Testing Stochastic Convergence across Chinese Provinces, 1952–2008

Abstract

Herrerías M. J. and Ordóñez Monfort J. Testing stochastic convergence across Chinese provinces, 1952–2008, Regional Studies. This paper analyses stochastic convergence across twenty-eight Chinese provinces over the period 1952–2008 using unit root tests that take into account structural breaks and non-linearities. The results suggest a significant degree of convergence in capital intensity, labour productivity and total factor productivity in China's provinces. However, the findings also reveal that provinces located in coastal areas of China have benefited from the economic reforms more than the central and western regions and have created small regional clusters.

Herrerias M. J. and Ordóñez Monfort J. 检验中国各省的随机收敛，1952年至2008年，区域研究。本文运用纳入结构性断裂与非线性考量的单位根检验，分析中国二十八个省份在1952年至2008年间的随机收敛。研究结果指出，中国各省在资本密集度、劳动生产力与总要素生产力方面具有显着程度的收敛。但研究发现亦显示出，位于中国沿海地区的省份较中、西部省份而言，自经济改革受惠更多，并且创造了小型的区域集群。

Herrerias M. J. et Ordóñez Monfort J. Tester la convergence stochastique à travers les provinces chinoises, de 1952 à 2008, Regional Studies. Employant des tests de racine unitaire qui tiennent compte des discontinuités et des non-linéarités, cet article cherche à analyser la convergence stochastique à travers vingt-huit provinces chinoises pour la période allant de 1952 jusqu'à 2008. Les résultats laissent supposer un taux de convergence non-négligeable de l'intensité du capital investi, de la productivité du travail et de la productivité globale des facteurs dans les provinces chinoises. Toujours est-il que les résultats laissent voir aussi que les provinces situées dans les zones côtières de la Chine ont profité des réformes économiques plus que les régions centrales et occidentales et ont créé de petits clusters régionaux.

Herrerias M. J. und Ordóñez Monfort J. Überprüfung von stochastischer Konvergenz in chinesischen Provinzen, 1952-2008, Regional Studies. In diesem Beitrag analysieren wir mit Hilfe von Wurzeltests unter Berücksichtigung von strukturellen Brüchen und Nicht-Linearitäten die stochastische Konvergenz in 28 chinesischen Provinzen im Zeitraum von 1952 bis 2008. Die Ergebnisse lassen auf ein signifikantes Ausmaß von Konvergenz hinsichtlich der Kapitalintensität, Arbeitskräfteproduktivität und Gesamtfaktorproduktivität in den chinesischen Provinzen schließen. Allerdings geht aus den Ergebnissen auch hervor, dass Provinzen an der chinesischen Küste von den Wirtschaftsreformen stärker profitiert haben als die zentralen und westlichen Regionen und dass sie kleine regionale Cluster gebildet haben.

Herrerias M. J. y Ordóñez Monfort J. Evaluación de la convergencia estocástica en provincias chinas, 1952–2008, Regional Studies. En este artículo analizamos la convergencia estocástica en veintiocho provincias chinas durante el periodo de 1952 a 2008 usando tests de raíces unitarias que tienen en cuenta los recesos estructurales y no linealidades. Los resultados indican un grado significativo de convergencia en la intensidad de capital, la productividad laboral y la productividad total de los factores en las provincias de China. Sin embargo, los resultados también indican que las provincias chinas ubicadas en áreas costeras se han beneficiado más de las reformas económicas que las regiones centrales y occidentales y han creado pequeñas aglomeraciones regionales.

• China
• Stochastic convergence
• Regional growth
• Non-linear unit root test

• 中国
• 随机收敛
• 区域成长
• 非线性单位根检验

• Chine
• Convergence stochastique
• Croissance régionale
• Test de racine unitaire non-linéaire

• China
• Stochastische Konvergenz
• Regionales Wachstum
• Nichtlinearer Einheitswurzeltest

• China
• Convergencia estocástica
• Crecimiento regional
• Test de raíces unitarias no lineales

JEL classifications:

• C20
• O18
• O40
• R11

Acknowledgements

The second author gratefully acknowledges the financial support of CICYT Project Number ECO2011-30260-C03-01 and Generalitat Valenciana Project Number PROMETEO/2009/098. Both authors acknowledge the comments received from two anonymous referees.

Notes

1. For the need to balance the sources of growth in China, see Aziz (2006) and Prasad (2009).

2. This is probably due to the use of traditional unit root tests that do not consider structural breaks or a non-linear form in the stochastic process. The use of unit root tests that incorporate one of these two possibilities increases the number of regions that converge in the reference area (Strazicich et al., 2004; Christopoulos and Tsionas, 2008).

