Regional Per Capita Income Growth: A Comparative Analysis

The study empirically evaluates the impact of globalisation on regional per capita income growth by taking into account the spatial correlation. The physical investment and human capital are used as the control variables in the empirical analysis, which have been emphasised in Barro and Sala-i-Martin (1995), Islam (1995), Mankiw et al., (1992), Solow (1956), etc. Human capital affects economic growth by improving productivity (Schultz 1975; Welch 1970; Romer 1990; Benhabib et al. 1992; Lucas 1988; Kremer and Thompson 1993). Productivity growth has a significant relationship with the quality of human capital, through the technological competence of the workforce. One and the same technology can be applied in two different firms, but the output

The sign of spatial correlation is issue specific. For instance, in the context of economic growth, the spatial correlation is expected to have a positive effect. However, in some cases, for instance the location of investment, the correlation could be negative or positive. The location of investment in one region may affect positively due to the effects of agglomeration or spill-over. This relation may be negative on the other hand, because the relatively strong business environment of a region, reduces the location of investment in its neighbouring regions. These kind of spatial interaction effects can be controlled through spatial regression models (Belotti et al 2016). Spatial autoregressive (SAR) model and Spatial Durbin model (SDM) are commonly used in regional economic development. The SDM model takes account the spatial effect of both dependent and independent variables (Belotti et al 2016). The SAR model is the spatial case of the SDM model, which takes into account the spatial effect of the dependent variable only (Ansellin and Bera, 1998). The panel representation of the SAR or spatial lag model can be specified as

ힺ is the spatial autoregressive coefficient or the coefficient of the spatially lagged dependent variable. This indicates the magnitude to which the income growth in one region is determined by the behaviour of its neighborhood. The sign of the value of the ힺ parameter indicates the sign of the spatial effects. The error term ε_it is again assumed normally distributed and independent of all the regressors, under the assumption that all spatial dependence effects are captured by the spatially lagged variable. Corresponding to the dynamic panel GMM estimator in equation (2), the dynamic SAR Model can be specified as follow (Baltagi et al. 2014).

Y_it=αY_(it-1)+ρ∑_(j=1)^n▒w_ij Y_it+δX_it+λZ_it+µ_i+ε_it (4)

This model can also be estimated by the difference GMM and system-GMM approaches like the non-spatial dynamic panel model. The study applies system estimation in the spatial data framework due to its advantage over the difference GMM as discussed before. The study applies both SAR and SDM specifications of dynamic panel GMM system