4.Results and Exploration of Empirical Analysis
4.1 Effect Test of Poverty-Reducing Factors
We estimated the poverty-reducing effects of several factors through regression; the results of the three methods are reported in Table 2. Specifically, columns (1) and (4) are based on the mixed least square method and dummy variables of province and time; columns (2) and (5) are based on the two-way fixed effect model and checked for robustness through heteroscedasticity, serial correlation, and cross- sectional correlation using xtscc command in Stata; and columns (3) and (6) are based on the generalized method of moments (GMM), which is applicable to models containing multiple endogenous variables. Given the different measurement units and dimensions of variables, we used the standardized regression coefficient to compare the size of coefficients without changing their symbol and significance. It should be noted that several types of specification have been made in GMM estimation according to actual circumstances in order to overcome the potential problem of endogeneity in the model. The socio-economic development variables are specified as endogenous variables because of their reverse causality with poverty and potential influence from certain unobservable common factors. The government fiscal intervention variables are specified as predetermined variables considering that the government tends to formulate a spending plan for the next phase according to poverty status in the previous phase in order to distribute fiscal resources. Moreover, we adopt the climatic and geographical variables of annual mean temperature, annual mean precipitation, annual mean sunshine duration, average altitude, and relief amplitude7 as purely exogenous instrumental variables (IV). As can be learned from the comparison, there is no significant difference between the estimated results of POLS and FE in terms of the size and significance of coefficients, but certain differences exist in GMM results. Therefore, endogeneity does indeed exist in the original model, necessitating the selection of IV and treatment with GMM results. Hence, the analysis and interpretation of the regression coefficients are also based on the estimated results of GMM. The Arellano-Bond (2) tests in columns (3) and (6) cannot reject the null hypothesis of no autocorrelation of disturbance term εit, and the Hansen test also shows that the instrumental variables have passed the over-identification test. Therefore, the choice of GMM estimation is appropriate.
(i) Estimated results for the period from 2000 to 2012. Economic growth, as shown in column (3), had a broad anti-poverty effect. The significantly negative coefficients for non-farm employment and infrastructure imply that economic growth created non-farm jobs and improved infrastructure, hence creating opportunities for poverty reduction. Transformations in various social structures and systems reflect differentiated poverty-reducing effects, and the coefficients of industrial structure and income gap are significantly positive, indicating that industrial upgrading and the urban-rural income gap were unfavorable to poverty reduction during this period. Both urbanization and marketization
have significantly negative coefficients and are conducive to poverty reduction by providing the poor with information and opportunity to generate greater income. In the public fiscal spending of government, the coefficient of compulsory education is significantly negative, demonstrating the effectiveness of the dropout protection mechanism; the coefficient of higher education is insignificant, which can be attributed to problems in the issuance of scholarships (Chen et al., 2013) and a possible lag effect in poverty reduction through spending on higher education; the significantly positive effect of vocational education reflects the poor quality of vocational education. The coefficient of social security spending is insignificant, implying that such transfer payments have failed to aid the poor. The insignificant coefficient of fiscal decentralization means that the current tax sharing and performance evaluation systems provide insufficient poverty reduction incentives for local government fiscal spending.
(ii) Estimated results for the period from 2013 to 2018. According to the column (6), none of the three economic growth variables are significant, which shows the diminishing trickle-down effect of economic growth. The significance of industrial structure and income gap has also diminished, meaning that neither of them has any significant adverse impact on poverty reduction. There is an increase in the poverty- reducing effect of urbanization, implying that new- type urbanization has benefited the deeply poor regions left out from traditional urbanization. The poverty-reducing effect of marketization is no longer significant, suggesting a decline in its pro-poor effect. A probable reason is that marketization created shocks to the otherwise poverty-reducing effect of private social network capital and other informal systems. Among the three variables of educational spending, the coefficient of compulsory education is no longer significant, which explains that the marginal poverty-reducing effect of compulsory education diminished after the universalization of nine-year compulsory education; the coefficient of vocational education is no longer significant either, implying that vocational education is no longer unfavorable for poverty reduction; the coefficient of higher education becomes significantly negative, suggesting that college scholarships alleviated the financial burden of students from poor households and made it less likely for them to drop out. In fact, most students from poor households who received college scholarships had benefited from the universalization of compulsory education, reflecting the long-term and continuous nature of poverty reduction through education. There is no significant change in the coefficients of social security spending and fiscal decentralization, which means that their problems persisted during this period.
