Human-Centered Risk Management
We have been seeing some really exciting growth on the Kiva Zip team recently. October was our biggest ever month for U.S. loan volume by 23%, and the chart below, plotting monthly loan volume since the inception of Kiva Zip in November 2011, is an encouraging one.
But in running and growing a lending program, we can never take our eye off the ball when it comes to risk management, maintaining a robust repayment rate, and repaying the generosity and dollars of our lenders, without whom there would be no Kiva Zip program.
This blog post aims to give a pretty comprehensive overview of how we think about Risk on the Kiva Zip program. It is broken down into four sections:
1) Our philosophy towards risk management;
2) The evolution of our repayment rate over the last four years, and the corresponding evolution of our approach to risk management;
3) How we analyze risk on the Kiva Zip team;
4) Some insights we have gleaned from the 1,600 U.S. loans that have begun paying back on Kiva Zip (or not!).
1) Our philosophy towards Risk
Our highest vision for the Kiva Zip program is to reimagine a financial system in which human relationships and interpersonal connections are restored. When it comes to risk, this vision is manifested in two principal ways.
Firstly, whereas conventional lenders make lending decisions based on financial data (credit scores, cashflows, collateral, etc.), on the Kiva Zip team we aim to make lending decisions based on the strength of a borrower’s character, and their standing in their community. We call this approach “social underwriting”, as opposed to the financial underwriting practiced by conventional lenders. There are several components of our social underwriting:
• The relationship between a borrower and their “Trustee” – usually a non-profit that supports small businesses in their community;
• The requirement for a borrower to make a $25 loan on Kiva Zip, to experience the community-oriented and reciprocal nature of the program, before they can receive a loan;
• The public crowdfunding phase, in which our 65,000 lenders democratically decide which small businesses will receive our 0% interest loans;
• And most importantly, the Private Fundraising Period, in which the borrower is asked to recruit a number of people from his or her network to lend them at least $25, before they begin publicly fundraising. We have found the number of lenders invited by the borrower to be an incredibly powerful predictor of a loan’s repayment rate (see section 4 below).
For a more comprehensive overview of our social underwriting approach, and why we are so excited about it, see this Next Billion blog post.
The second way in which our approach to risk management aims to be people-focused is in the types of small business owners we seek to serve. As a non-profit, our ultimate aim for Kiva Zip is not to make as much money as possible, but rather to create economic opportunities for entrepreneurs who might otherwise lack them. From African-American entrepreneurs in inner-city Detroit or Ferguson, Missouri; to Native American artisans in rural Minnesota; from returning veterans in Waterloo, Iowa; to ex-offenders in Dallas, Texas; from Vietnamese restaurant owners in San Francisco’s Tenderloin district; to El Salvadorean restaurant owners in its Mission district – we aspire to empower any entrepreneur who has the character and commitment to repay, even if they have no collateral, a damaged credit score, or insufficient cashflows.
But lending to such entrepreneurs is risky! That’s why banks stopped doing it. Kiva Zip’s repayment rate for ex-offenders, or businesses in their first year of operation, or borrowers whose household income is less than $30,000, or ethnic minorities, is lower than the average repayment rate. We could increase our repayment rate significantly by requiring that our borrowers have two years of documented business cashflows, or a FICO score of at least 600. But our reason for existence is to serve these entrepreneurs, who are deemed too risky by every other lender that considers their case.
So our long-term target floor for our repayment rate as a program is not 100%, or even 98%, but 90% (at least currently – as with all of our goals as a team, this target may change over time).
This target is partly based on feedback we have received from lenders in surveys (see the chart below). But it is also deliberately low, based on our own belief as a team that there is a real tension between the depth of social impact of the microloans we are making, and our long-term repayment rate as a program.
