Dataset and methodology
To measure economic growth, we consider historical annual data of real per capita GDP (y) in 1990 International Geary‐Khamis dollars. The main source is:
• Historical Statistics for the World Economy on Angus Maddison database: www.ggdc.net/maddison/Historical_Statistics/horizontal‐file_03‐2007.xls
To measure China's regional income inequality, we consider annual data of Gini coefficient (i). Our main source is:
• Kanbur and Zhang (2005). While the data of the years 2001‐2007 was computed through growth rates of Gini coefficient and the source was the National Bureau of Statistics of China: www.stats.gov.cn/eNgliSH/statisticaldata/yearlydata/
Because most macroeconomic variables are trended, time series can potentially create problems of finding spurious regressions when they are non‐stationary, (see Phillips, 1986 for an analysis of spurious regressions). Classical econometrics is not applied when process is non‐stationary and cointegration method should be applied. Therefore, as a first step we have to study the integration order of the series in order to applied cointegration method. One method is the two‐step procedure proposed by Engle and Granger (1987). However, this method assumes the existence of only one cointegrating relationship. Most general procedure was proposed by Johansen (1988) and Johansen and Juselius (1990), this test has the advantage of testing all the possible cointegration relationship.
Banerjee et al. (1993) highlight the important connection between a cointegration relationship and the corresponding long‐run equilibrium equation. Studying a cointegration relation is analyzing a statistical equilibrium between variables tending to grow over time. The discrepancy of this equilibrium can be modeled by a vector error correction (VEC) model which shows how after a shock the variables come back to the equilibrium. Gobbin and Rayp (2008) pointed out that a cointegrated VAR‐setting approach is the proper way to cope and avoid the problems of parameter heterogeneneity, omitted variable bias and endogeneity of the variables.