Subjective Expectations and Income

Processes of Micro Finance Clients in

India

Britta Augsburg (EDePo @ IFS)

Orazio Attanasio (ICL, IFS, NBER & CEPR)

Britta_a(at)ifs.org.uk

March 2010

Introduction and Motivation

This work is part of a research agenda on the measurement and use of subjective expectations data.

In this particular case we will use elicited probability distribution to estimate the features of the stochastic process for total income faced by the (potential) clients of a MFI.

In the future we plan to relate these features to various types of behaviour

We are also planning to extend the measurement to get at asymmetric information

Measuring expectation

The direct measurement of subjective expectations has recently being advocated strongly (Manski,2004).

Delavande, Gine & McKenzie (2008), Attanasio (2009) show that even in developing countries it is possible to elicit sensible data on the prob. distribution of stochastic variables sa income.

An important feature of many of the new contributions is an attempt to estimate the entire probability distribution of subjective expectations.

This potentially allows the estimation of more credible models that avoid strong assumptions (such as rational expectations).

In this particular case we will use elicited probability distribution to estimate the features of the stochastic process for total income faced by the (potential) clients of a MFI

Some words of caution

The direct measurement of subjective expectations is not without problems

A standard question or methodology that works everywhere does not exist.

Extensive piloting on the wording and ingenuity of the measurement tools is necessary.

Common features:

It is often easy to start eliciting the range of variation

Eliciting the distribution of probability over the relevant range is hard.

Examples to convey the concept of probability are important (rain, games of chance etc. )

Visual representations of the probability are useful (rulers, stones etc. )

Not only income

While (total family) income has often been the object measured by expectations questions, it is by no means the only one.

Many other examples exist:

Inflation

Effectiveness of contraception

HIV infection

Stock Markets

Profits

Particularly interesting in our context and for developing countries is the measurement of expected returns to investment.

Expected returns to education

Dominitz and Manski (1996): Wisconsin students

Attanasio and Kaufmann (2008): High school students in Mexico.

Returns to investment financed by MFI

Outline

The data: A survey of (potential) MFI clients in

Anantapur, Andhra Pradesh, India

The expectations questions

Validation of the questions.

Estimating income processes using the

expectations questions

Extensions in the near future

Conclusion

The Survey

The survey: was collected within an attempt to evaluate the impact of the loan and insurance products.

The institution: BASIX India

The area: Anantapur, Andhra Pradesh, India

The loan product: mainly targeted to the purchase of buffalos to start a dairy production (group sizes: 4-7), partly bundled with health insurance product (for buffalo)

The survey included: 'treated' villages (i.e. where BASIX, was operating the new loan

product)

'control' villages (where the program had not been started yet) –(comparable, chosen by BASIX staff)

The Survey – Cont.

Round 1

Jan-March 2008

Sample Size: 1041 (half BASIX customers)

36 ‘treated’ and 30 ‘control’ villages

Round 2

April-June 2009

Re-interviewed: 951 (91%)

Questionnaire:

extensive (~1.5hrs)

substantial expectations module (including questions about expectations on income and income/costs from/of dairy activities)

The Expectations Questions

Expectations questions were asked for: Total family income;

Income from dairy activities;

Health costs for the buffalo.

As in many expectations survey, for each of the relevant variables, we first elicited the 'range'

This was done with two questions. Maximum: Imagine that you have a very good year, every member of working age in

the household managed to have work, and there were no droughts or anything the like. What would be the maximum amount of income your household would receive in such a situation in one year?

Minumum: Now imagine the total opposite: the harvest is bad; animals get sick, finding work is not possible. What would be the monthly income of your household in such a situation?

The resulting interval was divided into three subintervals, determining three cut-off points: A, B, C.

Min MaxA B C

The Expectations Questions – Cont.

Having established the cut-off points we ask the value of the distribution function for each of these points.

In particular, for each point we ask the probability that the relevant variable is at least as high as that point.

To be able to do that we need to introduce the concept of probability.

This is done by giving the following example: We have a ruler here with a scale from 0 to 100. We will use this as an

indicator of how sure you are that it will rain in the future.

Having practiced the rain examples (we have data on that) (including probability of rain next week and next month)

prompting for increasing answers

... we move on to ask the question on the C.D.F.

The Expectations Questions – Cont.

How likely do you think it is that your income in the coming year will

be higher than (A/B/C) Rupees?

No prompting (for decreasing probabilities) was allowed

Having obtained a few points of the CDF we estimate the entire

distribution function by making some assumptions on the nature of the

distribution.

In particular, we assume that the distribution is piecewise uniform

2008 2009

min max min max

Validation

Given the difficulty in answering the questions we

need to validate them

Response rates

Internal consistency

Bunching

Some but not excessive

Mainly at 50

(problematic for log normality!)

co-movements with other variables.

