the FIRE and the wholesale/retail trade sectors. But since the absolute value of the
coefficient of the initial GDP per capita is greater than one (in absolute value) for the low
initial GDP group in the FIRE sector, the rate of convergence could not be calculated for
this group.5 Therefore, we admitted that the distribution of GDP per capita displays
convergence clubs only in the wholesale/retail trade sector. This means that in the
wholesale/retail trade sector, the rich and poor economies (states) exhibit different patterns
of convergence.6
[Table 6 about here]
The unconditional growth specification is largely a descriptive tool, as it does not
account for growth conditioning factors such as labor and capital inputs. Moreover, it takes
technological progress as exogenously given, rather than explaining it in terms of factors that
stimulate the growth of technology. We therefore continue by estimating an endogenous
growth model. Our endogenous growth model is based on the initial idea of “domestic”
effects of the human capital stock on economic growth, and the role of catching up to the
technology leader as developed by Nelson and Phelps (1966), and Benhabib and Spiegel
(1994). This model has formerly been applied in Pede et al. (2006) to investigate the pattern
of economic growth in US counties over the period 1963—2003.
The model starts by a simple specification based on a Cobb-Douglas production
function, which reads as:
5 When the coefficient of the initial GDP per capita is greater than one (in absolute value), it means that there is
leapfrogging and the rate of convergence I undefined.
6 As a caution note, it should be pointed out that the Chow-Wald test may not be very powerful in detecting
convergence clubs given that we only have 49 observations.
13