it similarly imposes a single autoregressive parameter over all units in the
panel. However, this test employs the Zellner’s seemingly unrelated regres-
sions (SURE) estimator (one equation for each cross-section). Given that in
standard SURE models T must exceed N, this test cannot be applied to pan-
els where the cross-sectional dimension is greater than the time dimension.
As such, it is maybe more suited to macro-econometric time series. The LLC
test, on the other hand, does not have this “limitation”, and is more suited,
contrary to our case, for small-T, large-N panels.
The ST test involves the hypothesis, for each equation, that the sum of
the coefficients of the autoregressive polynomial is unity. The null hypothesis
consists of the joint test that this condition is satisfied over the N equations.
Hence, under the null hypothesis, all of the series in the panel are non-
stationary stochastic processes. The asymptotic properties of the statistic are
unknown. Hence, Taylor and Sarno (1998) provide response surface estimates
of the 5% critical values, derived from Monte Carlo simulation.9 The main
advantage of this procedure is that, unlike the previous ones, using SUR,
it take into account the cross-sectional dependence of the errors. This is
particularly important in our analysis, where it is very likely that shocks are
connected across regions, and spillovers may have the effect of increasing the
process of convergence among some regions and divergence among others.
An important caveat of this test is that the null hypothesis can be rejected
even if one of the series in the panel is stationary. Hence, rejection of the null
cannot be taken as conclusive indication that each of the series is stationary.
4 Empirical Implementation
4.1 Total Factor Productivity
The Italian Statistical Office (ISTAT) has recently provided the national time
series for the period 1993-2003 of TFP. However, at regional level no official
9The response surface was estimated over sample sizes ranging from 25 to 500 obser-
vations per cross-sectional unit.
10