3. For an application of this notion of convergence for G7 countries, see also Cheung and Garcia (2004).

4. For recent surveys, see Temple (1999), De la Fuente (2000) and Islam (2003).

5. For example, see the following works using either σ- or β-convergence: Rozelle (1994), Jian et al. (1996), Chen and Fleisher (1996), Raiser (1998), Yao and Zhang (2001a, 2001b), Weeks and Yao (2003), Wang (2004), and Pedroni and Yao (2006), among others. On the other hand, only three papers on Quah's approach are found, namely: Bhalla et al. (2003), Sakamoto and Islam (2008), and Herrerías et al. (2010).

6. The authors highlight the fact that unit root tests that do not consider structural breaks give rise to misleading results if structural breaks are present in the data. However, the same rationale might apply if researchers made use of unit root testing with one structural break, and the data-generating process yielded more than one structural break.

7. They are expressed in real terms, based on the regional GDP deflator. The base year is 1953 = 100. In addition, for simplicity, the concept of provinces is used throughout this paper. However, in China there are twenty-three provinces, five autonomous regions, four municipalities and two special administration regions (SAR). Hainan and Tibet have been excluded due to the lack of data. In addition, this paper focuses on mainland China, and consequently Taiwan, Hong Kong and Macao have also been excluded. Chongqing is included in Sichuan province, given that it was part of it until 1997. This is standard practice in Chinese studies. In relation to this, there is much debate as to the quality of the Chinese statistics. However, Holz (2005), Chow (2006) and Bai et al. (2006) present quality Chinese statistics for examining long-run trends. Data were used here from National Bureau of Statistics (NBS) (2010), which provided information that was homogeneous enough, both across Chinese provinces and over time, to perform this study properly. The authors would like to thank Yanrui Wu for the data on capital stock; for further details in his work, see Wu (2008). There is a large body of research that has covered the pre- and post-reform periods in Chinese studies, using unit root test and co-integration techniques (for instance, see Chow, 1987; Li, 2000; Herrerías and Orts, 2010a, among others).

8. The subscripts have been omitted for the sake of simplicity.

9. The aim of this paper is not to obtain the precise estimates of TFP or to partake in the debate on TFP measurement. The objective is simpler and consists mainly in obtaining an approximation of the contribution of capital to the growth rate in order to reach TFP and to perform analyses with the other variables considered in this study. For a similar application of the measurement of TFP, see Herrerías and Orts (2010b).

10. The estimations are available from the authors upon request.

11. Y_t follows a non-linear but globally stationary process provided by –2 < γ < 0.

12. The authors thank the anonymous referee for this suggestion.

13. The inclusion of the Kapetanios et al. (2003) test seems unnecessary given that the Christopoulos and León-Ledesma (2010) test captures non-linearities as well as structural breaks. However, the Kapetanios et al. test has been applied not only to check for the existence of non-linear stationarity, but also with the idea of distinguishing between two degrees of convergence (long-run convergence and catching up). This distinction depends on the significance of the deterministic components in the unit root test. The Kapetanios et al. test allows for linear deterministic trends, whereas the Christopoulos and León-Ledesma test allows for non-linear trends. The difference in the specification of the deterministic components can thus affect the conclusions regarding the type of convergence dynamics observed. The authors thank an anonymous referee for pointing out this issue.

14. The authors thank an anonymous referee for raising this question.

15. The results are similar for K/L and TFP. All results are available from the authors upon request.

16. Lee and Strazicich (2003) reject the null frequently without incorporating non-linearities, whereas the Christopoulos and León-Ledesma (2010) test rejects the null allowing for non-linearities. A reasonable question arises. Which are the relative merits of both tests? In the present authors' view these tests are complementary rather than substitutes. It is believed that by using them jointly one has a deeper insight into the nature of dynamic behaviour of the series. It is difficult to compare the relative merits of both given that the null is different and the procedure to test for the null is completely different as well. However, something can be said if one looks to the power of both tests. The Christopoulos and León-Ledesma test has good power when the residual for the Fourier expansion, equation (10), is close to linear (Christopoulos and León-Ledesma, 2010, table 5). This implies that it has good properties if the series is linear with breaks. The Lee and Strazicich test has good power also when the process is linear with breaks, however the power decreases when two breaks are assumed under the null, but the true data-generating process (DGP) has no break. This is caused by the inclusion of dummy variables under the null trying to capture breaks that do not exist.

17. The authors thank an anonymous referee for raising this question.