Assessment of overall effectiveness of poverty targeting programs

To date, there is limited evidence on the overall effectiveness of PRC's poverty investments. The main challenge of the assessment is to isolate the effect of poverty programs, since progress or lack of progress in reducing the poor may reflect factors other than poverty investments, especially those that affect the pace of overall economic development. Some have argued, for instance, that poor areas stood to gain more from market and commercialization reforms since the planning system forced them into production patterns that went against their comparative advantage to a greater extent than in richer areas (Lardy, 1983). There is some evidence that income growth in poor counties was greater than nonpoor counties in some regions (Park et al., 1996; Tong et al., 1995). At the same time, most studies of income distribution across all of PRC find rising inequality among rural areas, suggesting that the poor are falling further behind the rich.

Jalan and Ravallion (1998b) assess whether being located in an officially designated poor county affects growth in household expenditures, controlling for geographic externalities and other community variables that are likely to determine income growth. Utilizing NBS panel data (1985 to 1990) on households in four southwest provinces (Guizhou, Yunnan, Guangxi, and Guangdong), they find that living in a national poor county increases consumption by 1.1 percent per year, though this gain is offset by growing divergence in consumption due to other reasons. The rate of return on poverty investments is estimated to be 12 percent²³. The authors themselves point out that this estimation may overestimate the program's effect since some public expenditures on the programs may not be included, funds may be used for consumption rather than investment purposes, and community variables may omit factors that give poor counties advantages in achieving income growth.

Growth regressions examining the impact of the use of poverty investments on different sectors in poor area economies, agriculture, rural industry, and state-owned enterprise, make up the core of the analysis in Rozelle et al, 1998. The study utilizes a data set to examine the sources, uses and effectiveness of targeted poverty investments in 43 poor program counties of Shaanxi Province during the years 1986-91. The authors adopt three separate sectoral growth models in which the rates of growth in output value per capita in year t is a linear function of the current-year poverty investments, poverty investments lagged one year, government expenditures per capita (for controlling for other investments), rural income per capita (for controlling for private investments since income per capita is a proxy of private wealth), human capital represented by the share of the labor force that had graduated from middle school in 1985, the beginning of the period under consideration, lagged output value (for controlling for the initial size of the sectors), and county and time-related fixed effects (county and time dummy variables), as well as population density as a proxy for relative abundance of labor, which may reflect the allocation of labor across sectors. In the agricultural growth equation, they also include as a regressor changes in the availability of agricultural land, while in the growth equation of state-owned industry, they include fixed investments in the assets in both current and lagged-year form as regressors because such information is available in this sector. The estimation results reveal that for the sample of nationally and provincially poor area program counties, targeted investment funds allocated directly to households for agricultural activities are found to have a significant and positive effect on growth. In contrast, investments in township and village enterprises or county state-owned enterprises do not have a discernible effect on growth. In an even more disaggregated part of the study, investments in agricultural infrastructure (such as, terracing or soil leveling and improvements) do not positively affect growth rates in agricultural output by themselves. These results suggest that the poverty investments targeted directly at households have a positive growth affect. However, this study suffers from several limitations, First, it is based on data from only one province. Similar work using data from other provinces is necessary to have confidence that the results for Shaanxi can be generalized to other parts of PRC. Second, another important source of poverty investments, namely, the Food-for-Work funds are excluded from the estimation, which is most likely to affect the estimation of the impact of infrastructure investments since the majority of FFW funds go to infrastructure construction. Finally, investments lagged one year may not be able to capture the long-term impact of certain poverty investment such as infrastructure.

