Divergence and long-run equilibria in Italian regional unemployment

Abstract

The Italian labour market is characterized by large and persistent regional unemployment differentials. This study uses recent panel unit root and cointegration tests to derive the long-run properties of the Italian regional unemployment disparities. The empirical evidence suggests that the stochastic convergence hypothesis can be rejected. However, the existence of a long-run equilibrium relationship among regional unemployment rates cannot be excluded. Some possible implications of this finding are sketched.

Acknowledgments

We would like to thank, without implication, Joakim Westerlund for comments. The opinions expressed in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of ISAE or its staff.

Notes

¹ A full understanding of the nature of unemployment differentials cannot disregard the study of the role of institutions (Brunello et al ., 2000, 2001), of social factors (Di Pietro and Urwin, 2003; De Paola et al ., 2005), and the impact of migration (Faini, et al ., 1997; Böckerman, 2004; Huber, 2004).

2 A preliminary analysis is carried out on the properties of the single series. Individual ADF Unit root tests are performed. The results show evidence in favour of the unit root hypothesis for all the series.

3 An alternative would be to consider regional differentials with respect to a national average. In the light of Martin's (1997) criticisms to this approach, it was decided to consider an explicit regional benchmark.

4 Arbia and Costantini (2006) use SUR (Seemingly Unrelated Regression) and Multivariate ADF (Taylor and Sarno, 1998) panel tests in order to take into account cross-sectional dependence in studying stochastic convergence of per-capita GDP among Italian regions. Österholm (2004) uses the Multivariate ADF test to investigate the presence of unit root in the unemployment rates in the US.

5 Gutierrez (2003) shows that Moon and Perron's (2004) tests have good size and power in finite samples for different specifications and different values of T (the time dimension) and N (the unit dimension), and that Bai and Ng's (2004) pooled tests of the null hypothesis that idiosyncratic components are non-stationary also have good size and power, especially when the Dickey-Fuller-GLS version of the test is considered, while the ADF test used to analyse the nonstationary properties of the common component has relatively low power. On the contrary, Choi's (2006) tests seem to be largely oversized. However, Gutierrez (2003) also shows that all the tests lack power when a deterministic trend is included in the data generating process.

6 With the exception of the tiny autonomous region Valle d’Aosta. Data are taken from ISTAT: ‘Forze Lavoro, Dati congiunturali, Capitolo 3, Dati regionali’.

7 The North-Central regions considered in this study are: Lombardia, Piemonte, Trentino Alto Adige, Veneto, Friuli Venezia-Giulia, Liguria, Emilia-Romagna, Toscana, Umbria, Lazio, Marche, Abruzzo. This classification differs slightly from the traditional one, where Abruzzo is considered among the Southern regions.

8 In this test the cross-sectional dependence is represented by a factor model in which the errors of Equation 1 are generated by both region-specific shocks and unobservable factors that are common across the members of the panel. See Westerlund (2005a) for details.

⁹ For further details on these procedures see Bai and Kao (2005) and Westerlund (2005b).