4.2 Contribution Decomposition of Poverty-Reducing Factors
The preceding section provided an interpretation of the poverty-reducing effects of numerous factors based on estimated GMM coefficients, which depicts an ideal scenario assuming that other variables remain constant. To determine the real effect of each variable on poverty reduction, it is necessary to substitute the coefficients into equation (1) to arrive at the final model, and use actual data for each variable to estimate their contribution to poverty incidence. To this end, we used two decomposition methods: Difference and variance decomposition. While difference decomposition is a factor decomposition of the change in poverty incidence over time, variance decomposition is an itemized decomposition of the inter-provincial variance in poverty incidence over time. Both types of decomposition can be utilized to quantify each factor’s contribution to poverty reduction. While the former focuses on the causes of lowering poverty incidence, the latter investigates the causes of interprovincial poverty incidence convergence to zero. As a robustness analysis, it is recommended to perform variance decomposition after difference decomposition. This technique also allows to consider the reasons for inter-provincial poverty convergence and to assess the driving forces behind overall poverty reduction.
4.2.1 Difference decomposition of poverty incidence change
The difference of poverty incidence can be decomposed into the weights of the differences
of variables and residual error using the approach for economic growth accounting. Similar to the calculation of TFP for growth accounting using the Solow residual, residual error in this section reflects poverty reduction efficiency, and changes in the residual error reflect changes in efficiency. The results of difference decomposition are shown in Table 3.
Over the period 2000–2012, economic growth factors may have contributed 65.41% to poverty reduction and social development factors may have contributed 83.39%. Economic growth resulted in the creation of non-farm jobs, and marketization enabled the labor force in deprived areas to migrate to cities in quest of non-farm opportunities. Meanwhile, economic growth increased government fiscal resources, improved infrastructure in poor regions, and accelerated urbanization and the equalization of basic public services, allowing the poor to share the benefits and opportunities of development. However, the government’s public fiscal instruments on the spending side failed to effectively reduce poverty and, in some cases, resulted in resource waste (-38.68%), with compulsory education spending (-9.11%)
and social security spending (-23.25%) both contributing negatively. Fiscal decentralization (-19.72%) lessened the contribution of fiscal spending even further. Local governments prioritized infrastructure while neglecting public services and transfer payments under the fiscal decentralization system, leaving them underfunded ( Zhang et al., 2010). Due to inaccurate targeting, elite capture, crowd- out, and induced effects, the limited financial assets were subject to further losses. The aforementioned issues occurred all across the relative poverty reduction system, resulting in inaccurate poverty identification, waste of poverty reduction resources, and a disordered organizational structure, showing a negative contribution of the residual (-10.62%). Poverty reduction during this time period was ineffective, and the poverty reduction policy needed revision.
From 2013 to 2018, the trickle-down effect of China’s economic growth decreased from 65.41% to 9.43%, and the pro-poor effect of social development also decreased from 83.89% to 35.64%, owing primarily to the decreasing contribution of marketization, which shrank from 76.41% to 17.14%. The reason for this is that the countryside trailed behind cities in marketization reforms, with diminishing marginal poverty reduction effects ( Zhou and Tao, 2016), potentially causing shocks to informal institutions such as private social network capital. Furthermore, the government’s increasingly strong role has distorted market mechanisms to some extent (Shen, 2020). The negative contribution rate of industrial structure (-36.69%) indicates that upgrading industrial structure underlines the challenging nature of providing jobs for the poor, far away from supporting pro-poor industrial projects. As in the previous period, government public fiscal spending’s contribution to poverty reduction was negative (-32.89%). This suggests that even during the targeted poverty reduction period, fiscal spending on public services and transfer payments was still susceptible to leakage, and that in the future fiscal system, special poverty reduction funds need to be replaced with conventional fiscal instruments to reduce operating and management costs. Despite a minor increase, educational spending contributions remained modest (13.57%), emphasizing the need to improve the poverty-reducing effects of vocational and higher education and promote endogenous poverty reduction capabilities for the poor. Finally, the contribution of residual error increased significantly over this time span, rising from -10.62% to 87.72%. Therefore, the targeted poverty reduction approach effectively addressed the major contradictions and challenges in poverty reduction, resulting in a large boost in poverty reduction efficiency. According to the IGDS, 39.25% of respondents attributed China’s remarkable poverty-reduction achievements to “targeted and effective poverty-reduction programs and policy support”.