We hope that, like us, many of you will be willing to accept a slightly higher level of risk in the service of deeper social impact. But some lenders (like the 51 respondents in the survey above, who told us they require a repayment rate of at least 90%) will always be more sensitive to risk. We would recommend that these lenders either (a) focus their lending on the main Kiva.org platform, where repayment rates are much higher (over 98% over the last 10 years), or (b) select their Kiva Zip loans carefully. One of the advantages of the Kiva Zip model is that there are many mechanisms available for lenders to do more in-depth due diligence on the loans they make. For example, Kiva Zip lenders can ask questions of their borrowers on the Conversations tab of their loan before deciding to lend $25; or they can explore the borrower’s Facebook page or Yelp profile to gauge whether their business has good reviews and a loyal customer base. Many Kiva Zip trustees have a very strong track record of repayment, which can also be a great indicator of risk for lenders – consider La Cocina, Hacienda Community Development Corporation, the Central Arkansas New Agrarian Society, or Northern Virginia Family Services.
2) Kiva Zip’s risk management: The story of the first four years
If mastering risk management in character-based lending is a mountain, then we just left base camp. We still get it wrong all the time – in our policies, in our systems, in our processes. I feel incredibly grateful for the grace and generosity of our lenders in bearing with our small team as we stumble into failures, and try hard to learn valuable lessons from them. Their forbearance reminds me of this quote of Franklin D. Roosevelt, which I read at his monument in Washington DC:
But while we have a long road ahead of us, we are nevertheless proud of the progress we have made as a team over the last four years – both in the improvements to our risk management that we have rolled out, and in the results that we have seen.
The chart below shows Kiva Zip’s cumulative U.S. repayment rate over the first four years of the program:
(A) When we first started, we were only doing a handful of loans in San Francisco, and so it was very easy to meet with every one of our borrowers in person, and extensively assess their business plan and character. As a result, the first 17 loans we made had a perfect repayment rate! But as we expanded the number of loans we were making, and our geographic footprint, it was clearly not possible to maintain this…
(B) In the summer of 2012 we had our first problems with delinquent loans. They were overwhelmingly concentrated in one trustee, which worked with low-income entrepreneurs in St. Louis, Missouri. It is often your most acute failures that teach you the most valuable lessons, and so it proved for our team here. This painful experience taught us of the dangers of risk concentration. Our response was to implement credit ladders for Trustees. Rather than being able to endorse as many loans as they wanted from the outset, Trustees were limited to three endorsements on day one, and could only endorse more borrowers after demonstrating a strong repayment rate on their first loans. The result is an extremely diversified portfolio of loans by Trustee – see the chart below:
I think this chart is a fascinating one, but it needs some explanation! Each one of the 610 green bars on the chart represents a trustee. The width of the bar is the amount of dollars of repayments that have come due on the loans endorsed by that Trustee, and the height of the bar represents the Trustee’s repayment rate across the loans they have endorsed. The bars are sorted by descending repayment rate. In total, $4,036k of repayments have come due, and the average repayment rate across all Trustees is 88.8%. I called out two of the larger Trustees (Institute for Veterans & Military Families in Syracuse, New York; and Centro Community Partners in Oakland, California) as examples; as well as highlighting the widest bar, which represents loans without a Trustee – where the endorsement is from the “Borrower’s Network”. I think there are few interesting things to note about this chart:
• Our portfolio is extremely diverse. Other than loans without a Trustee (which represent 6.5% of repayments), the largest Trustee (Union Kitchen in Washington, DC) represents only 1.6% of repayments. The largest ten Trustees represent only 13% of the portfolio – no longer can one Trustee have a significant negative impact on our overall repayment rate.
• A very large number of Trustees have a repayment rate of 100%. 423 Trustees (over two thirds) currently have a repayment rate of 100%. Collectively, these Trustees represent 47% of cumulative repayments due to-date.
• Only 104 Trustees have a repayment rate less than 80%. These Trustees represent 17% of repayments due to-date.
• 448 Trustees (representing 60% of repayments due to-date) have a repayment rate of over 95%.
The most exciting part about this chart is not what it tells us about the past, but what it can mean for the future. Our Trustee credit ladder system means that if a Trustee has a low repayment rate, we will gradually stop working with that Trustee. If we wind down our relationships with the 104 Trustees who have a repayment rate less than 80%, proceed cautiously with the 58 Trustees whose repayment rate is between 80% and 95%, and grow aggressively with the 448 Trustees whose repayment rate is above 95%, this bodes well for the future direction of our repayment rate.
