Study of social networks in urban slums in India

Everybody in Annawadi talks like this-

“Oh, I will make my child a doctor, and he will make us rich. It's vanity, nothing more.

Your little boat goes west and you congratulate yourself, 'What a good navigator am I" 

And then the wind blows you east.

-Abdul’s father, 

Behind the Beautiful Forevers, by Katherine Boo[1]

Background

The Indian microfinance industry is one of the largest in the world. The gross loan portfolio at the end of March 2019 was reported at USD 1.5 million.[2] Microfinance, which evokes images of farmers and women in rural India, also has a rapidly growing urban footprint.[3] Much of this growth can be attributed to a rapidly urbanizing India. 

Estimates suggest that about 68% of the world would be living in urban areas by 2050.[4] The highest contribution to this number is projected to be made by Indian growth.[5] A bulk of this population is likely to live in urban slums, and constitute a large population that will need access to microfinance loans. 

The group loan model, popularized by Prof. Mohammed Yunus and the Grameen Bank is the most common across India, irrespective of geography. A group of 5 persons, usually women, come together to guarantee each other’s loan in the absence of collateral. Repayment rates continue to remain high and the group solidarity is expected to be reinforced through group meetings, held at a weekly, fortnightly or monthly basis. Microfinance institutions, in the absence of other information, often use attendance as a proxy to predict group repayment, or potential default. 

Urban India, however, has a completely different rhythm from rural India. Women in urban slums are usually employed in various home-based enterprises in the neighbourhood, or in their own homes, or employed as domestic help in neighbourhoods nearby. Work begins early and extends till late at night, with women often bearing the burden of housework and income generation to supplement the household income. 

This makes it hard for customers to attend group meetings held by microfinance institutions during the day. While everyone is employed in small scale enterprises, or in jobs that gives them a small but regular salary, these jobs are precarious. Often a day off from work would result in loss of income, and regular absence could mean a loss of work. The insistence on attending group meetings, therefore, imposes a time cost and psychological costs, which are over and above the financial cost of the loan. 

Despite these additional costs, demand for microfinance loans is fairly high. We also know that the financial behaviour of these households is fairly complex[6], with multiple formal and informal loans, erratic income, regular expenses and unexpected shocks. Microfinance loans are often used for cash flow smoothening, to tide over unexpected shocks, demands from homes back in a village, payment for school fees or to repay other high cost debt. 

While the repayment of microfinance loans even in urban India continues to be regular, the microfinance industry association has begun to worry about low/declining attendance rates. The research question is therefore, are there other attributes that can predict group cohesiveness and willingness to repay on time? 

Decoding the microfinance model using game theory and social network analysis

The traditional Grameen bank model of micro-lending has elicited surprise and interest from academics all over the world because of the high repayment rates associated with it, despite the absence of traditional collateral. The model relies on social collateral, i.e. a loan is provided to a group, usually of 5 women, who guarantee each other’s repayment. In the event of default by any one member, the others pitch in to contribute and make sure that a repayment is made on behalf of the defaulting member. Since this is not likely to continue forever, the group is incentivized to self-select to make sure they don’t have to deal with an errant borrower. The microfinance institution, in effect, outsources its credit appraisal to the group and therefore lowers the cost of appraising each individual, and overcomes the problem of information asymmetry. In terms of social network analysis, if we assume A,B,C,D,E are members of a group. The guarantee structure would be represented as in fig 1 below.

Fig 1: MFI group and social collateral

The microfinance institution is not incentivized to check for any attributes of the group besides verifying their ‘Know Your Customer’ documents (id proof), as mandated by regulation. 

Since the repayment relies heavily on trust, there has been some attempt to explain the high repayment rate in terms of game theory in academic papers. Pelligra (2009) sets up the case as a sequential trust game between the lender (L), who decided to give or not give a loan to a borrower (B), who can choose to default or repay.[7] However, this assumes that the group self-selects effectively and has enough information about others’ behavior to do so. It also assumes that the payoff due to social ostracism, in case of default, is so negative that it would act as a dissuading factor.

