Risk-Coping in Low-Income Populations

Coping with income fluctuations important problem when incomes low

Agriculture has high-frequency fluctuations: intra and inter-annual variability

How to maintain consumption in face of fluctuating incomes?

1. Reduce income fluctuations: mitigate effects of production shocks on income

2. For given income realization, take action to maintain consumption

Sector/timing income/production consumption

ex ante: prior to realization of

shock

Contracts: share tenancy

Asset allocation:

crop diversification

irrigation investment

plot diversification

Occupational diversification

Save

Insurance contract

Social insurance arrangements:

marriage

migration

ex post: after realization

? Borrow

Sell assets (dissave)

Transfers

Labor supply

Production and risk

Does 1) absence of insurance, 2) dislike of risk affect productivity?

Structure

1. Utility of farmers:

c c 1 2U = V(: , F ) V >0, V <0 (dislike variability)

c cwhere : = average consumption, F = consumption sd

2. Technology (CRS):

Farm assets differ in two dimensions: profitability and

contribution to risk (e.g., irrigation pump versus plow)

B T: = Wf("): av. profits and average weather(T)

B T variability in profits and weatherF = W'(")F

where W = total productive wealth of the farmer

1... n" = {" " }, the farmer’s asset portfolio,

i =with " share of asset i in total productive wealth of farmer

f() reflects the productivity of the farmer’s portfolio

'() reflects the riskiness of the farmer’s portfolio

3. Constraints:

c B: = :

c BF = k(W)F k’<0, 0 # k # 1

Perfect insurance: k = 0 No insurance: k = 1

4. Behavior: maximize utility subject to technology and

constraints

Results (FONC):

i nPerfect insurance: f - f = 0 (production efficiency)

1 i n 2 i n TImperfect insurance: V (f - f ) = - V (' - ' ): k

implies:

2If farmers dislike risk (V <0) and are uninsured (k�0),

A. Inefficient production

B. Positive relationship between marginal

contribution to profit levels and to profit

variability across assets and farmers:

1 n 1 nf /f = ' /' overinvest in risk-reducing

assets

Absent insurance market means farmers must

trade-off profitability and risk reduction

C. If wealthy farmers are less risk-averse or better

able to insure, then wealthy farmers will have

more profits per unit of wealth and more variable

incomes per unit wealth

Important issue: Absence of insurance market implies that poorest

farmers are most inefficient because they are poor:

they choose to be inefficient!

But policy-makers, observing inefficient poor farmers but not knowing

the root cause, could conclude:

a. Provide technical assistance, training for poor

b. Not engage in equalizing land reform (why give land to the

inefficient?)

Thus, need to know whether its missing insurance, or incompetence:

Empirical questions: Do we find A, B, C?

Two challenges: characterize technology and measure weather

1. Specify a profit function:

We want a flexible function (not Cobb-Douglas!) to characterize

technology: normalized (by total wealth portfolio) quadaratic

kt i i ikt i j ij ikt jkt i i ikt t T t ktProfits/W: B = G$" + 1/2EE * " " + G(" T + ( T + ,

(Specification also includes year effects (prices) and a farmer effect)

kt i i ikPortfolio riskiness: ' = sqrt((G(" ) )2

2. Measure weather: what aspects matter?

What if principal asset, store of value is used for consumption-smoothing?

“Stocking out” then means out of business

Example: use of bullocks in Indian ICRISAT villages

What’s a bullock?

1. Bovine male used for traction - animal power

2. Use of bullocks (pair) necessary for agricultural production in monsoon

agriculture (bovine economy)

3. Ownership of bullocks also necessary: problem of rental (moral hazard again)

4. Good also as buffer stock: transportable

ICRISAT Facts: 1. Important part of portfolio

2. High turnover - sales and purchases (9.5% in national survey)

3. 60% of bullock sales to buyers outside the village

4. Turnover seems related to consumption-smoothing

5. Farmers own too few

Puzzle: If bullock ownership is necessary and bullocks are useful also as

store of value, why do farmers hold so few?

Is it the problem of stocking out because of the inability to insure/borrow?

What would happen to bullock ownership and the efficiency of production if

insurance and/or an assured source of income were provided farmers?

