the normalization for a log–linear equation given that .

9. Long-run inflation rates by metropolitan areas are set out in detail in Table A1 in the Appendix.

10. Italy has five NUTS1 regions: North-West, North-East, Centre, South and Islands. The last two are often merged because they have similar economic and socio-political features. We adhered to this practice in defining the Italian macro-regions. Note that NUTS is the French acronym for Nomenclature of Territorial Units for Statistics used by Eurostat. In this nomenclature NUTS1 refers to European Community Regions and NUTS2 to Basic Administrative Units, with NUTS3 reflecting smaller spatial units similar to counties in the USA. Local inflation rates are computed on the basis of surveys conducted in the main cities of NUTS3 regions.

11. The intuition being that small shops cannot stay on the market in the presence of economies of scale, which are present in the retail sector (see, for instance, Betancourt & Malanoski, 1999).

12. Although they have the expected signs.

13. The Ellison and Ellison test proved to be more successful than other nonparametric tests in detecting functional misspecification (see also Miles & Mora, 2003). Ellison & Ellison (2000) mention the possibility of using their test also to detect the absence of omitted variables.

14. For instance, we could not control for money growth, which may differ among cities possibly owing to either credit market segmentation and credit rationing or to different stages of development of the credit system, or to different liquidity preferences of lenders and borrowers (Dow & Rodriguez-Fuentes, 1997). Data on credit for Italian LLMAs exist but they are based on the location of banks and not of borrowers. Inserting the average growth rate of this variable between 1998 and 2002 into our regressions would not have returned a significant t-statistic. Further details are available from the author upon request.

15. Descriptive statistics of these variables are shown in Table 2, Part C. In this context we carried out also the following robustness check. We estimated a model regressing long-run inflation on the share of firms with more than two employees, productivity growth in the service sector and a constant. The first regressor turned out to have a coefficient equal to –0.0036 with a p-value of 0.039.

16. Given the temporal discontinuity in unemployment data after 2002, we also tried to change the average unemployment rate between 1998 and 2002 with the average unemployment rate between 1998 and 2005. The results were robust.

17. It might be argued that our dependent variable was estimated in a first stage regression and that this may have induced heteroskedasticity. For this reason we used robust regression analysis. However, following Lewis & Linzer (2005) we also computed a weighted least squares estimator and a feasible GLS one. The results were stable, as shown by Table A2 in the Appendix.

18. In this way the system is exactly identified, having two instruments for two instrumented variables.

19. Recall that exclusion restrictions cannot be tested (see for instance, Hsiao, 1983).