8.1. Introduction: Background, Scope and Methodology
From the poverty reduction point of view the success of MFIs would depend on their ability to reach out to the poor, provide services that cater to the needs of their poor members, reduce the current poverty level of their beneficiaries, and foster a process of graduation of the most disadvantaged sections among their clients out of poverty.37 The present chapter on "poverty reduction and graduation" addresses three-fold objectives. First, it seeks to assess the extent of poverty reduction under microfinance. Second, it attempts to capture the extent of vulnerability reduction under microfinance. Third, it provides evidence on the process of graduation of the poorest members of the MFI clientele out of poverty.

Reduction of Poverty and Vulnerability under Microfinance

The scope of the work is restricted to the following items of primary interest:
(a) Analysis of income-poverty trends by relevant member/ organizational categories between the first and the third round.
(b) Analysis of the poverty impact of microcredit based on the analysis of the determinants of household level consumption and poverty data.
(c) Analysis of the impact of microcredit on vulnerability of the poor.

proportionately allocate more resources to investment in human and physical capital even at low level of food consumption. By investing more resources now in capital accumulation they aspire to end up in sustained higher level in the subsequent period. The main hypothesis is that the proportion of such households would be higher in case of MFI members compared non-members. Accordingly, the scope of the work under this component of the study would include the following:

* This was prepared by Dr. Binayak Sen, Senior Research Fellow, BIDS, Dhaka, Bangladesh

1.

(a) Analysis of the pattern of investment in physical capital by relevant membership categories, controlling for their poverty status.
(b) Analysis of the pattern of investment in human capital by relevant membership categories, controlling for their poverty status

8.2 Poverty Measures

The measurement of poverty involves
(a) the specification of the income level below which a person is considered poor, - (the so called "income-poverty line") and
(b) construction of an index to measure the intensity and severity of poverty suffered by those whose income is below the poverty line. The most widely used measure of poverty is the so-called 'head count' ratio, i.e., the proportion of people living below the poverty line.

Sen (1976, 1981) and Kakwani (1980a, b) have proposed several criteria that a poverty measure must satisfy to be able to assess the changes in social welfare. First, an increase in income of a person below the poverty line, with the income of others remaining the same, must reduce the poverty measure (improvement in poverty situation). Second, a pure transfer of income to a poor person from a richer person without making either cross the poverty line must reduce the poverty measure. Third, a poverty measure is a characteristic of the poor and not of the general poverty of the nation. So no fall in income of the people below the poverty line can be outweighed by any rise in income of the people above the poverty line. Fourth, the poverty measure must be sensitive to income distribution among the poor. A given increase in income of a person far below the poverty line compared with the same increase for a person near the poverty line must reduce the poverty measure.

The head-count measure of poverty, which is simple to interpret and hence has appeal to politicians and policymakers, does not satisfy the above criteria. The measure does not show an improvement in poverty situation until a poor person has sufficient increase in income that pushes him or her above the poverty line. For example, if certain development activities (e.g. Rural Public Works or Vulnerable Group Development Program) focus on the extremely poor and succeed in raising their income but not enough to lift them above the poverty line, success will not be reflected in the head-count measure of poverty. In contrast, another development
2.

37 The term "poverty" and "income-poverty" is used interchangeably in this chapter.

activity (say credit for generation of self-employment) that focuses mainly on households nearer the poverty line may succeed in pushing them above the poverty line with the same or even smaller increases in incomes. From the welfare point of view, the former is a more desirable outcome than the latter, but one would draw contrary conclusions from changes in the head-count measure of poverty. If reduction in intensity and severity of poverty is the concern, greater weights must be given to the poorer households among the poor.

Foster et al. (1984) proposed a class of poverty measures that are additively decomposable and that satisfy all the criteria for an ideal poverty measure. This measure, known as the FGT index, is given by



where

n = the number of sample households,
Z = the estimate of the poverty line income,
q = the number of people whose income is below the poverty line,
Xi = the income of the ith individual among the poor households, and
a = parameter that reflects the society's weight given to the poverty problem.
It may be noted that the income of the (n-q) nonpoor households (with income above the poverty line) does not affect the poverty measure. The parameter a determines the weight given to the severity of poverty. If the value of a = 0, society does not distinguish among the poor and is merely concerned with their number. In this case, the poverty measure reduces to a head-count ratio that estimates the "incidence of poverty."