Bullock model:

max E0GT$t(1/1-()(Ct - Cmin)

1-(

Bt = Gj"jDtj + Gj"jDtjTt Dtj = 1 if the # of bullocks (Bt) = j

Ct = Bt - pbbt+1 if Ct > Cmin

Ct = Cmin if Bt + pbBt < Cmin (stock out and bakruptcy)

Bt = Bt-1 + bt

Estimates and simulations

Risk-Coping and Marital Arrangements: India

The efficient risk-sharing model eliminates idiosyncratic risk -

independent shocks to households - but not aggregate community-level

risk

Given spatial covariance of risk, especially in agriculture, want partners

in risk pooling arrangement who are spatially separated

1. Household migrants: evidence from Africa (Lucas and Stark,

JPE 1986) that remittances compensatory

2. Patrilocal exogamy: marry outside the village

households export daughters, import daughters-in-law

risk-sharing is centrifugal l force for family: incentive to

spread the family spatially

(returns to specific experience induce a centripetal force:

incentive to stay together)

marriage creates a risk-sharing partner:

1. Creates inventive to risk share: member of household

A in household B, so if household A still cares about

the former member of A, A will help B in bad times

(altruism)

2. Creates a monitoring capacity: if the former A-

member still cares about origin household A, will act as

resident agent providing state verification

Implication of marriage as a risk-coping institution:

1. Marital partners will come from spatially-separated families: marital migration

2. Set of destination households for daughters will be spatially diverse: diversified

portfolio of marital partners

a. two daughters will not marry partners from either their own village or the same

village elsewhere

3. Households linked by marriage will be similar with respect to permanent

characteristics: assortative mating on capacity to provide help (recall borrower-group

self-selection)

4. If family groups can transfer information at less cost, given that marital search over

long distances is costly, marriages will be between family members

5. If the cost of marriage is positively related to its remoteness (search costs), then

households better able to self-insure will have shorter-distance marriages (wealth)

Empirical questions:

1. Does distance really matter for reducing risk covariance? How much separation is

required to obtain covariance reduction?

2. How important are inter-village transfers, as opposed to intra-village transfers?

3. What proportion of marriages are spatially exogamous, occur among partners from

different villages?

4. Are the marital portfolios spatially diverse?

5. Is consumption-smoothing really augmented by marriage? Is the consumption of

households with more marital partners more insured against income fluctuations?

6. Are households that face more income risk and are less protected have marital partners

that are more spatially separated?

Evidence from a survey of 115 marriages in 10 ICRISAT villages in 1984: locations of

husbands of all daughters of head and locations of origin of all daughters-in-law

Findings: 92.2% of married women came from or went to another village (exogamy)

average distance = 33 km (sd=60), maximum distance = 750 km

in households with two or more married women (daughters or daughters-in-

law) 94% not from or located in same village

only 14 of 115 marriages involved partners who were not blood relatives

in value terms 59% of transfers came from outside the village (26.6% of

loans)

Findings on spatial covariance:

use times-series information on daily rainfall, profits, wages in 6

ICRISAT villages

measure: distance between each of the 6 villages = 15 independent pairs

regress: correlation coefficient rijk for each variable k between village i and

village j on the distance between i and j for the 15 village-pairs:

rijk = $kdistanceij, where k = rainfall, profits, wages

Findings on consumption-smoothing and marriage:

compute variance in consumption Fc2 and farm profits FB

2 for each ICRISAT

farm household over 10 years

want to know the relationship between a household’s consumption variability

and its profit variability, and how it is mitigated by household characteristics

regression across households:

Fc2 = *1FB

2 + *2FB(wealth) + *3FB2(# of marriages) + *4FB

2(marital distance) + *5FB2(# of migrants)

dFc2/dFB

2 = *1 + *2(wealth) + *3(# of marriages) + *4(marital distance) + *5(# of migrants)

expect: *1 > 0; *2,*3,*4,*5 < 0

I. Micro Credit Institutions

Problems of the credit market in a setting with imperfect information:

1. Adverse selection (hidden characteristics). Banks cannot tell poor risks from good ones

among borrowers. Terms of loan affect who borrows and ability to repay. What is an effective

method of screening?

2. Enforcement. How do banks ensure that clients comply with terms of loan? Terms of loans

affect choice to repay. What is an effective mechanism to enforce repayment?

3. Moral hazard (hidden action). Loan terms create incentives to increase risk, and thus reduce

ability to repay. How can clients be efficiently monitored so they utilize the loan appropriately?

4. Verification costs. Costly to distinguish ability to repay from choice to default. How can

these auditing costs be minimized?

These problems acute among poor:

1. If a project fails, poor more unlikely to be able to repay loan.

2. How do you punish poor people? little to expropriate (“grim strategies”)

3. Ratio of transaction costs (verification) to loan size high (why are loans small?).

But, also:

1. Rural villages are typically small and the poor are immobile: communities have a great deal

of information about members.