These assumptions change in an urban context, where an urban slum is mostly composed of immigrants. The absence of deep roots and lack of ownership of homes, unlike in a rural context, can allow people to shift rapidly and disappear into the anonymity of the city, altering the negative payoffs that the model assumes. This in turn, reduces the risk of social ostracism. Credit bureau history could eventually act as a dissuading factor but the cost of recovery is high. As an alternative, firms could ask customers for more information, but this imposes high costs on customers since it affects their privacy. It also imposes costs on the firm that now has to deploy resources to analyse the additional information which may or may not be predictive of the credit default risk. Additionally, increasing competition between firms in urban slums means that groups may come together without necessarily knowing each other, since there is a high demand for credit and a reasonably high supply which can dilute group formation criteria. 

In fact, in game theory terms, there are possibly two sets of games being played. The first is a simultaneous game between a group member and all other members of the group, who choose between defaulting and not defaulting. In terms of payoffs, this can be modelled as seen in figure 2. The second is a sequential game played between the microfinance institution and the group, where the firm relies on group attendance as a proxy indicator for group cohesiveness. 

In the simultaneous game, we assume that the payoff from any group member defaulting is higher to the person/people defaulting vs. the person repaying. The payoffs from repaying for the entire group is possibly the same as defaulting, since it means both parties have the same result in both cases i.e. they are denied further loans, but don’t repay the current loan or they both get further loans (on repayment).

Figure 2: Modelling group repayment

		Other group members
		Repay	Default
Group member A	Repay	(0,0)	(-1, 1)
	Default	(1,-1)	(0,0)

The game as it is currently set up would have a single pure strategy Nash equilibrium at both group member A and other group members defaulting. This means in a situation where the group members do not know each other, it is highly likely that they would default. Therefore, attendance is probably not the best proxy indicator to predict default.

Microfinance institutions possibly know this risk exists, and are likely to build in costs to help mitigate this risk. Since the interest rate in India is capped by regulation at 26% p.a., these costs are likely to be non financial, and psychological which include the threat of not receiving further loans from the same institution, or poor credit history and the low likelihood of being able to borrow from any other institution as examples. 

While these might work, it creates a sense of distrust between customers and institutions and incentivizes customers to find ways to game the system. Change of names, misspelt names, borrowing by another family member are ways in which it is possible to bypass the current system. It is also likely to lower loyalty displayed to a particular institution, since the relationship is purely transactional. This in turn lowers business value by increasing acquisition costs for MFIs.

Finding alternate attributes that would help explain group cohesiveness besides attendance at group meetings, would help ease the pressure on both customers as well as microfinance institutions that are beginning to operate in a rapidly urbanizing India. 

The role of social network analysis

Social network analysis can help play a role in identifying attributes that better serve the interests of both microfinance institutions and customers, and offer a more customer centric way of lending.

Research on the money management practices of poor customers by Rutherford & Arora (2009) and CGAP[8]suggest that poor people often cultivate relationships with others- family, friends, local stores, neighbors- who can help them in times of need.[9] This means that even in an urban slum, people are likely to have sub-groups that are connected to each other by features like attending a community festival, contributing for communal events financially or otherwise, borrowing and lending informally, buying and selling to each other, or offering credit if they operate local stores. While these financial transactions are hard to track in a survey unless we take a 'financial diaries' approach, the existence of this network also indicates that information is likely to be exchanged in the same network.

We hypothesize that identifying people in the slum who are also borrowers from the microfinance institution, who exchange information and analysing this information would help identify key persons who could influence a higher repayment or default, depending on their roles in the community.

Proposed methodology

(Subject to approval by the IRB)

It is proposed to conduct a survey amongst 20 groups (100 customers), selected randomly from the customer base of Centrum Microfinance Private Ltd (CMPL)[10], a wholly owned subsidiary of the Centrum Group of companies. CMPL works predominantly in urban and semi-urban areas in India. As a late entrant in the Indian microfinance space, the company began operations by acquiring an existing portfolio from First Rand Bank, in the bustling, crammed and chaotic neighbourhood of Dharavi, an area in Mumbai, popularized by the movie Slumdog Millionaire. 

Dharavi, however, is home to thousands of informal businesses with an annual turnover of $1 billion by some estimates.[11] The area has squat tenements that house both residences and businesses, and is home to multiple communities and linguistic subgroups who have migrated from all over India. 

CMPL faces the same challenge of low attendance rates that other urban microfinance institutions have experienced. However, they also see slums as ecosystems that are buzzing with activity and composed of a series of micro-supply chains. CMPL desires to find attributes that can indicate the extent of social capital that exists within these communities, which could include shared languages, shared norms, cultures, business interactions, friendship, etc, or others, and could act as predictors for loan repayment instead of attendance, and is willing to share how this could result in a potentially new way of delivering micro-credit in India.