If the value of a = 1, each poor is weighted by his or her relative distance from the poverty line. The society is indifferent between an absolute increase in income accruing to a person who is nearer the poverty line and the same incremental income accruing to the person who is further away from the poverty line. In this case, the poverty measure reduces to

3.

This is a measure of the aggregate poverty gap and shows the percentage of total income needed to be transferred from the nonpoor to the poor households to lift them above the poverty line. If this measure is estimated for the subsample of the poor people, we get Sen's (1981) concept of poverty gap ratio, which is a measure of the "intensity of poverty".

If a society is particularly averse to the inequality among the poor, the poverty measure must give higher weight to an income transfer to the poorer compared with a less poor household (Kakwani, 1980b). Thus, the value of a should be more than unity. An improvement in the distribution of income among the poor through a transfer of income from a moderate to a hardcore poor, even when the head-count ratio and the poverty-gap ratio remain unchanged, should be considered desirable from the welfare point of view. In recent empirical studies on poverty, a = 2 is commonly used. This is a measure of the severity of poverty and is estimated by



8.3 Microcredit and Income-Poverty Trends: Evidence from Panel Survey

This section attempts to capture trends in poverty by participation status in MFIs as emerging between the first and the third rounds, corresponding to 1997/98 and 1999/00. Altogether a panel of 1526 target households (defined as those with land less than 50 decimals) has been identified for poverty comparisons. In this present exercise we use the household income data for estimating poverty. Note that more relevant measures of poverty comparisons between the two groups would be FGT poverty-gap and squared poverty-gap since the target households usually have average income lower than the poverty line.

8.3.1 Participation Categories

Poverty is estimated for three major participation categories, as defined below:

(a) "regular participants" who were members of MFIs in all three rounds;
(b) "occasional participants" who were members of MFIs at some point, either during the period preceding the survey, or during one of the three survey rounds only;
(c) "non-participants" representing the traditional control group who never participated in the MFIs.

4.

8.3.2 A Faster Poverty Reduction Rate for MFI Participants?

Poverty comparisons are carried out for the eligible (target) group with land-size less than 50 decimals. This is done to control for the possible differences in the "initial conditions" (such as land and non-land assets) across participation status. Poverty-gap ratio, measuring the "depth of poverty", is selected as the suitable measure for comparison (it is also easier to interpret than FGT-2 measure). Poverty-gap ratio is simply the product of head-count ratio ("incidence of poverty") and income-gap ratio ("intensity of poverty").


It may be seen from the table that initial poverty condition was not similar across the participation categories in round-1 (Table 8.2). If one goes by the criterion of head-count index, the initial poverty situation was actually worse for the target participants than the non-participants (68.4 vs. 62.4 per cent). The conclusion is reversed when one considers the income-gap index (43.4 vs. 49.0 per cent). The poverty-gap index, which is a synthetic measure of the above two indicators, appears quite similar between the participants and the non-participants (29.7 vs. 30.6). Given these ambiguities regarding the initial poverty ranking between participants and non-participants, it is desirable to focus on the rate of change of their poverty 5. between the first and the third round. The contribution of MFI to poverty reduction should show up in the faster rate of decline in poverty measures recorded for the regular participants compared with the non-participants.

The results show that regular participants have registered a faster rate of poverty reduction than occasional participants and non-participants. This result is valid for both head-count and poverty-gap measures. Thus, in case of the regular participants the poverty-gap index has declined by 13.6 per cent between the first round and the second round (Table 8.2). This may be compared with only 10.8 per cent drop recorded for the non-participants group. Similarly, the proportion of population living in poverty has gone down by 12.4 per cent in case of the regular participants compared with only 6.9 per cent for the non-participants during the same period. The observed difference in the rate of poverty reduction may be ¾ at least partly ¾ attributable to MFI impact. It may be noted that the poverty reduction rate estimated for the occasional participants was higher than in case of non-participants, but lower than the matched decline registered for the regular participants.38
The fact of faster rate of poverty decline as per the poverty-gap measure compared with the rate of decline measured on the basis of head-count index has important implications for assessing the poverty impact of MFI programs. This implies that benefits from MFI have not been restricted only to those participants persisting in and around the poverty line. The less advantaged within the poor community has also benefited from participating in MFIs. This trend is pronounced for the regular participants than in case of occasional participants. The issue of the less advantaged within the poor community will be examined in the later part of this chapter in the context of assessing the impact of MFI on fostering the "graduation process" for the extreme poor.