2. Community may be able to impose social sanctions.

Credit sources: commercial banks, government banks, village moneylenders, other institutions

- micro credit

Micro-credit institutions exploit advantages of communities:

1. Require borrowers to form small groups in which all borrowers are jointly liable for each

other’s loans: joint liability lending

2. Intensive monitoring of clients.

3. Promise repeat loans for responsible borrowers as a group.

Grameen Bank

1. Lends to 2 million people in 36,000 villages (1999)

2. Lending funds obtained mainly from external institutions (subsidy?)

3. Loans made to self-selected groups of 5 people in the same village (optimal?)

4. Joint liability, with threat of no more loans in event of default

5. Compulsory weekly savings.

6. Surcharges on loans

7. 4 and 5 used to form emergency consumption insurance fund

8. Groups participate in non-financial activities (training)

Micro-Credit Issues

1. Why should we expect these institutions to solve fundamental credit problems? Theory

2. Are they necessary, compared to what? Contrast with collateral requirements. Theory

3. What is the evidence on:

A. Improvement in repayment rates.

B. Alleviating poverty

C. Solvency and sustainability: are these institutions profitable?

Framework

B = loan amount i = interest rate charged by institution

D = return on money institution can get if does not make loan

p = probability of project success (loan can be repaid)

u* = value of alternative use of borrower’s time

C = collateral (object of value that is forfeited if loan cannot be

repaid)

What makes for good collateral?

1. Appropriability 2. Absence of risk 3. Absence of

moral hazard

Examples: financial assets, animals, jewelry, land

Which is best?

Expected value of loan B made by institution (no-profit constraint):

E(B) = p(B(1+i) - B(1+D)) + (1- p)(C - B(1+D)) = 0

Participation constraint of borrower:

p(Y - B(1+i)) + (1- p)(-C) > u* (u* may vary across

borrowers)

Note: limited liability: cannot expropriate any monies if loan not

repaid, beyond C (may be due to poverty)

Roles of collateral and joint liability

Case 1: No asymmetric information or strategic behavior

C and i are perfect substitutes (both lender and borrower are

indifferent):

dE(B)/dC = (1- p) > 0 dE(B)/di = pB > 0

dv/dC = -(1- p) < 0 dv/di = -pB < 0

(but collateral requirement can be used to circumvent interest

ceiling)

Case 1: Adverse selection

Assume that there are two observationally-identical borrower types

a and b with different p’s: pa > pb

Problem: bank cannot distinguish, so bases loan cost on average of

p’s: low-risk borrowers subsidize high-risk; some low-risk won’t

want loans - repayment rates low as high-risk borrowers are

advantaged

A. Screening using collateral

a prefers loans with higher C and lower i compared to b

Bank can offer two different contracts: low C-high i and high C-

low i, lowering average risk of borrowers and increasing

repayment rates overall: collateral is a screening mechanism

But very poor have no collateral

B. Joint liability lending

joint liability lending substitutes for absent individual collateral.

Assumes that community members know who is a and who is b

Joint liability credit contract: if a borrower’s project succeeds, she

pays loan back but if her partner’s project fails, she also pays an

amount c to the bank - c is then collateral, pledged by partner

Expected value of i’s joint-liability loan with partner j:

E(v)ij = pipj(Y - r) + pi(1- pj)(Y - r - c), where r = Bi

Implication: positive assortative matching by riskiness

Expected gain from b partnering with an a rather than another b:

E(v)ba - E(v)bb = pb(pa - pb)c

Expected loss from a partnering with a b rather than another a:

E(v)aa - E(v)ab = pa(pa - pb)c

So a will never partner with b as long as c>0.

Bank can now screen by offering joint-liability contracts with

different i, c combinations. High-c offer leads to group forming of

low-risk borrowers, with therefore high repayment rates

Case 2: Strategic default and enforcement

Sometimes it is in the interest of borrowers not to repay even when

successful.

Decision rule:

Repay if Y - Bi > Y - C - S, where S=any sanctions

A. Enforcement using collateral

Borrowers now not indifferent between C and i:

Higher C lowers return to default; higher i increases default

gain

In high-risk (low p) settings with no collateral, banks cannot just

raise i in order to break even - that increases default risk, lowers

repayments; high i and low repayment rates go together

If no collateral, only recourse: threat of cut-off from future loans

B. Enforcement using joint-liability contracts

Joint liability credit contract: All in default if any member does not

repay, and no one in group ever receives a future loan

Decision rule now:

Repay if Y - 2Bi > Y - S

Does joint liability lead to higher repayment?

Consider two cases:

1. One group member defaults and other repays both: better

2. One group member defaults, the other is willing to pay

own debt but not partners: worse than individual liability

So benefit in this case not obvious, depends on distribution of Y’s

What are problems with threat of loan cut-off?

Requires credit monopoly - now Grameen bank’s monopoly

being eroded

Case 3: Moral hazard (p is now a choice)

A. Choosing risk with collateral requirements

Max {p(Y - r) + (1 - p)(-C) - 1/2kp2}

p* = (Y - r + C)/k

dp*/dr = -1/k < 0, dp*/dC =1/k >0

interest rate is like a tax on effort; increasing C induces more effort

But very poor have no collateral: lower effort and lower repayment

What is equilibrium level of p in the “collateraless” economy?