The survey will be restricted to these groups, all residents of Dharavi, to assess who are the top 3 people they receive information from, provide information to. We will also ask for frequency of interactions with others in the cohort. 

CMPL works with customers who are literate and can read Hindi/Marathi, so the survey instrument can be translated into the local language and administered over smartphones. Responses will be gathered in a google doc at the back end and will not be accessible to CMPL staff, who will only make sure that the survey is completed by the respondents. 

Operating with the hypothesis that some nodes may emerge as more influential than others, we propose to then run a series of centrality and sub-group tests to measure influence in terms of information flow.

The analysis would then be as follows:

1)    In-degree centrality: These are likely to be people who enjoy popularity or are considered people with some level of prestige within the slum. By extension, they are unlikely to default, since it would mean a loss of face. For an MFI, these are important customers to build loyalty with, and convert into possible brand ambassadors. If there is a risk of default, these are the people who are likely to mitigate the risk

2)    Betweenness: These are people who are likely to act as information brokers, and are an important constituent to identify when launching operations in a new area, or launching a new product/service. They are ideal to spread word about the MFI and again important to nurture as loyal customers

3)    Eigenvector centrality: These are people who enjoy informal leadership and would possibly make good group leaders. However, these are also likely to be people who are connected to other power structures in the slum, and could easily influence mass default, especially if they are connected to local politicians. For an MFI operating in a slum, it is helpful to know who these people are, and incentivize them to act in favor of the MFI.

4)    Closeness: These are people who are close to many other people and are capable of quickly mobilizing the network. Again, these are people who can both support repayment or incite mass default, and need to be incentivized accordingly

5)    Subgroup analysis would indicate the homogeneity of the various groups, sub-cultures and possible cliques

Based on the result, recommendations would include unpacking the ‘black box’ of the group to identify specific people both within the groups and across other groups, who could act as important stakeholders to both MFI and customers. This could possibly entail some alterations to the existing group model followed by the MFI. 

Limitations

1.     Contingent on IRB approval: The final project is due on Dec 7, so the survey will need to be conducted by Nov 15. If the IRB approval does not come through by this date, it may not be possible to conduct this research

2.     SNA analysis might have to be backed by qualitative surveys that the MFI is expected to conduct to verify the hypothesis and take next steps

---

[1] Boo, Katherine. Behind the Beautiful Forevers: [life, Death, and Hope in a Mumbai Undercity]. New York: Random House, 2011

[2] https://www.livemint.com/industry/banking/microfinance-industry-posts-38-growth-in-2018-19-1559564596042.html, accessed on 20 October 2019

[3] https://www.livemint.com/Politics/tPNyKYelixWcbXsiiRdMWI/Urban-MFI-borrowers-outstrip-rural-counterparts.html accessed on 20 October 2019

[4] https://www.un.org/development/desa/en/news/population/2018-revision-of-world-urbanization-prospects.html, accessed on 20 October 2019

[5] https://www.un.org/development/desa/en/news/population/2018-revision-of-world-urbanization-prospects.html, accessed on 20 October 2019

[6] Collins, Daryl, Jonathan Morduch, Stuart Rutherford, and Orlanda Ruthven. 2009. Portfolios of the poor: how the world's poor live on $2 a day. Princeton: Princeton University Press.

[7] Pelligra, V., (2011), Trustful Banking. A Psychological Game-Theoretical Model of Fiduciary Interactions in Micro-Credit Programs, in Sacconi, L. & Degli Antoni G. (Eds.), Social capital, Corporate social responsibility, Economic Behaviour and Performance, London: Palgrave MacMillan, pp. 80-100

[8] https://www.cgap.org/research/publication/how-poor-manage-their-money

[9] Rutherford, Stuart, and Sukhwinder Singh Arora. The Poor and Their Money: Microfinance from a Twenty-First Century Consumer's Perspective. Warwickshire, UK: Practical Action Pub, 2009.

[10] The student who is proposing the study is an independent director on the board of CMPL and has checked with the management on their willingness to conduct this study

[11] https://in.reuters.com/article/india-landrights-slum-idINKCN12B1KY, accessed on 20 October 2019