Zhang, Huang and Rozelle (2002) analyze the impact of participation in the national and provincial poverty programs on income growth in Sichuan Province. They classify all counties in Sichuan into program poor counties, non-program poor counties and non-poor counties. Using gross income per capita, designated or program counties started lower in 1985 and ended higher than non-program counties. Growth of real gross per capita income in program counties was positive and exceeded the very small rise for non-program poor counties. Increase in gross income per capita of poor program counties, however, did not keep up with increases in the non-poor counties. Although less evident, the poor program counties also outperformed non-program poor counties in terms of net income per capita. To examine the statistical significance of the differences in the growth rates among sub-groupings of counties, they regress the log of gross and net per capita income on a series of year and group dummy variables. The results show that growth rates of poor program counties were statistically indistinguishable from non-poor counties and non-program poor counties had significantly slower growth. They also use a single regression model to identify determinants of growth and examine the impact of PRC's poverty programs using six years data from 1990 to 1996 for 177 Sichuan counties. The growth of income is regressed on sets of independent variables representing resource endowments and the economic structure of the county, investment (by type) made through the fiscal system (which includes some but not all of poor area investments), and program participation²⁴. They find investments in agriculture, health and education, and electrification positively affect growth, though the effect on growth of some investments (e.g. those in "other" infrastructure projects) is not readily apparent. Another finding of interest from the growth regression is that the poverty program does positively increase growth, or more accurately, keep growth rates of poor program counties from falling as much as the growth rates of poor, non-designated counties. After accounting for endowments, structure, and beginning level of income, poor program counties grow slower than non-poor counties (by 2.95 percent per year less). However, this slower growth rate was still faster than non-program poor counties, which experienced growth rates 4.56 percent slower than those of non-poor counties. They authors attribute the higher growth rates in program poor counties toeither more effective use of poverty investments that go through the fiscal system, or to FFW or other poor areas programs that are not included in the fiscal investments. Considering that less than 20 percent of the total government poverty investments go through the fiscal system, the latter explanation is more likely to be true. There are several important differences between this study and Rozelle et al (1998) study. First, this study uses data from all counties in Sichuan while the other study only works with designated poor (program) counties, which enables this study to have the non-designated poor counties as a good comparison group. Second, investment data of this study are fiscal investments, excluding subsidized loans and FFW funds, while the other study includes subsidized loan and budgetary development funds (part of fiscal investments) in the regression. Incomplete coverage of poverty investments is a common limitation of both studies. Finally, this study uses expenditure data on health and education and finds a strong effect on income growth. However, it is hard to believe that expenditure on health and education (the majority of the expenditure goes to primary education) can generate big impacts on growth in such a short time period as that being studied.

Fan et al. (2002) develop a simultaneous equations model to estimate the various effects of government expenditure on production, inequality, and poverty through different channels. They conclude that poverty investments (measured as poverty loans) matter somewhat for growth and poverty alleviation, but not nearly as much as investments in other sectors of the economy. The study, using provincial data for the past 26 years between 1970 and 1995, shows that government spending on production-enhancing investments, such as agricultural R&D, irrigation, rural education and infrastructure (including roads, electricity, and communication) have all contributed not only to agricultural production growth. Moreover, these investments, which all have a public-goods aspect, also have a large and significant effect on the reduction in rural poverty and inequality. One of the most striking results is that large parts of the poverty and inequality-reducing effect are realized through improved access to rural non-farm employment. Government anti-poverty loans specifically targeted for poverty alleviation have the smallest impact on poverty reduction of any of the investment programs. This study has both strength and weakness in terms of assessing the impacts of poverty investments. The strength lies in its adoption of simultaneous equations model and the use of a panel data set lasting for 25 years. This is also the only study that uses provincial poverty incidence to estimate the impacts of public expenditures on poverty. The weakness of this study is also obvious. First, only subsidized loans are taken into consideration in the estimation and the other half of poverty investments in infrastructure, health and education, training, etc. are excluded. Second, even poverty loans do not enter into the simultaneous 'production equations' and therefore do not generate feedbacks in the way that infrastructure and other non-poverty investments do, most likely to generate biased estimation on the impacts of poverty investments. Finally, there are so many equations in the simultaneous system with a lot of strong assumptions that do notreadily hold, which is also likely to lead to biased estimation.

Using MOA county level data for all counties where data exist, Park, Wang and Wu (2002) estimated the impact of poverty reduction policy on average income growth in the poor counties. The growth in county i’s rural income per capita (Y) from period t-ô to time t is modeled as a function of the county’s status as a designated poor county made at the beginning of the period (Pit-ô), initial income per capita (Yit-ô), other initial characteristics (Xit-ô), county time-invariant characteristics (ãi), and prefectural time-varying factors (ëpt). The specification implicitly assumes that poor county designation is not endogenous to time-varying unobservables that differ within prefectures and are not correlated with initial characteristics. In the main specification, the sole X variable is grain production per capita, a commonly used poverty indicator in PRC. The error term consists of other time-varying unobservables and measurement error that are assumed to be uncorrelated with the regressors. A panel is constructed from data for each county for four time periods: 1981-85,1985-89, 1989-1992, and 1992-95. The first period predates the poverty program, the first poor county designations occurred during the second and third periods, and new designations were made during the fourth period. Information on growth rates before the poverty program began makes it possible to identify the effects of poor county status while also controlling, through county fixed effects, for unobservables that have persistent effects on growth. This also eliminates potential bias from the endogeneity of poor county designation to county unobservables that are time-invariant²⁵.