Take into account zero-profit constraint: p*r = D, then

sp*2 - Yp* + D = 0, so p* = (Y + sqrt(Y2 - 4Dk))/2k

B. Joint liability lending and risk choice

joint liability lending substitutes for absent individual collateral.

Assumes that community members can monitor group members’

effort

Again, joint liability credit contract: if a borrower’s project

succeeds, she pays loan back but if her partner’s project fails, she

also pays an amount c to the bank - c is then collateral, pledged by

partner

Assume that all choose the same p cooperatively, then the problem

solved by each partner is:

Max {p(Y - r) + p(1 - p)(-c) - 1/2kp2}

p* = (Y - r - c)/(k - 2c), with

dp*/dr = -1/(k - 2c) < 0, dp*/dc = (2(Y - r) - k)/(k - 2c)2 >0

What is level of p in the “joint-liability” economy?

Take into account zero-profit constraint: p*r = D, then

(k - c)p*2 - Yp* + D = 0

so p* = (Y + sqrt(Y2 - 4D(k - c)))/2(k - c)

Evaluating the Performance of Micro-Credit Institutions

Methods

1. Compare repayment rates and well-being of households in

villages with Grameen banks with households in villages

with no bank

A. Where are banks placed? program placement bias

2. Compare repayment rates, well-being of Grameen

borrowers and other loan takers within a village

A. What does theory say about those who form groups?

Theory says loan groups that form contain people

who are better risks - selection bias

3. Compare same Grameen loan-takers before and after the

introduction of the bank with non-loan-takers - no data

Example: Coleman, Journal of Development Economics, 1999)

1. Exploits phasing in of village joint-liability banks in NE

Thailand: sample of bank members and non-members in 14

villages, 8 of which had a bank for a number of years 6 were

assigned a bank but the bank had not set up. However, the

groups had formed in anticipation of the bank coming in

2. Compares members and non-members within villages

without a bank (control villages) to members and non

members in villages with a bank (treatment villages)

Assumes self-selected groups are the same, so only difference

is ability to obtain a loan - the treatment (72 outcomes)

Mobility, Economic Development and Growth

Contrast rural and urban India: both experienced economic growth, but

different patterns of mobility

Institutions play a key role in both contexts

Why Is Mobility in India so Low? Social Insurance, Inequality and

Growth (Munshi and Rosenzweig, 2005)

1, Increased mobility is usually a hallmark of economic development - individuals no longer

tied to land born on or occupations of their families.

2. India appears to be an exception among developing countries:

A. Occupation: For decades, caste-based occupations locked groups of individuals

to narrowly-defined occupations in Bombay

B. Migration:

1. India lags behind other similar countries: UNDP change in percent

urbanized, 1975-2000 by country

2. In rural India, rate of migration by men out of their village fell from 10% to

6% between 1982 and 1999.

Indeed, in most studies of Indian rural economy migration is assumed to be

absent (determinants of local public goods, returns and schooling choice)

0

5

10

15

20

25

30

35

40

45

China Indonesia India Nigeria

1975 2000

Figure 1. Change in Percent Urbanized, by Country, 1975-2000

C. Marriage: tradition to marry within sub-castes (jatis): restrict matches, given age

restrictions, to narrow pool.

1. Bombay: only 7.6% of individuals aged 25-40 in 1991 married outside their

jati in 2001

2. South Indian tea plantations in 2003: 6.2% out-married in same age group;

3. All rural India (except J & K state): 9.1% out-married in same age group.

Explanations for low mobility: ad hoc, incapable of explaining who is mobile

1. Low out-migration: tied to relative high rates of growth in agriculture

(Indian green revolution) - but accelerated growth in past 15 years in urban

industrial sector and a decrease in migration

2. Low out-marriage: tied to need for marriage between similar mates - but

inequality within jatis has risen, and yet no increase in out-marriage

STATE KARNATAKA

CODE CASTE1 ACHARAS2 ACHARS3 ADISAS4 AGASA5 ARAYARU6 BABU7 BEDIGAS8 BILLAVA9 BORI10 BRAHMAN11 CARPENTER12 CHRISTAIN13 DASOBU14 DEBAUGAS15 DEVOUGA16 EDIGAS17 GOLDSMITH (SUNAR)18 GOMOTARY19 GOUA20 GOWDAS21 HADIUA22 HORABALRU23 IJPAR24 JAIN25 JOGI26 KURBA27 KURUBA28 KUUABEE29 LAMBANI30 LARANIO31 LINGAITH32 LUGAYATUS33 LUNUATAT34 MADIDALAS35 MADRASI36 MARATHI37 MEDIGA38 MUSLIM (SAYED)39 MUSWAS40 NAI (BARBER)41 NAIKI42 NAUDABOBI43 NONDAVAS44 OKAURA45 REDY46 SCHEDULE CASTE47 SCHEDULE CASTE/SCHEDULE TRIBE48 SCHRI49 SIKH50 SWEEPER (BHANGI)51 VISHWA KARMA52 VODA BHAVI53 VYSHAS54 WASHMAN