China’s eastern, central, and western regions all reduced poverty with the same features as nationwide. Government public fiscal spending has the biggest negative effect in the eastern region. Local governments in the eastern provinces have not given adequate attention to poverty reduction and have failed to efficiently transfer fiscal resources to the poor despite significant socio-economic development. This issue is also evident in the efficiency of poverty reduction, with the highest negative contribution from 2000 to 2012. The contribution of the income gap became negative in the central region. It should be highlighted that educational investment made only a minor contribution to poverty reduction, and this contribution became the least important after 2013 possibly because of central region’s tremendous abundance of human resources. In the western region, socio-economic development contributed the least to poverty reduction, and marketization’s effect was even negative, indicating the country’s large regional development disparity. The government needs to maximize the role of socio-economic development in future poverty reduction efforts.
4.2.2 Inter-provincial variance decomposition of poverty incidence
Variance decomposition is extensively applied in the analyses of income gaps. Based on the logarithmic form of the Cobb- Douglas production function, Klenow and Rodriguez- Clare (1997) decomposed the variance of income into the sum of the covariance between income and TFP and the covariance between income and factor input. After substituting the regression coefficient into equation (1),
poverty incidence can be expressed as the linear aggregation (including the constant term and residual error) of each variable terms (the product of the variable and its coefficient). Hence, the variance of interprovincial poverty incidence in a given year may also be expressed as the sum of covariances between poverty incidence in the current year and individual variables, and the contribution of a given variable is the covariance as a share of the variance of poverty incidence. For instance, the contribution of socio
nj economic development in year t is expressed as ΣΣ cov( Yt, βjr ESDjrt)/ var( Yt) , and the contribution of
j= r= efficiency (residual error) is cov( Yt, ε' t)/ var( Yt) 8.
According to the statistics, the variance of inter-provincial poverty incidence decreased over the years, showing that poverty gaps in various provinces were shrinking and converging toward zero; therefore, the goal of eradicating absolute poverty was almost complete. The variance in inter-provincial poverty incidence over time is then decomposed to investigate which factors contributed to interprovincial poverty convergence and which others contributed to its divergence. Figure 1 depicts the variance contributions of socio-economic development, government fiscal intervention, and adjustment in poverty reduction strategy (as reflected in its efficiency).
Figure 1 shows that, with the exception of a significant increase in 20199, the contributions of economic growth and social development to the variance of inter-provincial poverty incidence lingered around 100% in most years. Economic growth and social development were the fundamental drivers of inter-provincial poverty gaps. Specifically, the variance contribution of economic growth is smaller than that of social development and it became negative after 2013 (data available upon request) and began to promote the convergence of inter-provincial poverty incidence. However, social development inequalities among provinces continued to limit inter-provincial poverty convergence.
Except for a significant decrease in 2019 (due to the same reasons described above), the contribution of provincial government public fiscal spending to poverty variance is close to zero, meaning that provincial fiscal spending will not lead inter-provincial poverty incidence to converge. To remedy the lack of synergy between provincial governments, the central government needs to provide additional fiscal resources to specific impoverished areas.
The variance contribution of poverty reduction strategy is slightly more than that of fiscal intervention by the government. However, when compared to socio-economic development, poverty reduction strategies are not the primary cause of inter-provincial poverty differences. In recent years, its contribution rate has decreased significantly and turned negative, becoming a key factor impeding inter-provincial poverty incidence divergence and promoting its convergence to zero. This suggests that the targeted poverty reduction strategy has accelerated the pace of poverty reduction in impoverished regions, making it possible to eradicate absolute poverty by 2020.