(C) In the summer of 2013, we continued to see a decline in our repayment rate, and we took actions as a team to turn this around. Led by our Risk Manager at the time (Daniel Jung), we implemented more stringent criteria for deciding which loans to post. For the first time in the history of the program, we began to ask for credit reports from borrowers, which we used to approve or reject loan applications. As you can see from the chart, this approach was successful in addressing our repayment rate problem, but we saw three problems with it: Firstly, our growth stalled, as we started to reject more loan applications based on an applicant’s damaged credit history. Secondly, this process was time-consuming and expensive – both for our small team, and for our borrowers. And thirdly, it was a pragmatism-mandated deviation from our intention to extend credit to small business owners of good character, even if they didn’t have a “thick credit file”.
(D) In January 2014 we found the solution to these three problems. There are few silver bullets in life or lending, but for the Kiva Zip team, the Private Fundraising Period that we launched in January 2014 was one. For over a year we had been encouraging borrowers to invite their friends and family networks to lend to them. Then we started tracking how many people they were inviting. And then we started financially incentivizing it, by matching loans made by “invited lenders”. But in late 2013, we had the following idea: Before we posted a loan to begin publicly funding on the Kiva Zip website, we would require the borrower to invite a certain number of lenders from their own network to lend to them in private. If the borrower reached the required number of lenders (which was initially 15 for loans over $2,500, and 7 for loans of $2,500 and under), they would begin fundraising publicly on the Kiva Zip lend tab. But if they didn’t reach the required number of lenders, they would quietly expire in private. The Private Fundraising Period solved each of the problems that our movement to a more conventional, credit report-based underwriting approach had created: Firstly, we were able to start growing quickly again – borrowers that would have been rejected based on their credit history were now given the chance to prove their entrepreneurial spirit, and that they had a trust network willing to lend to them. Secondly, the time-consuming work of sifting through a borrower’s credit report and digging into any discrepancies was removed. And thirdly, we were able to start saying “yes” again to financially excluded borrowers without a strong credit history. Furthermore, we found that the number of lenders invited by the borrower is a very powerful predictor of repayment rate. The Private Fundraising Period allowed us to manage risk effectively, without slowing down our growth rate or compromising on the depth of our impact.
In the summer of 2014, we were so intrigued by the potential of the Private Fundraising Period as a mechanism for managing risk that we decided to run a small experiment where we made loans without a Trustee. We called this the “No Trustee Pilot”. We have now made hundreds of loans without a Trustee. It does seem (as you would expect) that the repayment rate for loans endorsed by a Trustee is slightly higher than that for loans without a Trustee, but there are some clear advantages to the “No Trustee” model – for example, now small business owners can benefit from Kiva Zip’s 0% interest loans, even if there is currently no Trustee in their community.
(E) At the start of 2015, we iterated on the Private Fundraising Period again. We increased the average number of lenders that borrowers were required to invite in the Private Fundraising Period (see the chart below), but also got more sophisticated in calculating how many lenders borrowers were required to invite. In January 2015, we started to adjust the number of required invited lenders based on four principal factors:
1) Risk. Just as conventional lenders charge a higher interest rate to riskier borrowers, we began to “charge” riskier borrowers a higher number of invited lenders. Kiva Zip borrowers still pay 0% interest and no fees, but because startup businesses (or Transportation businesses) are inherently riskier, such borrowers are now required to invite a few more lenders. By contrast, borrowers with a large number of positive Yelp reviews are required to invite fewer lenders, because we know that a strong Yelp profile significantly lowers their riskiness.
2) Fundraising speed. We aim to serve our lenders as well as our borrowers, and so we want to provide lenders with more of the loans that they want to fund. Farmers are very popular on Kiva Zip, and so their invited lender requirements are lower than the requirements for borrowers in other industries.
3) Social impact. Unlike most conventional lenders, rather than making it harder for low-income borrowers to access a Kiva Zip loan, we lower the Private Fundraising Period requirements for low-income borrowers, even though they are higher risk, because those are the borrowers that we most want to serve.
4) Balance of supply and demand. At times of high site supply, we may increase the number of required invited lenders, to reduce the number of loans fundraising, and increase the amount of dollars coming from lenders invited by our borrowers.
