Endnotes

¹ Ravallion and Bidani (1994) describes a method of non-food expenditures based on a regression of food share on a constant and the log of (expenditures/food poverty line). It is straightforward to calculate the food share for households whose food expenditures exactly equal the food poverty line (upper line) or whose total expenditures equal the food poverty line (lower line).

² Non-food expenditure share of 40 percent used prior to 1998 dropped sharply to a 17 percent in 1998 due to the change from a fixed share to regression estimation. Data made available to the author by the NBS show that in 1999, the non-food expenditure shares of the oor, defined as those with incomes below 850 yuan per capita, in Guizhou, Gansu, and Henan were 27, 33, and 49 percent.

³ Chen and Ravallion (1996) calculate CPI's for the poor that grow significantly faster than the overall CPI in two of four provinces. Khan and Riskin (2000) find that the CPI for the poor is four percent higher than the overall CPI for the period 1988 to 1995.

⁴ Nor do they allow for regional variations in the food bundle itself.

⁵ Riskin and Li (2001) oddly use a national poverty line while deflating incomes by provincial price indices, which will produce unpredictable bias in the change of the poverty headcount.

⁶ Personal communication with China Center of Preventative Medicine, Ministry of Health.

⁷ 1 US dollar equals to 8.26 RMB yuan.

⁸ The 8-7 poverty reduction plan was launched by the central government with the objective of bring the remaining 80 million absolute poor in the rural areas out of poverty within 7 years (from 1994 to 2000).

⁹ Income standards for poor county designation are different from official poverty lines, which are mainly used to calculate poor population and not available till 1993.

¹⁰ LGPR renamed these poor counties as key poverty reduction counties.

¹¹ Beginning in 2002, the interest rate was adjusted to 3%.

¹² It is straightforward to weight type 1 and type 2 errors differently. Just as for poverty measures, one can also give greater weight to larger targeting income gaps by using higher order weighting terms (Foster, Greer, and Thorbecke, 1984).

¹³ There is no obvious reason for choosing the official poverty line since the goal of the program is to target individuals below the poverty line rather than counties with average incomes below the line.

¹⁴ Since the targeting income gap for all counties is 77 yuan and all the income gaps are from one fifth of the mistargeted counties, the average magnitude of mistargeting in mistargeted counties is 385yuan.

¹⁵ Targeting income error formula is the same as for targeting income gap except the poverty line Z is the income of the threshold county and the summation is divided by D instead of N. Targeting rank error replaces income difference with income rank difference.

¹⁶ The MOA data is known to show more poverty in PRC's southwest and less in the northwest in comparison to the SSB data (World Bank, 1992). Both SSB and MOA data are available for poor counties in 1994 and 1995. The two series have a rank correlation of 0.89 and 0.92 in the two years.

¹⁷ Part of the measured bias against southwest provinces may be due to biases in the MOA versus SSB data. However, interviewed officials in Beijing confirmed that the number of poor counties in the poorest provinces was limited to preserve balance among provinces.

¹⁸ It is worth noting that the government only provides interest subsidies for subsidized loans and the capital money are from the state owned financial institutions. However, the losses from subsidized loan program are most likely to be taken by the government.

¹⁹ The World Bank implemented three large poverty reduction projects in the Southwest, QingbaMountainous Areas and Western PRC with a total input of $568 million US dollars.

²⁰ We have added average outstanding loans directly to the average funding for FFW and development capital. This gives disproportionate weight to the subsidized loan component of poverty funds, and also may introduce a bias towards progressivity if lower incomes correlate with lower repayment and thus higher outstanding loans.

²¹ Among 592 poor counties, 532 counties have complete data for all four years.

²² Agricultural Development Bank was set up in 1994 to manage policy loans for agricultural sector, and took change of subsidized loan for 4 year. Beginning in 1998, the management of subsidized loan was assigned back to ABC.

²³ The authors also have estimated an econometric model of consumption growth at the farm-household level allowing for individual effects with nonstationary impacts using the same data set. After controlling for (observed and unobserved) household heterogeneity, they find that living in an area with poor natural conditions reduces consumption growth, namely, poor areas tend to grow slower because of geographic externalities. Without allowing for the geographic externalities, the estimated rate of return from poverty programs is zero.

²⁴ In their year-to-year growth equation, endowment and structure variables are also in year-to-year form. Right hand side investment and endowment variables are lagged one period to help avoid endogeneity problems.

²⁵ The main equation is


To implement the fixed effects, first we rewrite the last equation as follows:


where small y and x denote logs, and the "~" superscript denotes differences from regional means, where regions can be defined as prefectures or provinces. We allow for the effect of the poverty program to be different for the period of original designations (1985-1992), captured by and the period of new designations (1992-95), captured by Imposing these restrictions, controlling for region-time effects, and implementing county fixed effects by taking first differences yields a system of 3 equations:



Here we replace t with explicit time subscripts (0=1981, 1=1985, 2=1989, 3=1992, 4=1995). The coefficients on the poverty status variables should be interpreted as the effect of the poverty program on counties in the same prefecture in the same period with the same starting income and grain production levels and controlling for time-invariant unobservables.

²⁶ Ravallion and Jalan (1998b) also find that without accounting for the geographic externality, the programs' impacts are zero.

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

Comment(s)

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