Has the above result any bearing on assessing the poverty reduction experience in general in rural areas of Bangladesh for the period under consideration? The declining trend in poverty

38 The qualitative conclusion about improved performance of regular participants of MFIs compared with the non-participants does not change if one uses different price data other than the prices derived from the PKSF survey for setting the poverty line. Using the BBS prices for the rural areas for the period we find a trend of poverty increase across the participation categories. But, the results still show MFI regular participants in favourable light: participants have registered a much lower increase in poverty than the non-participants. The unsuitability of BBS price series for making poverty comparison for the PKSF panel sample over the period under consideration has been pointed out earlier in Box 1. 6.

8.3.3 "Small Is Beautiful"?

Within the category of participants, however, there is a considerable variation in the poverty reduction rate by MFI characteristics (Table 8.3).

As per the preferred measure of poverty-gap index, the reduction rate in case of the relatively small MFIs has been quite comparable to (if not higher than) the large MFIs. Thus, the poverty-gap index has declined by 39-42 per cent for the two relatively small NGOs such as "Sabolomby" and "Prottayshi" compared with 3-10 per cent drop observed for BRAC and Grameen Bank. The poverty reduction rate for MFI members with multiple membership with both large and small NGOs has been assessed at 11 per cent, while the matched figure for "all others" is 9 per cent.40 The other aspect noteworthy from the table is that small MFIs have been equally (if not more) successful in reaching out to the poorer households than large MFIs. This can be judged by comparing the level of poverty-gap index between small and large MFIs for the first round. Thus, both Grameen and BRAC members have lower initial poverty rates compared with Sabolomby and Prottayshi (25-31 vs. 35-38 per cent). While the issue of small vs. large MFIs from the poverty reduction point of view needs to be explored further the evidence is clearly dismissive of the view that small and local MFIs are necessarily less capable in implementing anti-poverty policies than large and national MFIs. The former should get as much policy support as the latter have enjoyed to date, especially when it comes to expanding the credit accessibility of the target poor as well as addressing the issue of covering the left-outs.

8.4. Determinants of Income-Poverty: Does Participation in MFIs make a Difference?

Table 8.4 shows that standard FGT measures are lower for target participants compared with target non-participants (i.e. households from the same landowning group of up to 50 decimals).

39This result is in congruence with the recent results for HES 2000 showing decline in rural poverty between 1995/96 and 2000
40 The only notable exception in this sample appears to be " Noabeky", which did not register any decline in poverty, which may be due to the fact that the area of its operation was severely affected by the 1998 flood. 7.
This result holds true for both "within program village" and "across program and control village". However, it is not clear whether participation in MFIs per se or some other factors (such as varying initial conditions) are driving this result. Accordingly, the question we ask in this section is: does participation in MFIs reduce poverty controlling for the differences in initial resource endowments at household and community levels? This has been tested by carrying out analysis in three stages, as described below.

8.4.1. Multivariate Analysis of the Determinants of Consumption

A typical multivariate model of consumption determination has been estimated for the poor (Table 8.5) as well as for the target group as a whole (Table 8.6).41 The results show that, controlling for possible household level differences in asset endowment, literacy level, relevant demographic and labour characteristics as well as variations in village-level income-earning environment measured with respect to flood-propensity, irrigation and average affluence, participation in program (the so-called "pure effect" of program participation) is a significant explanator of average poor's consumption (Table 8.5). Similar results are noted when the model is separately run for the target group (Table 8.6). Within the group of program participants, there is some indication that average consumption tends to rise with increase in the length of membership (Table 8.7), displaying signs of household level graduation.