4.3 Robustness Analysis
When actual conditions are considered in our baseline analysis, the choice of particular variables may have an unpredictable impact on the results. For instance, economic growth variables are limited to agricultural development, non-farm employment, and infrastructure. However, economic growth has many facets, and focusing on only three variables may not properly reflect their impact on poverty reduction. Besides, the choice of variables for educational spending, particularly higher education, has spatial spillover effects10. Our baseline analysis created indicators only based on local (provincial) spending in the China Education Funding Statistical Yearbook to investigate the effects of educational spending on poverty reduction in various provinces, but such selection may not fully exclude the spatial spillover effect given the recruitment of students from other provinces. Moreover, the selection of the social security spending variable. In the “general public budget spending” section of provincial statistics yearbooks, we have currently selected the “social security and employment spending” item. This component, however, comprises not only low-income allowances, but also payments for retirees from administrative and public institutions, as well as administrative expenses for civil affairs services. As a result, it may not be suitable to calculate the amount of allowances for the poor through transfer payments using “social security and employment spending”.
To answer the above three concerns and strengthen the robustness of the empirical results, we created three robustness analysis schemes by supplementing and substituting the variable design. The first robustness analysis scheme established a comprehensive variable economic growth in the baseline scenario based on variable selection to measure the level of economic growth based on provincial nominal GDP growth rates. The second robustness analysis scheme restructured the higher education
variable to include central government spending and investigate the spatial spillover effect of central fiscal allocations. The third robustness analysis scheme changed the variable social security spending to
11
exclude “pension funds of administrative and public institutions”, which account for nearly one third of total spending but are not intended to subsidize low-income people and thus have little effect on poverty reduction.
The system GMM approach is used in all the three robustness analysis schemes. The results are generally robust and show no significant deviation from the baseline regression. The newly included economic growth coefficient is not significant, showing that the three variables of agricultural development, non-farm employment, and infrastructure may include the poverty-reducing conduits of economic growth. In the second robustness analysis scheme, both the significance and absolute values of the higher education coefficient have increased, indicating that most universities that have received central government fiscal allocations have outstanding academic performance and have created spillover
effects by enrolling non-local students. As such, in order to exclude the spatial spillover effect to some extent, we exclusively used local data and excluded central government data in the formulation of indicators based on our analysis. In the third robustness analysis scheme, the coefficient of social security spending turns significantly positive, implying its leakage has aggravated after excluding pension spending for administrative and public institutions. That is, such transfer leakage indeed occurred in the allowance and relief for the low-income people rather than in administrative spending. The difference decomposition results of the three robustness analysis schemes are generally consistent with the baseline decomposition (regression estimates and difference decomposition results are available upon request).
5. Further Analysis
The eradication of absolute poverty in China does not imply that poverty reduction has been completed once and for all. Widening wealth disparities demand ongoing research into long- term strategies to deal with relative poverty while consolidating poverty-reduction results. Meanwhile, during the transition period, poverty reduction policies need to be reasonably stable. Hence, two further analyses will be carried out: First, the Foster–Greer–Thorbecke (FGT) indicator system will be established to analyze the factors of poverty depth and severity in order to draw attention to poverty gaps within poor groups. Second, the PVAR is used to analyze the lag effect and long-term influence of several factors on poverty in order to determine which policies and factors should be kept stable during the transition period and which needed to be corrected immediately.
5.1 Analysis of FGT Indicators
To better identify income distribution within the poor group, FGT indicators include the scope, depth, and severity of poverty. Following a discussion of the effects of several factors on the scope of poverty (poverty incidence), we will examine their effects on the depth and severity of poverty to determine the extent to which poor groups have benefited and to call attention to income disparities within poor groups.
