Project Everest

Proposed Experiment

[Problem/Proposed Experiment]: SoCon Fiji - Is microfinance forgetting about the people? - June 2018

The Problem
One of the core challenges facing the microfinance industry is how to accurately and efficiently model credit scores - in other words, how can we best predict the whether someone will pay back their loan? This often comes down to: how much do you know about the consumer (read: what data do you have), and how accurate is this information in telling you about their ability and willingness to repay?

In the developed world, banks (being the main sources of loans) have access to troves of financial transaction data about you - and they use this data, in conjunction with other demographic factors, to predict your credit worthiness. However, in developing countries where a significant portion of the population are unbanked, this sort of data is hard to come by.

In such a data-poor environment, microfinance institutions are forced to price in uncertainty premiums on loan rates. Once touted as the big revolution in development, recent studies into microfinance suggest that high interest rates (50+%) and overbearing repayment schedules (can be weekly) has limited its effectiveness. This is one of the paradoxes of finance - the cost of borrowing for an average Australian is 5-10%, while the rural farmer in Malawi is charged over 50%.

Idea Origination
So, the question is: how do we find out more about the people we are lending to? The current trend in microfinance fintech (financial technology) startups is alternative data. These are often non-financial sources of information such as agricultural, social network, and telecommunications data, which can help predict credit worthiness. But, have they missed a trick by overlooking the closest source of information to the lender?

What if people you know can help guarantee your ability to repay? Now, this is not something new - microfinance institutions have enacted tactics like group lending to redistribute risk. Group lending is where a loan is given not to one individual, but a group - if one person within the group cannot repay their portion of the loan, the others in the group make up the payment amount. The strength of this idea is that it engages communities and social ties. It makes sense that the people in your network are likely to know whether you will be able to repay a loan. 

However, this process has not been digitised (and hence costly), so its scalability has been hampered. In addition, group lending is inefficient in the fact that the loan rate is determined by the likelihood that the group will repay, not per individualSo how can we digitise an individual-based lending model based on how other people close to you judge your ability to repay?

What It Could Look Like:

1. You open up an account and are given a certain amount of ‘reputation points’.
2. The more ‘reputation points’ you have, the more money you can borrow and at a lower rate – it essentially proxies a credit score. For each level of loan amount, there will be a certain amount of 'points' you'll need. You can either accumulate these points yourself and/or borrow points from others. When you get a loan, these points are then locked away until you pay back the loan.
3. So, how do you build up your ‘reputation score’? Through a few ways: 
  a) You earn a certain amount of ‘reputation points’ by repaying a loan you have borrowed. The greater the loan, the more points you will receive back. 
  b) You can ‘sponsor’ your friends & family when they apply for loans – when they pay back, you earn back your points plus some. 
  c) By connecting your account to “hard metrics”/financial data/or other sources of credit information, you can also earn more ‘reputation points’.

What this system effectively does is put the credit risk modelling in the hands of those who know you best. On top of this, the people who have the most points (i.e. the ones who have proven themselves to be reliable), have the most influence in the system. This, in conjunction with other sources of data, can help increase the predictability of the model. Currently, there is no such service on the market to my knowledge. 

Obviously, how this will be implemented will be key. For example, the economics around the value of the points, the incentives to sponsor a friend in their application, the reward for repayment, etc. are important considerations in developing this out.

While this is something I've only recently come up with (and a lot more work needs to be done), I would love to get some feedback. Thanks for taking the time to read this.

Tagged users
edited on 5th September 2018, 23:09 by Justin Hakeem

Andrew Vild Jun 19, 2018

Awesome idea you have.

Key questions that need to be answered to make this happen in practice:
- What work needs to be done in order to make this happen?
- What is the critical data that can easily be collected to start creating risk profiles?
- What is the most basic MVP that would allow this to be tried and tested so the iteration process can begin to occur?

My opinions on this are:
1.Work that needs to be done:
- old-school data collection of a set data size to understand access, affordability and understanding of the microfinance space
- competitive analysis and case studies of other organisations that may have already achieved this in Africa or Asia that just hasn't caught our attention.
- reach out to ambitious, but low hanging fruit in terms of key partners, stakeholders or people with critical information

2. Critical data
- National Statistics and Surveys that have been completed previously
- Success stories of other companies doing similar thing
- Local telecomm data sets on their users spending and habits
- Local banks statistics and reports on loans, repayments, spending and habits (understand that we will have demographic data on "banked" individuals as opposed to "unbanked")

3. Basic MVP
- Micro loans through digital money eg. Airtel Money in Africa
- Physical loans with follow ups
- Loans through community elders

Challenges exist in how to create a risk profile, how to measure risk, how to build trust and credit history.