Figure 2a: Distribution of the Number of jatis per Village, 1999

0

2

4

6

8

10

12

14

16

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

0

2

4

6

8

10

12

14

1950-59 1960-69 1970-79 1980-89 1990-99

Rates of Out-Marriage (Hindus), by Decade, Rural India 1950-1999 (N=31,529)

0

2

4

6

8

10

12

14

1970-75 1975-79 1980-85 1985-90 1990-95 1995-2002

Rates of Out-Marriage (Hindus), by Quinquennia, Mumbai 1970-2002 (N=5,406)

HYV Yields (Rupees per acre) and Real Agricultural Wages, 1971-1999

0

2

4

6

8

10

12

14

16

18

1971 1982 1999

HYV Yield/100 (1971 rupees)

Agricultural Wage (1982 rupees)

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

Between Villages Between Jatis Within Villages Within Jatis

1982 1999

Figure 2: Changes in Gini Coefficients for Household Wealth, 1982-1999Between and Within Villages and Jatis

Our explanation for low mobility:

Mobility reduces ability of risk-sharing networks to function

Out-marriage: lower ability to sanction via family

Out-migration: lower ability to sanction individual who is away

Those who cannot be sanctioned are higher risks, and will be excluded from

networks (Grief)

So mobile individuals lose network services

Risk reduction a fundamental problem of rural areas in India - many studies

Such studies reject perfect insurance, but high degree of smoothing

But Indian studies ignore caste networks - village is relevant group

Imperfect commitment models imply quasi loans used to smooth rather than

transfers

But literature on rural credit focuses on local money lenders and banks,

ignoring caste loans implied by limited commitment models

We show using new data providing both sub-caste affiliation and loans by source and

purpose for national panel of rural Indian households

1. Caste loans are as important as money lender loans in portfolio.

2. Caste loans have superior terms - lower interest rates, lower collateral

requirements.

3. Using consumption data, importance of jati as the risk-sharing group.

4. Consistent with a limited commitment model of risk sharing with wealth

inequality indicating who benefits least from networks, we find:

A. Higher-wealth households within jatis pay lower interest rates on caste

loans, but not on bank loans or money-lender loans

B. Households with out-married or out-migrant men are significantly less

likely to receive caste loans - but this is not causal

C. Exogenous changes in relative wealth within the jati affect the probability

of mobility - relatively more wealthy always exit first (identification strategy)

D. Overall growth has little effect on mobility, given networks, but growth in

inequality within jatis lowers mobility - overall inequality does not matter

Data Sets

1. NCAER ARIS 3-year panel: 4,118 households in 259 villages in 17 states surveyed in each

of three crop years: 1968-69, 1969-70, 1970-71

A. Information on consumption, income, location, HYV use, village infrastructure

2. NCAER 1982 REDS: 4,979 households in same villages, except for Assam: 1971 panel

A. Information on consumption, income, location, HYV use, village infrastructure

B. Information on loans, by source and use: obtained in reference year and outstanding at

beginning of period

3. NCAER 1999 REDS: 7,578 households from the 1982 households, except for J&K state

A. Information on consumption, income, location, HYV use, village infrastructure

B. Information on loans, by source and use: obtained in reference year and outstanding at

beginning of period

C. Sub-caste (jati) identified for all immediate relatives of head and spouses

D. Marital and migration histories for all immediate relatives of heads

Figure 3Location of ARIS-REDS Villages and District Boundaries

Caste Networks and Consumption Smoothing: Loan Data

1. Are loans from caste members a significant part of loan portfolios? Yes

2. Are caste loans important for consumption smoothing? As important as moneylender

loans

Use 1982 loan information: loans by source (caste, bank, moneylender) and purpose:

investment, operating expenses, consumption, contingencies (illness, weddings)

3. Do caste loans have superior terms compared with bank and money lender loans?

Use 1982 and 1999 loan information:

Average interest rates: lower than bank or moneylender

Percentage of loans with no interest charged: higher than bank or moneylender

Collateral requirements: smaller proportion for caste loans

Table 1: Loans Received by Source and Purpose, 1982

Loan source: Caste Bank Moneylender Other(1) (2) (3) (4)

Total loan value (%) 12.33 46.30 12.19 29.18

Loan value by purpose (%):

Investment 17.07 26.47 16.83 39.63

Operating expenses 6.08 53.47 1.82 38.63Contingencies 42.61 20.56 27.48 9.35Consumption 23.11 15.08 47.42 14.39

Note: N=1,423 loans received by the sampled households in 1982.Loan values sum up to 100 across the four sources in each row.Investment includes land, house, business, etc.Operating expenses are for agricultural production.Contingencies include marriage, illness, etc.