Factors that significantly affect the scope or incidence of poverty will likewise affect poverty depth and severity, according to the regression results (available upon request). Poverty-inducing and poverty-reducing factors both have an increasing impact on the scope, depth, and severity of poverty. In other words, the depth and severity of poverty are more susceptible to those factors, and some adverse situations for poverty reduction, if not addressed promptly, can result in more depth and severity of poverty, as well as widening wealth gap. Certain factors, according to our regression analysis, may limit the scope of poverty without reducing the depth and severity of poverty. That is, those characteristics may only benefit people around the poverty line and rarely reach the truly impoverished. For example, urbanization may benefit poor communities on the edges of towns but does nothing to help the destitute in remote areas.
Economic growth and social development, according to the decomposition results (available upon request), contribute less to poverty depth and severity than to the scope of poverty. That is, while socioeconomic growth is important for poverty reduction, the majority of beneficiaries are the moderately poor, who face less poverty and are more likely to benefit. The extremely poor, on the other hand, confront a more dire situation and are less likely to gain. As a result, the government must target the extremely poor. The results of difference decomposition also revealed that government public fiscal investment may help to reduce the depth and severity of poverty. Despite transfer payment leakage due to targeting inaccuracy and elite capture in the process of reducing the scope of poverty, government public fiscal spending remains the most effective means of improving living standards for the deeply poor who are excluded from socio-economic progress.
5.2 PVAR Estimation
Variables such as economic growth and social development, as well as government public fiscal spending, may all have complicated and dynamic interactions with poverty incidence. As a result, the panel vector autoregression (PVAR) model has emerged as an appropriate choice. This model combines the benefits of the VAR model and panel data, and it considers all variables to be endogenous in order
12 to investigate the interaction and lag effects, making it suited for “large N and small T” short panel data. To detect the long-term trend, we use the PVAR model to analyze dynamic interactions between variables and poverty incidence (model details are available upon request).
We split two systems for separate investigation since incorporating multiple variables into the same model will increase the number of parameters to be estimated at the expense of flexibility. System 1 includes variables such as poverty incidence, economic growth, industrial structure, income gap, urbanization, and marketization, whereas System 2 includes variables such as poverty incidence, compulsory education, higher education, vocational education, social security, and fiscal decentralization (see Table 1). System 1 is obviously meant to examine the dynamic interactions of economic growth and social development variables with poverty incidence, whereas System 2 is intended to examine the dynamic interactions of government public fiscal spending with poverty incidence.
Before proceeding with the PVAR analysis, we first test data stationarity using three unit root test methods of LLC( Levin- Lin- Chu, 2002), IPS( Im, Pesaran & Shin, 1997), ADF- Fisher( Fisher’s Augmented Dickey Fuller method)(Fisher,1932), and the results indicate that all the variables belong to stationary sequences. Thus, the optimal number of system lags is determined based on the three information criteria of AIC(Akaike information criterion), BIC(Bayesian Information Criterion), and HQIC(Hannan-Quinn Information Criterion) for model selection, and results show that System 1 is of the second order and System 2 is of the first order. In other words, socio-economic development has a longer-term impact on poverty incidence than government fiscal spending. After determining the order of lag, the data are substituted into the model for GMM parametric estimation as well as the model’s stationarity and Granger causality tests. The results indicate that the model is generally stationary, and that all variables in both systems jointly compose the Granger cause of poverty incidence, allowing for pulse response and variance decomposition (test results are available on request). The PVAR model, on the other hand, is just a generic description of the correlation between variables in the system, and the GMM estimated results have no economic significance. Pulse response and variance decomposition, on the other hand, may be more relevant. While the former may show the path of unilateral dynamic effect of one variable after experiencing shock, the latter may demonstrate the contribution of variation(structural shock) in a specific variable to overall variation(forecasted variance) over the forecast period. Both can be used to observe the short-term fluctuations and long-term trends in lag effects.
Because the purpose of this paper is to investigate the effect of each variable on poverty incidence, we only reserve resluts of the pulse response and variance decomposition using poverty incidence as the response variable or forecast variable (additional results are available upon request). Figure 2 depicts the pulse response results, while Table 4 depicts the variance decomposition results.
