How Effective has Targeting Been?

К оглавлению
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 
119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 
136 137 138 139 140 141 142 

Some errors of targeting and some misappropriation are inevitable in any

economic environment and more can be expected in low-income countries.

Further, the very modest level of resources directed at the schemes would

also limit their impact, even given far lower targeting errors. However, the

consistent picture that emerges from the available evidence is that while some

schemes may have had a modest positive effect on the poor, in our casestudy

countries in general, trends in poverty reduction have been driven by

macroeconomic developments – the rate and pattern of economic growth

– rather than by targeted interventions.

There is a vast literature on the relationship between growth and poverty,

which concludes there is virtually everywhere a clear negative relationship,

although its strength varies between countries with different social, economic

and political structures. This can be illustrated for our country cases. Warr

(2000), for example, examines changes in poverty incidence (the headcount

ratio based on offi cial poverty estimates) across a set of countries including

India, Indonesia and Thailand. He fi nds elasticities of poverty incidence (the

proportionate change in the headcount ratio relative to the proportionate

change in GDP per capita) of –0.9 for India, –2.0 for Thailand, and –0.7 for

the Philippines.17 For PRC a similar exercise fi nds an elasticity for poverty

incidence of –0.8 (World Bank, 2001). Estimates are also available for the

income poverty elasticity, that is the relation between growth (change in

mean income) and the change in the income of the poor (normally taken

as the bottom quintile). For the Philippines the income poverty elasticity

(defi ned as the ratio of the latter to the former) is found to be relatively low

at 0.54, whilst for Indonesia the comparable elasticity is 0.71 (Balisacan

and Pernia, 2003; Balisacan et al., 2003). In both countries there is a clear

tendency for the elasticity for different quintiles to rise as one moves up

the income scale, although this is particularly marked in the Philippines.

In other words, although the poor benefi t from growth they do not benefi t

as much (both proportionately as well as absolutely) as the better-off.18

Similar results with growth accompanied by a strongly worsening income

distribution are found for PRC, with an implicit poverty elasticity of around

0.5 (Stern, 2001).19

These results imply that growth reduces the headcount index of poverty

and raises the income of the poor, although often not by as much as it

raises the income of better-off groups.20 However, the issue remains of the

impact of poverty-targeted programs discussed here, either in reinforcing

the positive effects of growth or in protecting the poor at times of recession.

As noted above, it would be unrealistic to expect a dramatic impact even in

the presence of more accurate targeting, given the modest budgets allocated

to these funds.21

Given the high leakage rates reported above and the administrative costs

involved in reaching the poor, one would expect that these schemes involved

relatively high costs of transfer per unit of benefi t received by the poor.

Estimates of the optimal degree of targeting, as discussed above, are rarely

available. However, in a simulation exercise for the Philippines, Balisacan and

Edillon (Chapter 6 in this volume) report that simple geographic targeting

provides the maximum benefi t to the poor for a given program cost, as

compared with other schemes, once the administrative costs per applicant

reach a modest fi gure (roughly 50 per cent of the daily minimum wage in

Metro Manila). The implication is that, once administrative costs are allowed

for, relatively simple forms of targeting dominate the alternatives.

Few rigorous cost effectiveness studies of alternative targeting schemes

are available. For India a comparison of employment guarantee schemes and

food subsidies suggests that at best the cost of transfer is nearly double the

benefi ts received by the poor. Approximate estimates suggest that the cost

of transferring a rupee to the poor through the Maharashtra Employment

Guarantee Scheme in its early years (Rs 1.85 per rupee transferred)

compared very favorably with both the later national employment scheme,

the Jawahar Rozgar Yojana (Rs 2.28 per rupee transferred) and the food

subsidy program under the Public Distribution System (Rs 6.68 per rupee

transferred) (Dev and Evenson, 2003). Separate estimates for the impact of

the Employment Assurance scheme in three states (West Bengal, Gujarat

and Haryana) found the cost per job per day to be Rs 200–300, which is

well in excess of wage rates, which were roughly in the range of Rs 35–50

(Srivastava, Chapter 2 in this volume).