8.4.2. Multivariate Analysis of Determinants of Poverty

For the completeness of the argument, we estimate the above model, this time taking poverty-gap (measuring the "depth of poverty") and squared poverty-gap (measuring the "intensity of poverty") index as dependent variable. The results show that participation reduces poverty of the program members measured with respect to the both dimensions of poverty (Tables 8.8 and 8.9).42 There is also some indication that poverty reduction effects of the programs become more pronounced with increase in the length of membership. These findings reinforced signs of household level graduation. However, the explanatory power of the model is rather modest. Only 41 See, annex 1 (part b) for the description of the model.42 See, annex 1 (part a) for the definition of poverty measures. 8.
a quarter of the variation in poverty rates could be explained by the proposed model. This shows the need for further probing into the issue of determinants of poverty.

To sum up the discussion so far, the key finding of this section has been that participation in micro-credit programs has a positive and significant effect on poverty status of the program household even after controlling for the possible differences in the initial conditions and selectivity biases across participation categories.
9.
8.5 Microcredit and Vulnerability of the Poor

Rural households face, on a regular basis, different kinds of crises. They originate from a wide range of events, such as, natural calamities, income decline due to household members' illness and subsequent withdrawal from the labor market, unanticipated extra expenses (such as, on medical treatment or for social ceremonies), etc. The present Study finds that while the incidence of crisis faced varies across landownership groups the incidence of such crises is quite similar across households within the same wealth/ asset status. Thus, within the land-size group of 50 decimals the incidence of crisis (i.e. the number of crisis-events faced by a household in a given period) is similar across target participants and target non-participants. In short, participation in the MFI programs does not affect the incidence of facing these crises by the poor households. However, as we shall observe below, where participation in MFIs makes a difference is the manner in which such crises are being coped by the poor households. MFI members have the ability to take recourse to "soft options" for crisis coping, while the non-participants within the poor group have to rely on "hard options".

Poverty reduction is not just about devising means for income generation; it is also about devising ways for preventing income erosion. Microfinance can cut both ways, i.e., by providing the poor the means for generating income as well as by helping them to better protect against anticipated and unanticipated risks and shocks. In this section some results are presented which indirectly highlight the risk-insurance aspect of microfinance. Risks and shocks in sociological terms are better known as "crisis" (equivalent to "shankat" and "bipad" in the vernacular). Some shocks are covariate (as in the case of natural disaster), some are systematic (as in the case when the poor are routinely exposed to infectious health risks) while others are idiosyncratic (such as those associated with personal insecurity). Whatever the nature of the shocks they can have longer-term consequences depending on the extent of material and moral damage they entail and the means of crisis coping available to the poor.

The means of crisis coping can be broadly grouped into two categories. First, one may analytically separate out positive coping methods, which rely on the soft options such as borrowing at zero or low interest, and/or mobilising help and support from the patrons and kinship network in a variety of forms, material and moral. Second, one may also highlight 10.

negative coping methods, which are based on the hard options such as dissaving and asset sales/mortgage, with long-term implications for debt and recovery. The divide between these two methods of coping is often blurred, as in the case of "curtailing family consumption expenditure", which may be a relatively soft option for the not-so-poor, but may be the last resort for the extreme poor. There is also the third important analytical aspect in discussing coping methods, which is the category of "no coping", i.e., when the minimum means for coping is not available. Accordingly, one may hypothesise that the incidence of "no coping" would be less for the program members than for the non-members. Similarly, the MFI members may have greater options for deploying positive coping methods than would be case for the non-members.

Data on the aggregate number of "crisis-events" along with the information on coping methods have been compiled for the first and the third rounds, and for the eligible group disaggregated by participation status, as defined in the study. Three points are noteworthy from Table 8.10 First, if one focuses on the two principal categories of "negative coping" such as asset sale/mortgage and dissaving, the stark difference between the regular participants and occasional participants clearly comes through for the round-1 data. Together they represent about 10 per cent of the total crisis events for the group of regular participants as opposed to 20 per cent for the non-participants. Second, the difference in the incidence of "negative coping" is less apparent in the round-3. This is because both program members and non-members have been affected by the preceding year's flood, lead to asset/savings depletion across participation status. Third, what appears to be a consistent difference between participants and non-participants in both rounds relates to the incidence of "no coping", As expected, such incidence is lower in case of MFI borrowers compared with the non-borrowers. For the round-1 it is assessed at 11 per cent for the regular participants vis-a-vis 16 percent for the non-participants. The corresponding figures for the round-3 is 36 and 40 per cent.11.