In Figure 2 (a), the response of poverty incidence in System 1 to its own shock is still positive and the highest, implying the existence of inertia in poverty. Poverty incidence has relatively small responses to economic growth and Theil index, implying that economic growth and income gap under the current data trend are no longer the primary determinants of poverty. Poverty incidence significantly responds to industrial structure, urbanization and marketization, most of which are related to job opportunities and income growth. Compared with the lack of material wealth and income, the deprivation of information
and opportunities is becoming the main contributor to chronic poverty. Notably, marketization has a long-term positive effect on poverty, implying that it is urgent to address certain factors that are not propoor in marketization.
In Figure 2 (b), the response of poverty incidence in System 2 to its own shock is still positive and the highest but is starting to diminish. Poverty incidence responds negatively to compulsory education, vocational education, and higher education. This means that education has a significant lag effect in reducing poverty. Higher education creates the greatest effect in reducing poverty, and therefore holds the key to overcoming the intergenerational transmission of poverty. Fiscal decentralization has a negative effect on poverty; owing to fiscal decentralization, the government developed a preference to invest in infrastructure, creating a crowd-out effect on financial resources that otherwise could have been used to reduce poverty. Yet eventually, infrastructure is conducive to poverty reduction. Social security spending has a positive effect on poverty overall, suggesting the existence of significant leakage13.
Table 4 reports the results of variance decomposition for 20 periods under both systems. Results of System 1 indicate that when the forecast is conducted for the 20th period, contribution of change in poverty incidence to its own forecasted variance is around 30%, whereas for System 2 is around 50%; this implies that other variables in System 1 have a greater influence on poverty incidence than in System 2. In other words, socio-economic development has a longer-term effect on poverty compared with government fiscal spending. This result also reflects the existence of inertia in the change of poverty, which necessitates adjustment in the poverty reduction strategy in order to rectify poverty-inducing factors. Moreover, Table 4 also indicates that industrial structure and urbanization have a considerable influence on poverty forecast. Therefore, industrial development and new-type urbanization should serve as major contributors to poverty reduction in the future. According to the decomposition results, however, education has a limited contribution to the variance of poverty forecast, implying that the role of education in reducing poverty has yet to be unleashed. In comparison, per capita social security spending contributes significantly to poverty, indicating that flaws in social security system will create an adverse long-term impact on poverty reduction.
6. Concluding Remarks and Policy Recommendations 6.1 Concluding Remarks
Using provincial panel data from 2000 to 2019, this paper provides a multi-stage and multifactor decomposition of China’s poverty reduction process to explore contributors to China’s poverty reduction and identify China’s poverty reduction experiences and problems. We have reached the following findings after classifying poverty-reducing factors into three categories: Economic growth and social development, government public fiscal spending, and adjustment in the poverty reduction strategy.
First, economic growth and social development are the key drivers of poverty reduction in China. According to the variance decomposition based on GMM estimation, economic growth and social development both greatly accelerated the process of poverty reduction from 2000 to 2012. Economic growth has raised government fiscal income and created non-farm jobs, while marketization has allowed the movement of rural labor to cities for non- farm careers. Because of government infrastructure investment, urbanization and equal access to urban and rural public services have made significant progress, allowing poor regions to share in the benefits of development. However, between 2013 and 2019, the trickle-down effect of economic growth diminished, and social development’s contribution also saw a decrease as the pro-poor effect of marketization declined. Meanwhile, variance decomposition based on GMM estimation suggests that economic growth and social development are the primary contributors to inter-provincial poverty gaps, although economic growth’s contribution is smaller and has begun to promote inter-provincial poverty incidence convergence.
Second, government public fiscal spending has failed to protect the poorest. According to variance decomposition, social security spending through transfer payments contributed negatively to poverty reduction, reflecting problems such as targeting error and elite capture; the modest contribution of educational shows the untapped potential of education in poverty reduction. Another conclusion is that provincial public fiscal investment has failed to increase the convergence of inter-provincial poverty incidence. Therefore, it is necessary for the central government to coordinate and distribute central fiscal
assets to poor regions.