Another important improvement that we made in early 2015 was to hire Suzanna Rush as our Operations Manager. Suzanna brought dedication and diligence to our delinquency management, and implemented a number of improvements in our risk management systems and processes, with immediate results. She has the Herculean task of managing the communication with all of our delinquent borrowers (along with everything else she does!), and she tackles the task with a perfect blend of the efficiency and compassion that we want to epitomize the Kiva Zip program. For a more thorough overview of the delinquency management processes that Suzanna has embedded since joining the team, please see this blog post that she recently wrote.
We will keep striving to make iterative improvements to our underwriting and delinquency management processes and policies. Recently, as another component of our “social underwriting” approach, we started requiring borrowers to make a $25 loan before they are posted to begin fundraising themselves. And next year we plan on building automated monthly repayments so that borrowers do not have to manually send us their repayments every month. As well as saving borrowers time, we hope this will give us a small boost in our repayment rate.
If you have other ideas for how we can improve our risk management, please don’t hesitate to suggest them to us at contactZip@kiva.org!
3) How we analyze Risk on the Kiva Zip team
We strive to be a very data-driven team, and every month we review and analyze our progress as a team – both on the growth side (e.g. how many loans did we fund, how many new lenders signed up, etc.), and on the risk side. We look at four Risk indicators every month, and in the interests of total transparency, here’s how we have done on each of them over the last four years:
(A) Repayment rate
This is the same chart as the one above, and represents our cumulative repayment rate on Kiva Zip. This is very simply calculated as (1) the total dollars of repayments that we have received to-date (excluding “advance” payments where borrowers pay ahead of time), divided by (2) the total dollars of repayments that have come due to-date. At the time of writing, $4,036k of repayments have come due, and $3,582k of repayments have been paid, so our cumulative repayment rate is 88.8%.
As you can see, our cumulative repayment rate has been gradually creeping up since the summer of 2013, when it bottomed out at 84.9%.
Our plan as a team is to throw a big 90s-themed party when we hit 90%! And all of our lenders are invited!
(B) Collection rate
Our collection rate is a more “real-time” indicator of our repayment rate. It is not a cumulative number, but is calculated each month. The denominator of the formula is the amount of repayments that came due in a given month, and the numerator is the amount of repayments that were paid on-time. In October 2015, $224k of repayments came due, and $174k were paid ahead of their due date. This represents a collection rate of 77.5%.
The long-term collection rate trend mirrors the cumulative repayment rate trend. Although it has been creeping very slightly downwards over the last year or so – something we’re watching closely.
(C) Recovery rate
Our recovery rate shows how well we are performing in our management of delinquent loans. The denominator is the amount of repayments that were not paid on time in a given month, and the numerator is the subset of these repayments that were caught up within 30 days of being missed. For example, if $20k of repayments were missed by borrowers in a given month, and $6k of these were collected within 30 days of their due date, our recovery rate that month would be 30%. In October 2015, $50k of repayments were missed, and $20k were paid ahead of their due date. This represents a recovery rate of 40%.
Our recovery rate has been relatively stable over the last couple of years. In late 2014 it was starting to drop, but Suzanna’s arrival on the team got it back on track!
(D) Cohort analysis
While our collection rate and recovery rate give us more of a “real-time” view of our risk than our cumulative repayment rate, they still have significant limitations for helping us to analyze our risk management performance in the recent pass. For example, our October collection rate is mostly driven by the performance of loans that we made over the last two years. It’s difficult to ascertain from our October collection rate how we performed in July or August of this year. That’s where cohort analysis comes in.
Cohort analysis is the best way (that we have come up with so far!) to give us an immediate view of our risk management performance – i.e. “Did we do a good job in underwriting loans last month?”
In the chart above, the green line shows the repayment rate of loans 60 days after they were disbursed, broken down by the month in which they were disbursed. (The light grey bar shows how many loans were disbursed in that month). So, for example, the 70 loans that we disbursed in January 2015 had a repayment rate of 95% after 60 days (the orange bar and diamond on the chart).
Where we see the cohort performance dip in a given month (e.g. May 2015, when it dipped to 92%), we dig into which loans are delinquent, and what we could have done to improve our underwriting.
We also look at the cohort performance of loans after 120 days, and 180 days.
