Abstract: This paper proposes to combine the standard analysis method of economic convergence with spatial econometrics to explore regional convergence based on a total of 240 cities in China. To investigate the kind of spatial autocorrelation and agglomeration, the Moran's I statistic is used, finding that the existence of strongly positive global autocorrelation of GDP per capita and what's more, the local spatial structure is rather stable. The findings suggest that the non-spatial models applied to analyse β-convergence suffer from the risk of misspecification and a spatial model is competent. The results based on the spatial models indicate the existence of absolute convergence between cities. Taking into account effects results in a significant faster rate of convergence. The sensitivity test of the absolute convergence with respect to assumption of a common steady state and robustness over space suggest that the finding of absolute convergence is not stable. The mechanism of diminishing return and technology spillover is both important for absolute convergence. Finally, a set of regional policies are discussed.

Key words: spatial autocorrelation, spatial econometrics, &beta, convergence, convergence mechanism 1558