Guangdong: A catalyst for economic growth and exports in hunan province

Abstract

This paper uses the concept of Granger-causality to analyze the link between export expansion in the rapidly growing Guangdong province and GDP growth and exports in Hunan, its adjacent northwest neighbor province. Data cover the 1978 to 2001 period. A long-run equilibrium relationship is found between the variables and a long-run positive causality is detected from export expansion in Guangdong to both GDP growth and exports in Hunan. Hence, the results seem to support the unbalanced regional development policy implemented by the central government in the late 1970s and early 1980s.

Keywords:

• Spillover effects
• backward linkages
• error-correction
• China
• JEL classifications: C22
• F14
• F15
• O10
• O53

Notes

 The four SEZ are classified into two broad categories: (i) comprehensive economic system zones covering industry, agriculture, commerce, services, housing, and tourism (Szhenzhen and Zhuhai), and (ii) primarily export processing and tourism-oriented zones (Shantao and Xiamen) (see Prybyla, 1986).

 In particular, there are large differences in annual growth rates between provinces in the respective regions.

 At this time the coastal provinces together accounted for almost 25% of China's industrial output. The gross output value of agricultural products was also higher in Guangdong.

 Share of rural population in 1978: Guangdong, 84%; Hunan, 89%.

 This idea was widely used as a planning tool in developing countries during the import-substitution era in the 1950s and 1960s. Reidel (1976) criticized Hirschman's view on backward linkages for implicitly assuming that demand creates its own supply. Instead, Reidel argued that in a country engaged in international trade, such an assumption would be naive.

 See also Naughton (1999) for issues on inter-provincial trade barriers in China.

 To increase the degrees of freedom one could increase the frequency of the data by using quarterly or monthly data. This, however, is not possible because such data are not available.

 Each time-series was estimated by OLS and checked for parameter constancy by (i) plotting the calculated 1-step residuals including error bands of

around zero, and (ii) performing a scaled recursive Chow-test. Neither test violates the assumption of parameter constancy. The graphs are excluded due to lack of space and are available upon request to the author. Note that the models are estimated as if they were stationary, i.e. without considering unit roots.

 ADF equation:

, where k is the lag length of the dependent variable, ▵ is the first difference operator, and ε _t is assumed white noise. Note that the ADF-test is a single-wise unit root test.

 Critical values (MacKinnon, 1991) at 5% and 1%: level form; −3.011 and –3.785 with a constant, and; −3.645 and –4.469 with a constant and trend. In first difference form; −3.02 and –3.807 with a constant, and; −3.659 and −4.5 with constant and trend.

 To determine the lag length of the VAR, three versions of the system were initially estimated: a three-, two-, and a one-lag version. The AIC, SBC, and a likelihood ratio test were used to test that all three specifications were equivalent. All tests rejected the null hypothesis that all the specifications were equivalent. The tests suggested that a VAR with two lags, i.e. a VAR of order 2, should be used in the cointegration procedure. The different test statistics are: VAR 1: Log-likelihood = 163.33, SBC = −12.002, AIC = −13.203. VAR 2: Log-likelihood = 171.69, SBC = −12.567, AIC =  −14.352. VAR 3: Log-likelihood = 161.64, SBC = −11.735, AIC = −13.683. Hence, the VAR with the largest (absolute) value of the AIC and SBC criterion is chosen.

 λ _i are the estimated values of the characteristic roots (eigenvalues) obtained from the estimated π matrix.

 The t-values are: 0.565; −2.938, and; 1.068 for α ₁, α ₂ and α ₃respectively.

 The graphs are excluded due to lack of space and are available, upon request, from the author.

 The Granger-causality can be exposed either through the statistical significance of: (i) the lagged ECs (η _i ) by a t-test; (ii) a joint test applied to the significance of the sum of the lags of each exploratory variable (ϕ, ϕ, γ) in turn, by a joint Wald χ ² test; or (iii) a joint test of all the set of terms described in (i) and (ii) by a joint Wald χ ² test, i.e. taking each of the parenthesized terms (ϕ, η) and (γ, η) in the first equation; (ϕ, η) and (γ, η) in the second equation, and; (ϕ, η) and (ϕ, η) in the third equation. See, for example Ahmad (2001) for details about Granger-causality.

 The VECM was estimated by Full Maximum Likelihood (FIML) using two lags. A number of diagnostic tests of this VECM are reported in the appendix (Table A.2). There is no evidence of serial correlation or heteroscedasticity in the residuals. Testimony of deviations from normality does not appear either.

 The estimated coefficients are presented in appendix (Table A.1).