A robust analysis of the evolution through time of regional unemployment disparities
requires an ample number of observations. Pooled estimation enables us to use 176 and
96 observations for the high and low unemployment rate panels, respectively. The pooling
of observations on a cross section of regions over several time periods can increase the effi-
ciency of econometric estimates.23 Within each group of regions, our estimation captures
differences in economic behaviour solely through fixed effects (i.e. differing constants in
the estimated equations), while the coefficients for the explanatory variables are taken to
be identical across groups of regions.
The data sources are (i) Datastream, (ii) the BD-MORES dataset, elaborated by the
Direccion General de Analisis y Programacion Presupuestaria (Ministry of Economy) and
the University of Valencia, and (iii) the Spanish Labour Force, elaborated by the Spanish
Statistics Institute (INE). The sample frequency is annual and the period of analysis is
1980-1995, due to data limitations.24 Table 2 gives the definitions of the variables.
Table 2: Definitions of variables
Regional variables lit : labour force uit : unemployment rate (= lit — nit) wit : real wage (=labour income per employee) kit : real capital stock popit : working age population |
National variables oilt : real oil price bt : real social security benefits per person taxt : indirect tax rate impt : real import prices |
All variables are in logs except for the unemployment rate uit, and the indirect tax rate, taxt . |
Dynamic panel data and nonstationary panel time series models have attracted a lot
of attention over the past few years. As a result, the study of the asymptotics of macro
panels with large N (number of units, e.g. countries or regions) and large T (length
of the time series) has become the focus of panel data econometrics.25 We test if it is
appropriate to use stationary panel data estimation techniques by performing a series of
unit root tests.
In particular, we test the order of integration of the national variables using the KPSS
unit root test.26 Table 3 presents these tests and shows that for all four national variables
23 The advantages of using panel data sets for economic research are numerous and well documented in
the literature. See, for example, Hsiao (1986) and Baltagi (1995) for a detailed exposition of stationary
panel data estimation.
24 The reason for restricting our analysis to the 1980-1995 period is twofold. First, the regional capital
stock series are obtained from the BD-MORES dataset which currently covers the 1980-1995 period and
is expected to be updated for the period 1980-2000. Second, in 2001 the Spanish Statistics Institute
(INE) introduced fundamental changes in the Labour Force Survey (mainly related to the definition of
labour force) in order to make the survey comparable to the Eurostat standards. The induced structural
break in the labour force and unemployment rate series implies that the figures for these series are not
compatible to the ones prior to 2001.
25 Banerjee (1999) and Baltagi and Kao (2000), and Smith (2000) provide an overview of the above
topics and survey the developments in this technical and rapidly growing literature.
26 Kwiatkowski-Phillips-Schmidt-Shin (1992) proposed the following statistic to test the null hypothesis
of stationarity:
VT , s2
KPSS (κ) = T≡t=⅛⅜'
11