The operations of the National Food Authority in the Philippines,

particularly through its rice subsidy, have been the subject of several cost

effectiveness assessments. For the early 1990s costs are again roughly twice

the sum transferred to consumers (Subbarao et al., 1996). However, NFA

rice is sold in special retail outlets in a form of self-targeting, and much

will leak to the non-poor. Assuming a 50 per cent leakage rate, more recent

cost effectiveness estimates for the NFA rice subsidy suggest that in 1997 it

costs Pesos 4.2 per peso of benefi t received by poor consumers and Pesos

2.5 per peso of benefi t in 1998. Much of this mis-targeting will have been

due to a regional misallocation with some of the poorer provinces being

under-represented, relative to their share in poverty, in the receipt of NFA

rice (Manasan, 2001).

In addition, however, it is important to remember that despite high leakage

and high cost, some of these schemes may nonetheless have been infl uential

in protecting the poor at times of adverse shocks. This is the judgement on

some of the many schemes introduced in Indonesia at the time of the Crisis

of the late 1990s, particularly in relation to health and education initiatives.

For example, there is some evidence that the education scholarship program

helped in keeping up school enrolment rates and reducing drop-out rates

from poor families. Similarly the Health Card scheme to allow free access

to public health facilities is credited with stabilizing the utilization rate of

such facilities by the poor (Perdana and Maxwell, Chapter 3 in this volume).

Cost and leakage may have been high, but some real benefi ts appear to

have been created.

Apart from these analyses of the cost of transfers to the poor, a few

detailed quantitative assessments of the longer-term income effects of this

type of program are available. Of our case-study countries, the most work

has been done for PRC. From a regression model Rozelle et al. (1998)

fi nd some positive income effects from direct lending to households in

poor counties in Shaanxi 1986–91; however, funds allocated directly to

enterprises in these counties do not appear to have any positive effect on

growth. Zhang et al. (2002) look at Sichuan province and compare growth

across program poor, non-program poor and non-poor counties. Allowing

for a range of other factors they fi nd that program status does appear to

have a positive effect on growth. Hence, whilst non-poor counties grew more

rapidly, the gap between poor and non-poor counties is lower when counties

have a designated poor status and receive poverty funding commensurate

with this designation. An even stronger result is provided by Park et al.

(2002) using a regression model, which makes growth across counties a

function of initial income, other initial characteristics (principally grain

production), time invariant characteristics, including poor county status,

and a number of time-varying factors. They fi nd that designation as a poor

county increases household per capita income, over that otherwise expected,

by 2.2 per cent annually in the 1986–92 period and 0.9 per cent annually

in 1992–95. When this rate of increase is compared with the amount of

funding to poor counties this gives a rate of return of between 12 per cent

and 16 per cent depending on the time period.22 This evidence needs to be

qualifi ed, however. First, even accepting the regression specifi cation as a

means of establishing the counterfactual in the absence of designation as

a poor county, the study makes no claims to know how the extra income

within the counties concerned was distributed. There need be no inevitable

assumption that the incomes of the poor grew by the same rate as average

incomes in the poor counties. Second, the authors make clear that their

results may be an over-estimate as they have not been able to include all

costs of the targeting programs. Third, their returns must be compared with

the opportunity cost of capital in China at this time, which was probably

relatively high, given the rapid growth rate, and may have been at least 12

per cent or more (which implies that equivalent or higher returns could have

been obtained on investment elsewhere in the economy).

A detailed examination of the impact of public spending on poverty in

PRC, which gives a less positive assessment of the poverty loans program, is

provided by Fan et al. (2002). Using a simultaneous equation model, that has

now been applied to a number of countries, they assess the effect on poverty

in terms of numbers pulled above the poverty line due to a given amount

of different public expenditures. By far the highest poverty effect is due

to education, followed by agricultural research and development (R&D).

Poverty loans have a relatively very small (and statistically insignifi cant)

impact per unit of expenditure. They have the smallest poverty effect of any

category of expenditure included (only 13 per cent of that of education, 15

per cent of that of R&D, and roughly one-third of that of roads).23 Similar

studies have been done for India and Thailand using the same model, but

only the India study includes poverty loans (covering rural and community

development and employment programs) as a separate expenditure category

(Fan et al., 1999). For the Indian case in terms of poverty impact the relative

ranking of the poverty expenditure category is higher than for PRC (it is

fourth behind roads, R&D and education).24 However, per unit its impact

is still well below that of these other categories of expenditures, being 17

per cent of that of roads, 30 per cent of R&D and 88 per cent of education.

No doubt the targeting errors reported in this chapter are a major part of

the explanation.