Table 2: Terms of Loans, by Source and Year

Year:Source: Caste Bank Moneylender Caste Bank Moneylender

(1) (2) (3) (4) (5) (6)

Interest rate 10.70 14.88 16.99 8.23 10.16 30.63(0.50) (0.47) (0.42) (0.91) (0.23) (2.30)

Percentage zero-interest loans 34.87 0.27 2.84 59.78 0.17 15.07

Percentage loans requiring collateral 16.23 48.95 18.99 1.31 83.21 24.78

Note: N=3,158 loans received by the sampled households in 1982 and 1999.Statistics are weighted by the value of the loan.Standard errors in parentheses.

1982 1999

Table ExtraTest of Perfect Insurance, Full Sample

Dependent variable: log own-consumption(1) (2)

Log own-income 0.321 0.321(0.032) (0.032)

Village log-consumption 0.783 0.784(0.035) (0.029)

District log-consumption -- 0.002(0.034)

R-squared 0.900 0.900

Number of observations 12,338 12,338

Note: regressions use three years of data 1969-71 for each household.All regressions include household fixed effects.Standard errors in parentheses are clustered at the state-year level.

Table 3: Tests of Full Risk-Sharing

-

-

-

Dependent variable: log own-consumption (All housholds) (1) (2) (3) (4)

Log own-income 0.184 0.174 0.171 0.172 0.168 (0.038) (0.040) (0.041) (0.041) (0.042)

Village log-consumption 0.735 0.725 0.635 0.576 0.638 (0.039) (0.041) (0.052) (0.057) (0.048)

Jati log-consumption - -- 0.232 0.216 0.239(0.038) (0.038) (0.041)

District log-consumption - -- -- 0.095 --(0.059)

Caste log-consumption - -- -- -- -0.024(0.080)

R-squared 0.883 0.824 0.826 0.826 0.825

Number of observations 5,394 3,543 3,543 3,543 3,387

Note: regressions use three years of data 1969-71 for each household.All regressions include household fixed effects.Standard errors in parentheses are clustered at the state-year level.Only jatis with more than 10 sample households are included in columns 1-4.Caste in column 4 measures broad hierarchical category in each state.

Mutual Insurance Models

Assume two identical (=wealth) individuals and two iid payoffs H (high) and L (low)

HH LL HL, LHA. 4 states of the world: P , P , P P

HH LL HL LH B. P + P + P + P = 1

HL LH C. P = P (equal wealth condition)

1. Perfect Insurance (Rejected in Townsend test, importance of loans for consumption

smoothing):

Consumption in any period = (H + L)/2 obtained via transfers of (H - L)/2

Ratio of marginal utilities equal across all states

2. No commitment (Coate and Ravallion, 1993):

In H,L state, incentive to quit network for person with H based on comparison of

gain from deviating H - (H + L)/2 versus future permanent loss of insurance

Transfers < (H - L)/2

3. Limited commitment (Ligon, Thomas and Worrall, 2002):

In H,L state, H person given promise of future compensatory transfers in all future

periods with equal payoffs by L, until new H,L or L,H state occurs

Expectation of future transfers in are such that in H,L state H does not deviate

Note: transfers are like loans in that they imply future payoffs, although state-contingent

What happens if partners are unequal in wealth? Two characterizations

1. Mean-preserving spread in wealth via change in probabilities

HL LH HL LHAssume P > P , with (ÄP = - ÄP ) so mean-preserving spread

Rationale: irrigation for some

A. Less wealthy individual is now more likely to be a net borrower

And, to maintain participation in the H,L state for the wealthier individual:

B. Future compensatory transfers must be higher when states equal (worse

loan terms for low-wealth L): interest rates on loans out higher

C. Transfers to wealthy when he is an L (borrower) have lower compensatory

transfers (better loan terms for high-wealth H): borr. interest rates are lower

2. Mean-preserving spread in wealth via change in payoffs in H states

Rationale: HYV availability for some

A. Wealthier agent better off compared to before (for given transfers) but

given declining risk aversion, as before benefits less from insurance in future

L state

So ambiguous results for loan position and interest rates on loans out

and in (note in first case declining risk aversion reinforces)

Now, introduce social sanctions imposed by network:

1. Raises cost of reneging, so improves loan terms for all

2. Ensures that those who leave the network will be less able to obtain loans

on favorable terms because those with lower possibility of sanctions are

undesirable insurance partners (Greif): cannot participate

Implications for Network Exit and Stability

1. Implication is that an increase in inequality requires that transfers be responsive so that

participation constraint is met for the wealthiest

2. Limits on responsiveness: A. Subsistence: consumption floor

B. Norms of redistribution: customary obligations of wealthy

3. If rigidity, then increase in inequality increases “tax” on wealthy - who are net lenders

If there is network exit, then will be by the wealthiest

Then scope for redistribution lowered for poorest, network gains less for them

4. Mobility, given sanctions, is associated with loss of network benefits - exit=mobility

A. Out-marriage: lowers capacity to sanction via families

B. Migration: lowers capacity to sanction individual who is away

Does mobility lead to loss of insurance network benefits?