8.6 Rethinking "Graduation": Emerging Trends

8.6.1 Graduation: What it is not

Graduation cannot be seen as a process of graduating out of MFIs with which the poor is currently associated. MFIs provide important financial services to the poor based on membership criteria. This explains continued presence of the poor in MFIs even when there is perceptible progress in economic and social dimensions of well-being.

There are two approaches to capturing the idea of graduation at the level of target households:
(a) graduation defined as "process" of escaping from long-term trap of poverty (call it the "process view"),
(b) graduation defined as the ability to cross some pre-identified poverty thresholds (the "threshold view").

8.6.2 "Process View" of Graduation

If the objective is to map the "process" graduation as way of climbing out of long-term inter-generational poverty, then graduation may be seen as a process of accumulation of critical minimum "self-development" ability to participate in economic growth. The key question is whether the MFIs are able to aid the poor in this process of sustainable poverty reduction. Using a growth accounting framework, we postulate that the rate of income-poverty reduction would depend on the rate of physical capital accumulation, human capital accumulation, and the rate of technological progress. Households and economic agents--even at similar level of current income (consumption)-- may differ as to their investment behaviour with respect to the above three variables of interest. The emphasis would be to check for systematic differences in the rate of physical capital accumulation, human capital accumulation, and technological progress across relevant member and organizational categories. 12.

8.6.3 "Threshold" View of Graduation

If the objective is to measure the number of "graduated" people, then the focus shifts to counting as graduated everybody who has crossed the threshold(s) in various dimensions of critical minimum ability (however defined). Some (tentative) examples are given below:

(a) Ability to overcome hunger (with implications for nutrition-productivity links): households able to overcome extreme food-poverty
(b) Ability to foster human capital (with implications for future occupational choice, productivity, and growth): households able to support primary and secondary education of boys and girls without dropout
(c) Ability to enhance security against shocks (with implications for vulnerability): households able to deploy "positive coping" such as raising emergency loans from multiple sources
(d) Ability to promote voice and freedom (with implications for bonding and bridging):
Households without any extra-economic coercion, or some relevant index of "powerlessness" suitably defined.

In this chapter we have mainly focused on three aspects for which we have detailed cohort data disaggregated by extreme poverty and participation status. These are
(a) ability foster human capital,
(b) ability to accumulate physical capital, and
(c) ability to enhance security against shocks.

8.6.4 Investment in Human Capital

A higher proportion of regular participants invest in human capital. As may be seen from Table 3.11, the share of investors in human capital during the first round of the survey was 56 per cent for the regular participants compared with 48 and 41 per cent recorded for occasional and non-participants, respectively. By the third round the matched share has gone up to 68 per cent for regular participants as against 59 per cent for occasional participants and 57 per cent for non-participants. This conclusion does not change even if one focuses on the extreme poverty group. Between the two round the corresponding share of investors in the participant extreme poor group has increased from 61 to 75 per cent. In contrast, the matched share for the non-participant extreme poverty group has gone up from 43 to 64 per cent. In short, those among the extreme poor who participated in MFIs exhibited a more pronounced tendency to invest in human capital development of their families compared with their non-participant counterparts. 13.

8.6.5 Investment in Physical Capital

A higher proportion of regular participants invest in physical capital as well. This is true for both the survey rounds (Tables 8.12 and 8.13). Here we discuss the findings of the last round only and restrict our comments to the comparative situation of the extreme poor between the two rounds (Table 8.13). The results show a clear edge of the participant extreme poor households over non-participant extreme poor households for a broad range of asset categories. The comparative examples capturing the proportion of asset owners between the two categories display the following pattern: radio (25 vs. 19 per cent), wall clock (40 vs. 29 per cent), rickshaw/van (14 vs. 8 per cent), bi-cycle (20 vs.12 per cent), hand tubewell (28 vs. 19 per cent), tree (84 vs. 71 per cent), chair/table (63 vs. 44 per cent), and bed (91 vs. 81 per cent). A small proportion of extreme poor participants also own shallow tubewell and engine operated boat, which is non-existent in case of their counterparts.

