variation indicates the dispersion of productivity levels while the Moran’s I statistic is a
measure of spatial autocorrelation. In Figure 2, positive values of Moran’s I are observed in
all sectors, denoting that states with similar productivity levels are spatially clustered.
However, the degree of spatial clustering varies between sectors. A persistent decline of
Moran’s I statistic is observed in the manufacturing and construction sectors from the early
1970. This is indicative of spatial defragmentation within these sectors over time, which may
have been caused by similarity in term technology or geography. In the other sectors, the
Moran’s I statistic is rather stable or follows an irregular trend. The mining sector shows a
good example of relatively stable Moran’s I statistics over time, which may be explained by
the fact that this industry is not footloose. The trend of the coefficient of variation of the
states’ GDP per capita is not similar across sectors either. A declining trend of the
coefficient of variation denotes σ-convergence,3 which suggests that the disparities of GDP
per capita across states are becoming less pronounced. Figure 2 shows no real evidence of σ-
convergence in any of the sectors considered. None of the sectors shows a steady decline of
coefficient of variation over the period 1969—1997. However, a relatively stable GDP
dispersion is observed in the construction and government sectors.
[Figure 2 about here]
4. Endogenous growth model with technological leadership
This section starts with the estimation of an unconditional convergence model in the
tradition of Barro and Sala-i-Martin (1991) for each of the sectors. The unconditional growth
3 The presence of β-convergence is a necessary although not a sufficient condition for the occurrence of σ-
convergence (Barro and Sala-i-Martin 1995).
10