Poverty dynamics in rural China revisited: do assets matter?

Abstract

This paper uses an asset-based approach to examine poverty dynamics in rural China over the period 1989–2006. The analysis documents a significant structural component in the poverty dynamics of households. The lack of profitable agricultural asset accumulation plays an unneglectable role in causing households to be trapped in persistent poverty. The escape from poverty is increasingly dominated by stochastic upward mobility rather than by structural movement in terms of asset accumulation. This could threaten the prospect of poverty reduction in rural China. It is argued that future reform and policy-making should pay more attention to building households’ asset base.

Keywords:

• poverty dynamics
• poverty transition
• persistent poverty
• agricultural assets
• rural China

JEL Classifications:

• I32
• O12

Acknowledgements

The author is grateful to the insightful comments on an earlier draft from two anonymous referees, Albert Park, Katsushi Imai, Adam Ozanne, Bernard Walters, Xiaobing Wang, Nick Weaver and the seminar participants at the University of Manchester. This work was supported by the Ministry of Education of China (MOE) Project of Humanities and Social Sciences [Grant No.: 13YJCZH231] and the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry. The financial support from Brooks World Poverty Institute and the North American Foundation for the University of Manchester at the early stage of this research is also gratefully acknowledged.

Notes

1. Data come from Poverty Monitoring Report of Rural China 2008 published by the National Bureau of Statistics (NBS).

2. Author’s calculation based on data from Poverty Monitoring Report of Rural China 2008 published by the NBS.

3. The share of agricultural income was 70% among sample household in 1989. This proportion declined over time along with income diversification and more off-farm opportunities (typically rising migration from rural to urban areas), but still accounted for more than 50% of household income. Agricultural income is particularly important for the poor as many of them are unable to engage in out-migration (Du, Park, and Wang 2005). The share of agricultural income in our sample is on average 7 percentage points higher for the poor than for the non-poor.

4. The CHNS used a multi-multistage, random cluster process to draw 3795 sample households covering 15,917 individuals in both rural and urban China from nine provinces in 1989. Rural households were defined as those who reported they permanently lived in home villages at the time of interview and were tracked by the “dwelling” rule in subsequent waves. Consistent questionnaires were used throughout different waves. Specific sampling methods and survey conduction can be found at http://www.cpc.unc.edu/projects/china [accessed July 28, 2013]. The discussion on the representativeness and appropriateness of using the constructed panel at the household level based on the CHNS can be found in Imai and You (2013).

5. This definition follows Benjamin, Brandt and Giles (2005) and is recently applied by Imai and You (forthcoming) to the CHNS in studying dynamics of household poverty. Due to data limitations, we cannot include more reliable sources of consumption expenditure. To verify our consumption data, we compared them with the Rural Household Surveys (RHS) collected by the NBS and find similar mean consumption for each sample province and trends of consumption changes over time. Our constructed household consumption has similar mean and time trend over the sample period of that of the RHS. Furthermore, as we will show in Figure 1, our constructed consumption also generates consistent poverty measures with past literature. Therefore, the validity of our consumption data could be reasonably believed.

6. Farm machines include large or medium-sized tractors, walking tractors, animal carts, irrigation equipment, power threshers and household water pumps. Land owned by households is not included, since in rural China land is allocated equally by local/village officials according to the number of household members. In our sample, the average household per capita farmland is very stable and lies between 0.79 and 0.96 in different waves. Land is not a tangible asset that households can accumulate or divest easily. That said, we do experiment with land when constructing the asset index. The shape of agricultural asset dynamics holds broadly same as before.

7. The constructed asset index is essentially an ordinal concept and thus can be either positive, negative or zero.

8. The only exception is the consumption poverty measured against the Chinese government’s official poverty line.

9. This estimate does not mean that 15% of the sample households were always rich over the period 1989–2006 as there were 2–3 years gap between two consecutive waves. It should better be understood as an upper estimation of the size of the always non-poor population. Similar interpretation also applies to the 3% persistently poor in the next sentence.

10. If using the Chinese official poverty line which is approximately 80% of the adjusted US$1.25/day, more people (32%) did not experience poverty in any wave and less people (1%) were always poor in every wave. Sixty-eight per cent fell in poverty at least once and 49% at least twice. Persistence of poverty becomes less severe as a result of low poverty line, while frequent transitions are still salient.

11. It should be noted that McCulloch and Calandrino’s (2003) framework is open to weakness. We may have overstated the transitory poverty by using their methodology. The limitation of their method is that their transitory poverty calculations do not reflect the case that many households may escape as a result of increasing mean consumption. Even if these increases were perfectly linear and steady, they would show up as transitory poverty.

12. Carter and May (2001) caution that cannot be used directly to classify whether the household is stochastically poor, because it contains genuine entitlement failures as well as other unobserved disturbances, such as measurement errors. A less precise from estimating Equation (2(2) ) would exaggerate and in turn overstate the stochastically poor. To improve the accuracy of the estimate of the stochastically/structurally poor and make the estimates less sensitive to the performance of Equation (2(2) ), they suggest an alternative test of one-side hypothesis presented in the following paragraph.

13. Carter and May (2001) note that it is impossible to further reduce this bound for the structurally poor. We will return to this point in Section 5.

14. Refer to Appendix 2 for the IV estimation results of the household livelihood regression (Equation 2(2) ).

15. Data come from Poverty Monitoring Report of Rural China 2008 published by the NBS.

16. One may concern that a household is likely to be classified as “structural poor” not because its failure in asset accumulation but simply divesting its wealth to smooth consumption. This is not a serious problem in our case. Following Barrett et al. (2006), we plot the volatilities (measured by coefficients of variation) of household per capita income and consumption over household intertemporal agricultural asset deciles. We actually find asset smoothing rather than consumption smoothing behaviour for the poorest 12.5% of households in the asset distribution as they have larger consumption variability compared with that of income. Extremely poor households have to defend their limited productive assets to survive rather than conducting consumption smoothing, e.g. rural Zimbabwe (Hoddinott 2006) and Kenya (Barrett et al. 2006). They would be classified as “structurally poor” if they encounter accumulation failures as well. By contrast, those lying above 40% in the asset distribution in general suggest consumption smoothing behaviour as their consumption variability is smaller than that of income. However, these households are less likely to be misjudged as “structurally poor” as they are relatively wealthy in the asset distribution even after using their wealth as a buffer and therefore can afford consumption smoothing.

17. At the regional level, Lang, Barrett, and Naschold (2013) combine estimated returns to various assets for geographically defined sub-groups and traditional poverty maps to better target candidates for policy intervention in terms of asset transfer schemes.