Data presented in Tables 8.12 and 8.13, however, represents comparison for the panel households disaggregated by participation status. A more defensible comparison would be to carry out cohort comparison, i.e., track the change of the same group of extreme poor households over the survey rounds. This is attempted in the next section.

The Cohort Comparison: Results for the Extreme Poor Group

In this section we present the results of the cohort comparison for the extreme poor group disaggregated by participation categories. Here we consider the following dimensions, namely, asset, education, and (emergency) credit availability from the informal market. Each is reviewed in turn. 14.

Comparison over Asset Dimension

The results for change in asset acquisition are given in Table 8.14. Here we construct an aggregate score of major assets for each household in the extreme poor group for both the rounds. The following aspects are noteworthy. During the first round the total asset score estimated for the extreme poor group was only slightly higher in case of regular participants compared with the non-participants (2.79 vs. 2.22). There has been a sharp divergence since then: the matched figure for the regular participants has gone up to 6.37, while that for the non-participants increased to 5.11. The faster increase in asset accumulation in the group of participants is suggestive of the favourable impact of MFIs on the graduation process of the extreme poor through the physical capital accumulation channel.

Comparison over Educational Enrollment

The school enrollment of children has been higher for the extreme poor participants group for both the rounds and for both primary and secondary levels. This further reinforces the observation previously made with respect to higher share of investors in human capital among the participants category in general. Thus, gross enrollment ratio at primary level was 1.03 for the first round in case of regular participants compared with 0.76 for the non-participants (Table 8.15). There has been a noticeable reduction of the enrollment gap between the two categories for the primary level, possibly reflecting all-round expansion of primary education in rural areas due to favourable public policy such as Food for Education. However, in case of secondary level of education where private decision to educate children is the major explanator, the participants have a clear edge over the non-participants. Thus, in case of secondary enrollment the observed rate is twice as high in the participant group compared with the non-participants. This finding is suggestive of favourable impact of MFIs on the graduation process of the extreme poor through the human capital accumulation channel.

Comparison over Credit Accessibility

The credit access to informal market for meeting emergency contingencies is an important indicator of the graduation process. Previously we have observed that MFIs have a favourable impact on reducing vulnerability of the poor households in general. Participants are likely to 15.

deploy positive coping method involving "soft option" such as interest-free loans while the non-participants face the risks of using "hard option" such as asset sale and dissaving. Data presented in Table 8.16 further suggests that credit accessibility to "neighbours" has increased for the extreme poor regular participants while it has considerably reduced for the non-participants. The average size of loan that can be mobilised from the neighbours is twice as high in case of regular participants compared with the non-participants. In contrast, the dependence on moneylenders, who advance relatively high interest loans, has slightly declined for the participants but slightly gone up for the non-participants. In general this finding combined with the previous observation on vulnerability-coping is suggestive of favourable impact of MFIs on the graduation process of the extreme poor through the informal (emergency) credit access channel. 16.

Note: "Eligible" group refers to conventional target group status with land ownership less than 50 dec. of land

Table 8.3
Poverty Dynamics by Organizational Status, Round-1 and Round-3:
Results for the Eligible Group

Note: Large 4 MFIs refer to Grameen Bank, BRAC, Proshika and ASA. The poverty estimates for the last two MFIs are not shown separately because of the small sample size.

Table 8.4
Poverty Estimates by Program Participation and Target Group Status,
1997/98, Consumption Expenditure Data

Note: Estimated food poverty line is Tk. 4906 per person per year, non-food poverty line ¾ Tk. 3070 per person per year, yielding total poverty line of Tk. 7976 per person per year. The food poverty line estimation procedure follows Ravallion and Sen (1996). The non-food poverty line is based on the upper poverty line method as described in World Bank (1998).
Source: Based on the PKSF Census data for 1997/98.

Table 8. 5
Determinants of Consumption of the Poor