1. Estimate “effect’ of exiting on benefits: caste loans, consumption variability

Correlation should exist between ability to obtain caste loans and exit, but

could be because those who are not able to get caste loans more likely to exit

2. Assess implications of model that incorporates loss of benefits:

those with least to lose (most to gain) exit (stay)

A. For given average jati wealth, those with greater own wealth more likely to

leave (relative wealth, more likely net lender) if adjustment imperfect, but:

1. Higher wealth households may have more outside options in “open”

marriage market too (reinforce)

2. Higher-wealth can afford to invest in mobility; opportunity cost

higher (oppose)

B. For given own wealth, increase in average jati wealth should decrease exit

Also show: higher average jati wealth, for given own wealth, increases jati loan access

Are the data on loan terms consistent with the model?

Wealthy within caste group should give and receive caste loans on terms that are more

favorable to them compared to the less wealthy for caste loans:

interest rates lower for loans obtained, higher for loans given out

Sample divided into two groups based on whether household wealth was

above or below the median within each caste group in survey period

Compare terms by within-jati wealth group also for bank and money lender

loans

Results:

1. Caste loans: Wealthy pay lower interest rates, lend at higher rates

2. Bank loans: Wealthy pay same rates as less wealthy

3. Money lender loans: Wealthy pay higher rates (monopsonistic

discrimination?)

Table 5: Interest Rates by Source and Household Wealth

Loan source:Wealth category: High Low High Low High Low

(1) (2) (3) (4) (5) (6)

Borrowing 10.08 11.98 12.09 12.08 28.78 14.22(0.83) (0.69) (0.25) (0.26) (2.04) (0.66)

Lending 10.83 9.20 -- -- -- --(1.82) (2.16)

Note: the household is the unit of observation.Interest rates are computed by pooling loans in 1982 and 1999.Standard errors are in parentheses.The cut-off separating low and high wealth is the median wealth level within the jati in each year.

Caste Bank Moneylender

Loan Access and Network Exit (Mobility)

Do those households in which immediate relatives have married out or migrated (men) receive

less caste loans?

Exit causes access to drop

Low access lowers cost of exiting

Out-marriage household: any immediate relative of the head (sibling, child) married someone

from another jati prior to the survey

Out-migrant household: any brother or son of the head left the village prior to the survey

Results:

Out-marriage household 30% less likely to receive a caste loan

Out-migrant household 20% less likely to receive a caste loan

Does exit cause loss of network services? Indirect tests

Table 6: Out-Marriage, Out-Migration, and Access to Network Loans

Reported statistic:Network exit: No Yes

(1) (2)

Measures of exit:

Married outside jati 6.17 4.76(0.25) (0.66)

Migrated outside village 6.30 5.27(0.28) (0.43)

Standard errors in parentheses.

Percent households receiving a caste loan

Specification and Estimation: Model implies that both own and jati-level wealth affect the

household’s equilibrium loan position (net lender):

it it t i itb = "w + $W + f + , "<0, $>0

itwhere b = household i’s loan position at t

itw = own wealth at t

tW = average wealth in the jati at t

if = household fixed effect

We estimate using two surveys (1982 and 1999) aggregating households at the dynasty level:

it it t it)b = ")w + $)W + ),

it it it itNote: cov()w ,), ) � 0 cov()W ,), ) � 0

So, use instruments to predict wealth change:

1. Land (acreage) inherited by the household prior to 1982

2. Initial (1971) conditions characterizing state of village HYV availability and use

Table 6: Descriptive Statistics, Panel Sample

Year: 1982 1999(1) (2)

Panel A: Loan Value by Source

Caste loans-in minus loans-out 44.21 41.34(31.55) (13.83)

Caste loans 71.42 81.72(11.43) (10.78)

Bank loans 393.96 235.39(89.54) (35.03)

Moneylender loans 47.77 46.13(7.61) (10.42)

Panel B: Marriage and Migration

Out marriage 0.07 0.09(0.01) (0.01)

Migration 0.10 0.06(0.01) (0.01)

Panel C: Wealth and Access to Banks

Household wealth 4831.91 20311.48(163.98) (1408.72)

Jati wealth 4609.11 20103.21(81.18) (1182.78)

Bank in village 0.19 0.36

Standard errors in parentheses. Statistics are computed using households in the 1982-1999 panel.Statistics computed using jatis with at least 10 sample households.