Third, adaptive adjustment in the poverty reduction strategy is essential to raise the efficiency of poverty reduction efforts. According to difference decomposition, the poverty reduction contribution of residual error was negative over the period from 2000 to 2012, indicating the existence of unquantifiable factors that diminish poverty reduction efficiency; over the period from 2013 to 2019, the contribution of residual error to poverty reduction substantially increased and turned positive, indicating that the implementation of the targeted poverty reduction strategy effectively addressed various contradictions and problems in the previous poverty reduction work and therefore significantly increased poverty reduction efficiency. Meanwhile, variance decomposition also indicates that the poverty reduction strategies and systems of various provinces have to some extent contributed to the inter-provincial poverty gaps, but are not the primary contributors compared with socio-economic development. The implementation of the poverty reduction strategy in recent years began to promote the convergence of inter- provincial poverty incidence, coordinating poverty reduction work in various provinces, accelerating the catch-up pace of poor regions and expediting the process of countrywide poverty eradication.
Fourth, although socio-economic development in the current stage may substantially reduce the scope of poverty, it does little to benefit the deeply poor; despite the leakage of government public fiscal spending in reducing the scope of poverty, it is one of the few effective means in assisting those who are left out from socio-economic development.
Fifth, there is a a longer lag and a greater extent in the poverty-reducing effects of economic growth and social development compared with government public fiscal spending, although the adverse effect of marketization calls for attention. There is a long lag in the poverty-reducing effect of educational spending, indicating that education is of far-reaching significance to poverty reduction. The long-term effect of social security spending on poverty is positive, highlighting the need to address the problem of elite capture to avoid long-term adverse impact on poverty reduction.
6.2 Policy Recommendations
The Central No. 1 Document of 2021 called for “establishing a five- year transition period following the eradication of absolute poverty to shift the priority from poverty reduction to countryside vitalization”. The primary task during the transition period is to address various problems during the poverty reduction period and extend the functions of the poverty reduction system to incorporate countryside vitalization into a comprehensive and institutionalized regular poverty reduction framework, as well as to develop long-term mechanisms for addressing relative poverty.
( i) Strengthening the poverty- reduction momentum of socio- economic development through inclusive economic growth and the countryside vitalization strategy centered on marketbased mechanisms and industrial development. The government should focus on addressing the problem of unmarketable agricultural products while maintaining stable economic performance and a stable employment environment. More efforts are needed to enhance the role of the market in resource allocation, to create a unified urban and rural land market, to ensure market entry and exit systems, and to facilitate the transaction of rural collective construction land. Financial institutions should develop the rural land finance guarantee business in order to improve the real right of use of rural housing plots. Furthermore, industrial development should contribute to rural vitalization by fostering feature industries based on local conditions, extending industrial chains, bringing farm products to market through leading enterprises, exploring collective economic modes such as cooperatives, creating prestigious brands, and promoting the spillover effects of tourism in reducing poverty (Wang et al., 2020).
(ii) The poverty-reducing efficiency of government public fiscal spending should be increased by improving poverty prevention monitoring and sorted support, as well as promoting equal access to
basic urban and rural public services, focusing on incentivizing local governance and empowering poor populations. Local governments should build capacity for poverty reduction and rural vitalization, promote a bottom-up self- governance model, and improve remuneration for grassroots cadres. To eliminate information asymmetry, the government should embrace digital information technology. Steps should be taken to improve poverty monitoring and categorized poverty relief, to use big data platforms for poverty reduction with the assistance of early warning and rapid response to poverty recurrence, and to improve monitoring and special groups such as migrant and rural left-behind residents. Priority should be given to poverty prevention and long-term support in order to provide the poor with opportunities for sustainable development. Progress should be made in equalizing access to essential public services between cities and the countryside, as well as promoting public-interest education, healthcare, and other welfare programs in poor regions to encourage people to stay in the countryside. Vocational schools should prepare students for local employment. College graduates should be encouraged to return to the countryside and use their knowledge and talents to help alleviate poverty. Experiences for achieving money and prosperity should be shared in order to inspire others to follow.