The authors estimate the three equations simultaneously using an iterative feasible 3SLS procedure, imposing appropriate cross-equation restrictions and using different instruments for the three equations. The instruments are lagged variables for income, grain production, and poverty status, and vary by equation because they are plausibly exogenous only when predetermined. Thus, the instruments for equation one are values in period 0, for equation two values in periods 0 and 1, and for equation three values for periods 0, 1, and 2. The estimation result shows that household net income per capita increases 2.2% and 0.9% faster in poor counties than in non-poor counties during the periods of 1986-1992 and 1992-1995. Without fixed effects, the effect of the poverty program is negative in both periods, although not statistically significant in the second period. One explanation for the different results is that counties with unobservables deleterious to growth are more likely to be designated as poor²⁶. Alternatively, the program's impact could be exaggerated in the differenced regressions if changes over time are benefiting poor counties, such as if poverty designations are going to counties with improved political connections which also facilitate growth, or if there is reporting bias associated with being a poor county. The effects are larger than those found by Jalan and Ravallion (1998b) for the period of 1985-1990 in four southwest provinces (discussed above).

Based on our measurement of program impact on rural income growth, it is possible to estimate the rate of return on poverty investments. In real terms, poverty spending during 1985-92 fell and then recovered to about its initial level, averaging 9.5 billion yuan per year (in 1995 yuan), equivalent to 89 yuan per person or 14 percent of rural income. Based on the 2.28 percent impact on incomes, the poverty program on average increased rural income by 13.8 yuan per person per year. This suggests a rate of return of 15.5 percent, somewhat higher than the 12 percent estimated by Ravallion and Jalan (1998). For the 1992-95 period, the rate of return is still 11.6 percent despite increased spending and smaller program effects, because the approximate doubling of the program's coverage reduced spending per capita to 55 yuan. Our estimates of program impact are open to different interpretations. Critics will argue that performance was much worse than we describe, because we do not account for all expenditures—we exclude administrative costs of the programs, matching or supplementary funds provided by local governments, relent poverty loans, international donor funds, and funds from a vast array of government and private initiatives. Some argue that the total of such spending is greater than official poverty alleviation funds (Xie, 1994). Thus, our estimates of positive impact on incomes could be overstating the rate of return on poverty investments by more than 100 percent. Second, indirect evidence of low repayment rates on subsidized loans and suspected substitution effects make the relatively high rate of return surprising. Third, it is possible that some funds are being diverted to direct consumption which is showing up as income, leading us to overstate investment returns. Fourth, differenced regressions remain subject to bias from time-varying unobservables that disproportionately benefit poor counties within the same prefecture. Finally, our results provide no evidence on the distribution of benefits within counties, so high impacts do not necessarily benefit the poor within poor counties. Other factors, however, may bias our estimates downward. First, if targeted programs also benefit poor counties not designated as poor, then leakage may dilute the measured impact on targeted counties even though the absolute effects are large. This is also true if provincial governments substitute budgetary allocations away from counties supported by national poverty alleviation funds, or initiate programs targeted at poor counties not designated as poor. Also, if consumed funds are being consumed directly and not being reported as income, benefits may be greater than suggested by the impact on income. We have empirical evidence that designated poor counties have fewer budgetary funds than non-designated counties ceteris paribus, pointing to slight selection or substitution effects that should lead to downward bias in program impacts. Poorer relative performance in 1992-95 is consistent with our knowledge of aspects of program implementation. The pattern of spending on subsidized loans shifted away from agriculture (households) toward industry (firms and intermediary organizations), despite the greater return to the former (Rozelle et al., 1998). The budgetary crisis in poor counties became acute beginning in the early 1990s and worsened over time (Park et al., 1996). On the other hand, benefits of Food-for-Work infrastructure (a program without significant funding until the early 1990s) may take more time to be realized, so that the lack of program impact for the most recent period may be premature.

Unfortunately, data do not permit the authors to separately estimate the extent to which specific programs affect income growth. We have data on county fund allocations only for the years 1994-96, and given the shortness of the panel, it is impossible to properly control for unobserved heterogeneity and time-varying factors. Despite these reservations, we estimate a model of third-period growth as a functionof average funding levels during 1994-96, including provincial dummies and initial period economic variables, as well as minority and revolutionary base status. We find no significant effect of poverty alleviation funds, except for a slight negative effect for subsidized loans.

Evidence from the above studies suggest that poverty programs in PRC have positive impact on household income growth and poverty reduction in poor areas, or more accurately, have kept poor regions from falling further behind, but the impacts from other investments seem even bigger. The efficiency of poverty investment is decreasing with the decrease of the rural poor population and increase of poverty funds, possibly because of the worsening targeting problems and irrational use of some poverty funds, e.g. subsidized loans are used inefficiently in rural enterprises in poor counties. Investments in agriculture, education and heath seem more promising than investments in industry in poor areas. Due to the lack of reliable poverty data at disaggregated (county) levels, none of the above studies has managed to disaggregate gains to poor and non-poor from poverty programs, and this requires future research.

Download this Discussion Paper [ PDF 375KB| 60 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.