Table A: First-Stage Estimates

Dependent variable: HH wealth change

Jati wealth change

HH wealth change

Jati wealth change

(1) (2) (3) (4)

Inherited land 13.84 0.02(2.56) (1.47)

Inherited land (jati avg.) 47.98 77.81(15.56) (25.09)

Inherited unirrigated land -- -- 14.66 -0.44(4.20) (1.77)

Inherited irrigated land -- -- 13.63 3.61(6.13) (6.61)

Inherited unirrigated land (jati) -- -- 26.27 55.32(9.91) (19.13)

Inherited irrigated land (jati) -- -- 87.04 117.48(14.92) (45.97)

HYV in the village in 1971 x 10 1.66 -1.85 1.09 -2.78(2.80) (1.73) (2.61) (1.81)

HYV in the village in 1971 x 10 18.36 29.96 14.74 26.35(7.44) (11.92) (5.92) (10.77)

IAADP district x 103 5.72 11.56 3.42 8.92(3.84) (4.89) (3.30) (4.22)

Village bank in 1971 x 103 -0.33 -2.91 -0.65 -3.33(2.71) (2.98) (3.29) (2.89)

Table A: First-Stage Estimates

Bank change (1982-1999) x 103 -0.27 -5.00 -1.49 -6.20(3.79) (4.20) (3.56) (4.69)

F statistic 7.79 3.24 32.97 3.68

p-value 0.0008 0.0328 0.0000 0.0146

R-squared 0.087 0.198 0.100 0.219

Number of observations 2094 2094 2,094 2,094

Standard errors in parentheses are robust to clustering at the state level.Dependent variables are computed as the change between 1982 and 1999.All variables in the regression are excluded from the second stage except bank change (1982-99).Regressions restricted to jatis with at least 10 households in sample and households with heads at least age 35 in 1982

Table 7: FE-IV Loan Estimates

Loan source: Bank Moneylender(2) (3) (4)

Household wealth -0.009 -0.060 0.004(0.004) (0.031) (0.004)

Jati wealth 0.007 0.020 -0.006(0.003) (0.026) (0.004)

Bank in village 141.654 453.976 31.762(68.575) (340.690) (96.710)

Constant -73.021 219.995 12.409(75.098) (124.592) (68.736)

F-statistic (over-id) 1.46 1.39 2.14p-value 0.26 0.28 0.11

N 2,094 2,094 2,094

Standard errors in parentheses are robust to clustering at the state level.

Caste

Table 8: FE-IV Out-Marriage and Out-Migration Estimates

Instrument set:Specification:Dependent variable: Out marriage Migration Out marriage Migration Out marriage Migration

(1) (2) (3) (4) (5) (6)

Own wealth x 10-6 0.62 1.24 0.63 1.41 1.84 5.06(0.34) (0.55) (0.31) (0.57) (1.22) (2.32)

Own wealth* -- -- -- -- -1.24 -3.70Above-mean x 10-6 (0.95) (1.86)Jati wealth x 10-6 -0.93 -0.64 -0.88 -0.91 -1.19 -1.68

(0.36) (0.48) (0.34) (0.45) (0.61) (0.97)Bank dummy x 10-2 -0.70 -0.25 -0.69 -0.35 -0.72 0.15

(0.85) (1.71) (0.84) (1.73) (0.85) (1.63)Constant x 10-2 2.94 3.03 2.81 3.14 2.94 2.87

(0.83) (2.14) (0.86) (2.10) (0.71) (2.02)F-statistic (over-id) 0.56 0.25 2.23 1.07 0.93 0.79p-value 0.76 0.95 0.09 0.44 0.54 0.69

N 896 925 896 925 896 925

Standard errors in parentheses are robust to clustering at the state level.Regression use 1982 and 1999 data and are run using differenced variables.Instruments include inherited land, initial HYV adoption in the village in 1971, bank in 1971.

RestrictedSymmetric wealth effects Asymmetric wealth effects

Full

Implications of the Point Estimates

1. What is the effect of average wealth on mobility? " + $

Fourfold increase in wealth 1982-99:

Out-marriage: -0.4 percentage points

Out-migration: +0.75 percentage points

2. What is the effect of a within-jati wealth increase (hh)? "

Median standard deviation of within-jati wealth in 1999: 13,318 rupees

Out-marriage: 0.8 percentage points (9 percent)

Out-migration: 1.9 percentage points (32 percent)

3. What is the effect of a mean-preserving increase in within-jati inequality?

Transfer 13,318 rupees from below- to above-mean wealth households

Need responses to differ by own wealth: re-estimate

Out-marriage: -1.7 percentage points (19 percent decline)

Out-migration: -5.0 percentage points (84 percent decline)

4. Seeds of self destruction? Will selective exit, which reduces average jati wealth, lead to

more exit? $

Discard top 10% of households in jati with average wealth at the mean of all jatis

(20,445 Rupees) - reduction in average jati wealth by 8,852 Rupees

Out-marriage: 0.8 percentage points

Out-migration: 0.8 percentage points