(iii) Strategic adjustment and system upgrade are needed to further enhance sustainable poverty reduction capabilities, allowing social forces to participate while prioritizing urban and rural cooperation and regional equilibrium. The approach to poverty reduction should evolve from targeted poverty reduction to a simple and systemic approach to poverty reduction. Simple approach to ensuring the cost effectiveness of poverty reduction is required; many poor groups, such as the elderly and migrant people, should be covered by generic poverty criteria. Furthermore, the complexity of relative poverty highlights the role of private organizations and informal systems in bringing together many stakeholders for poverty reduction on the basis of a capable government and well-functioning market (Huang, 2014; Xie and Song, 2021). It is also advised that cities and rural areas use distinct relative poverty lines to target poor groups in an accurate and simplified manner. Poverty reduction strategies and measures should be integrated to improve efficiency and encourage urban-rural integration, with a focus on the needs of the urban poor. Furthermore, poverty reduction programs should be tailored to local circumstances in order to tap into human capital reserves in the central region and accelerate marketization in the western region. To close regional gaps and establish regional cooperation, less developed regions should be incorporated into priority development plans.
1 As can be learned from variable specification, n1= 3 means ESD in the economic growth category includes three variables, i.e. agricultural development, non-farm employment, and infrastructure; n2= 4 means ESD in the social development category contains four variables, including industrial structure, income gap, urbanization, and marketization. Hence, ESD for socio-economic development contains a total of seven variables. The definition and value of mk are comparable to those of nj, and we have m1= 3, m2= 1 and m3= 1, i.e. GOV for government intervention in poverty reduction includes five variables, as detailed in “variable specification”.
2
China’s poverty line was 100 yuan in 1978 and 1,196 yuan in 2008 before raised to the current level of 2,300 yuan since 2010.
3
Some studies have substituted poor population with those living below the subsistence protection ( dibao) line. However, the dibao line’s frequent - in some years substantial - increases have led to a rise in poor population. This treatment method is therefore not adopted in this paper.
4 Growth rates need to be calculated for some variables. Hence, data for all variables start from 1999; the sample size is the number of provinces times the number of years (1999-2019), and the result is 651; the growth rate related variables start from 2000, hence the sample size in this case is 65131=620. The “poverty incidence” variable is missing for certain provinces in certain years, hence the limited sample size of 533.
5
Refer to Wang et al. for the calculation method. China Provincial Marketization Index Report (2018) [M]. Beijing: Social Science Literature Press, 2019. This report employs data of 2016 and before, and data of 2017 through 2019 is extrapolated based on trending.
6 Reference: Zhang et al. Will Fiscal Decentralization Help Reduce Poverty? Inter-Provincial Evidence after the Tax Sharing Reform [J]. Journal of Quantitative & Technical Economics (JQTE), 2010, 27 (12): 3-15.
7 Annual mean temperature, annual mean precipitation and annual mean sunshine duration are from the China Meteorological Yearbook of various years, the average altitude data of various provinces are obtained from Internet search, and relief amplitude data are calculated referencing Feng et al. (2007).
8 Residual ε' t= εt+μ; contribution of the constant term: cov( Yt, cons)/ var( Yt)= 0, and the constant term includes the common temporal trend of various provinces φt.
9
The absolute values of contributions of various factors in 2019 are subject to substantial increases or decreases, as the goal of eliminating absolute poverty approached its near completion. As a result, the incidence of poverty in provinces converged to zero with minimal variance. As a result, when the contribution ratios were computed, the denominator shrunk considerably.
10 The compulsory education indicator is based on educational spending for each rural compulsory education student, whereas the vocational education indicator is based on secondary vocational education spending. Such schools will only recruit local students, with a low possibility of spatial spillover.
11
Since 2007, the itemized data of pension spending for administrative and public institutions for individual provinces have ceased to be separately reported, but national data have continued to include such itemized data. As a result, based on existing ratios before 2007 and the rising trend of national statistics, we computed the proportion of pension spending for administrative and public institutions in social security and employment spending for various provinces since 2007.
12 The meaning of “large N and small T” is: there are many individuals at the individual level, butfewer periods at the time level.
13 The meaning of social security expenditure leakage is that social security expenditure has not been fully distributed to the truly impoverished people, and a part of it has been distributed to non-impoverished people.