Validation – Internal consistency

Response Rates / Logical response errors

2008 2009

Total no. Of Obs. 1041 951

Information on Income 1039 950

Min/Max given 1030 950

% not given but answers to min/max given 15 6

total 29 6

wrong violation of monotonicity 5 17

wrong 'direction' 2 4

one prob missing 4

total 7 25

TOTAL no. of Obs. Available1005

(96.5%)

919

(96.6%)

Validation - Bunching

0

100

200

300

400

500

600

700

0 5 15 25 30 40 46 55 60 68 75 78 85 90 96 100

thr A 2008 No thr B 2008 No thr C 2008 No

0

50

100

150

200

250

300

350

400

450

500

0 5 15 25 30 40 46 55 60 68 75 78 85 90 96 100

thr A 2009 No thr B 2009 No thr C 2009 No

Bunching of percentages

(probability of income being higher than threshold A / B / C)

Validation – Correlation of probabilities

Correlation of rain and income probabilities

(2009 sample)

Total Household Income

Full Sample

Households whose primary

income is not derived from

agricultural activity

probability of… prob>A prob>B prob>C prob>A prob>B prob>C

…rain tomorrow-0.05 0.08* 0.18* -0.17 -0.12 -0.02

0.16 0.03 0.00 0.12 0.26 0.85

…rain within the next week0.02 0.08* 0.08* 0.00 -0.01 -0.03

0.51 0.02 0.02 0.99 0.94 0.77

…rain within the next month0.06 0.04 -0.05 -0.06 0.06 0.02

0.08 0.22 0.10 0.60 0.60 0.81

Validation – Co-movements

Co-movements with other variables

(current, ‘typical’, expected income – 2008 and 2009)

0.2

.4.6

.8

De

nsity

6 8 10 12 14 16log of total income (y_tot)

Total 2008

Typical 2008

Expectation 2008

kernel = epanechnikov, bandwidth = 0.1506

Kernel density estimate

0.2

.4.6

.8

De

nsity

8 10 12 14log total 2009 income

Total 2009

Typical 2009

Expectation 2009

kernel = epanechnikov, bandwidth = 0.1134

Kernel density estimate

Estimating income processes

The properties of the stochastic process generating

individual income (persistence, variability) are important

to study consumption and saving behaviour

Subjective expectations data can be used to estimate the

dynamic process of individual income

Attanasio and Di Maro (2008), Mexico,

This paper, India.

Modelling individual incomes

Modelling individual incomes - continued

Estimating the income process

Results 2008

Estimating the income process

Results 2009

Estimating the income process

Results Pooled

Estimating the income process

Results First difference

Estimating the income process

Results Difference

Measuring asymmetric information

A potentially important application of the methods to measure expectations is to try to get to asymmetric information.

Consider individuals A and B

In a village, In a micro lending group, ...or could be the loan officer and a client

One can ask

individual A about her own future income and about B's future income

...and B about her own income and about A's future income.

Compute the variance of A's income as perceived by A and as perceived by B

B's perception of A's income should have a larger variance

The difference between the two is an index of the extent of asymmetric information

Conclusion

We have used subjective expectations to estimate the income processes of MFI clients.

In the processes we have validated the expectations data

We will relate properties of the income processes (at the individual and group level) to repayment behaviour

.... and investment behaviour

We plan more work on measurements and use of asymmetric information.

Further thought is necessary to get at issues of moral hazard.

Expected and Current Income

Obs

Income Ln Income

min max median mean std.dev min maxmedia

nmean

std.de

v

2008

Typical 1013 0 3,200,000 50,000 71,184 122,112 0.00 14.98 10.82 10.81 0.89

Total – last year 1019 0 637,000 55,400 72,901 63,505 0.00 13.36 10.92 10.92 0.80

Expected Average 995 1,388 1,145,000 56,250 73,948 70,547 7.21 13.41 10.91 10.90 0.76

Expected Std. Dev. 995 1,502 8,189,208 74,418 119,947 321,103 0.06 9.82 1.46 1.69 1.37

Min 1019 1,000 500,000 40,000 54,112 49,074 6.91 13.12 10.60 10.59 0.81

Max 1019 1,500 800,000 70,000 88,921 73,088 7.31 13.59 11.16 11.15 0.72

2009

Typical 918 3700 720,000 60,700 70,571 52,443 8.22 13.49 11.01 10.98 0.64

Total – last year 918 3700 720,000 60,100 70,502 55,173 8.22 13.49 11.00 10.97 0.64

Expected Average 918 1,475 542,500 59,000 67,513 47,778 7.28 13.20 10.96 10.90 0.68

Expected Std. Dev. 915 2,650 2,042,071 80,065 95,497 100,714 0.04 7.25 1.53 1.57 0.78

Min 918 1,000 1,000,000 40,000 48,847 50,825 6.91 13.82 10.60 10.53 0.75

Max 919 1,100 600,000 70,000 81,265 56,417 7.00 13.30 11.16 11.10 0.69