Users tagged:

Reply 3

William Lee Jul 1, 2018

Yep hit the nail on the head here.
How to collect data and how to collect repayments are the critical challenges to solve. The basic MVP sounds doable and will illustrate the potential of payment/collection systems nicely.

Reply 0

Ella Grier Jun 21, 2018

I like the idea Will. There are definitely many group lending models for MFIs operating in developing contexts and it is almost more popular than financing on an individual basis for exactly the reasons you have described.
What I like is the incorporation of a proxy credit score and how, cumulatively, those seeking financing can meet a ‘threshold’ score for a particular type of loan. It seems like a system that would work and I think our biggest challenge would be creating that entry level ‘score’.

A significant risk factor that could be incorporated into this is financial literacy. RangDe, one of the bigger MFI’s in India, ran a pilot program that assessed the performance among micro-credit borrowers and saw vastly higher repayments rates from those who had received business mentoring during their loan cycles. As a result, they established financial literacy kiosks in its target communities that are designed to suit local needs. We could digitize educational modules to be undertaken before loans as a means of increases their credit rating. It would be in the interest of those seeking finance as well as those acting as guarantors to be able to provide advice.

Reply 5

Wade Tink Jun 29, 2018

Love this. How do we create an experiment out of it to test first assumptions? My suggestion is that a post be created as a proposed experiment for the micro-finance team in July Malawi. Starting with what is to be learnt, how will it be measured and what method will achieve it.

Reply 0

William Lee Jul 1, 2018

Hey Tink, the attached file is a summary of what I believe is required to get a microfinance project up and running.

I think the social score element to this is something to be overlayed once we have the foundations set up.

The way to validate/test this social score idea in July would be to provide loans with one group requiring social proof and another without. Given there are only 4 weeks, it may be hard to see immediate results from this.

Given this, I think priority should be given to building the foundations of data collection and payment systems first, before delving into the complexity of credit scoring just yet.

Reply 0

Andrew Vild Jul 3, 2018

Status labels added: Proposed Experiment, Under Review, Problem

Reply 0

Liv Hendy Jul 4, 2018

Can't engage in this discussion fully on how to action these points but just wanted to add a potential gain that exists in actually getting people to engage with these token systems and implementing them. Concepts like these have worked before, but only if we can find a way to piggyback on existing cultural norms and just supplement those interactions for a representative token (ie your reputation score).

Token reinforcement (within the scope of Behavioural Economics) has been a powerful tool in order to shape behaviour that exists externally to economic transactions but has a direct relation to economic productivity. For mitigating risk this would use those existing social and cultural expectations and pressures to repay debt, but again these systems rely on these underlying notions of cognition that responds to these tokens as if they were inherently valuable (accepting that these reputation points cannot be a volatile token as they represent credit between us and our consumers).

I know its a bit left of centre, but highly recommend some behavioural economics studies that have been conducted roughly within this space, it's something that I'm passionate about. Currently these processes are used heavily in the health sphere (with treating addiction, autism and behavioural issues) but we can clearly see cultural associations within the sphere of borrowing and lending, there should be no limit why these fundamental human reactions can't bolster our opinions.

Reply 2

William Lee Jul 4, 2018

Sound awesome Liv, can you send me through these studies when you have the time?

Reply 0

Liv Hendy Jul 4, 2018

Alot of it has come from my brain (no references needed) but this website explains it in a nutshell -

and have attached a thick study on this whole topic. Interesting because its linked so closely but cannot find any evidence of token economies linked specifically with micro-financing models

Reply 1

View all replies (2)

Amber Johnston Jul 4, 2018

Jay this is very relevant to the credit score ratings you were exploring the other night. I would be interested in how your ideas work with this.

Users tagged:

Reply 0

Kurt Michl Jul 6, 2018

Awesome Will,
I would love to see if we can test this concept on the ground in Malawi!

Reply 0

JayP Jul 6, 2018

Cool idea!

I think there is great value in a reputation based system when credit modelling. It works well as a measure of credit risk as this gives another non-traditional/alternative data source improving the predictive capacity of the model.

Human behaviour from Economics point of view would greatly suggest that this incentive based system would give an accurate picture of not just the individuals but the surrounding individuals' levels of risk. i.e. someone who is known to be untrustworthy will not have many around them staking their financial reputation on them.

Furthermore, the somewhat gamification of this could make it more accessible to a wider market as well as giving it an ability to self-sustain if matched up with some machine learning. In this way, digitisation is reached when it comes to informational asymmetries problems as a person's reputation level will act as a signal to financiers. This can be compared to how a choice of excess level is a good indicator/signal of what a person believes is their true riskiness to an insurance company.

Reply 1