One thing that’s interesting to note in these charts is that (pretty obviously) a cohort that starts off bad is probably going to stay bad, and (even worse) deteriorate over time. For example, the May 2015 cohort of 77 loans had a repayment rate of 92% after 60 days, 92% after 120 days, and 91% after 180 days. If we make mistakes in our underwriting, it weighs us down for years to come!
The chart above illustrates that point. The 39 loans that we disbursed in May 2014 had a perfect repayment rate after 60 days, and they stayed strong over the next year. After 360 days they had a repayment rate of 95%. By contrast, the 39 loans that we disbursed in June 2014 started off badly (a repayment rate of 93% after 60 days), and stayed bad (repayment rate of 88% after 360 days). It is common to see a steady deterioration in repayment rate over the lifetime of a cohort, as some businesses fold or some borrowers are hit by financial hardships. But occasionally cohorts can “catch up” – the 33 loans we disbursed in February 2014 had a 94% repayment rate after 60 days, but a 97% repayment rate after 180 days! (Although it dropped back to 94% again after 360 days).
Given that we are still disbursing less than 100 loans per month, the statistical significance of volatility in our cohort performance is still relatively small – i.e. the “bad” cohort of May 2015 was driven by only six delinquent loans. Did we really do a bad job of underwriting in that month, or was it just bad luck that a few bad loans happened to cluster in that cohort?
But the general trend is a very illuminating one. It seems that our cohort performance wobbled a little bit in the summer of 2015. We’re keeping a close eye on this metric right now, and will take steps to get it back up again if we need to.
4) Some things we’ve learned that we think are interesting!
As you might have guessed by now, we like data and analytics on this team! We’ve now had 1,609 loans have a payment come due, and so we are starting to be able to glean some interesting insights about the repayment rate of different types of borrowers and businesses. If you have any other cuts that you would be interested in seeing, please don’t hesitate to let us know, and we would love to share them with you!
Here are some things we’ve learned from our repayment data so far:
(A) You are good at choosing who to lend to!
We grouped loans into bands according to how fast they fundraised on the Kiva Zip website. Loans that fundraised very quickly (less than 4 hours to fund $100) have a very high repayment rate (94.6%). Loans that fundraise very slowly (more than 25 hours per $100) have a very low repayment rate (84.1%). Averaging across these eight bands, loans that take less than 13 hours to fund $100 have a 93.3% repayment rate, compared to 84.9% for loans that take more than 13 hours to fund $100.
This is one of the main reasons that we think some expirations on Kiva Zip are a good thing! We have seen that our lenders are good judges of repayment rate, and we want to empower them to help us manage our risk. If we had 0% expirations, we would lose this crowd-sourced component of social underwriting.
(B) When borrowers invite lenders from their own networks, good things happen!
The chart above shows the strong correlation between the number of invited lenders on a loan, and the repayment rate. 294 loans without any invited lenders have a repayment rate of 85.9%, compared to 89.9% for 1,315 loans with at least one invited lender – even one makes a difference! 904 loans with at least 10 invited lenders have a repayment rate of 92.0%. 331 loans with at least 25 invited lenders have a repayment rate of 93.8%.
(C) The length of time a borrower has known a Trustee is a small indicator of repayment rate
Another positive data point for social underwriting – the repayment rate for the 651 loans where the Trustee has known the borrower for more than a year is 90.1%, compared to 87.0% for the 707 loans where the Trustee has known the borrower for less than a year.
(D) Beware of Comcast.net!
Borrowers with an @comcast.net email address have a repayment rate of 79.9%, compared to 82.4% for @yahoo.com, and 87.7% for @gmail.com!
(E) Look for a Yelp profile…
Borrowers that include their Yelp profile on their Kiva Zip loan page have a 97.4% repayment rate (based on 219 loans), compared to 87.1% for borrowers without a Yelp profile.
(F) …And a business Facebook page
Borrowers that include their business Facebook page on their Kiva Zip loan page have a 91.6% repayment rate (based on 1,090 loans), compared to 83.0% for borrowers without a business Facebook page (519 loans).
So if you are a lender and you are sensitive about risk, look for loans where the borrower includes a link to their Yelp profile and/or business Facebook page!
(G) Florists and Farmers!
The table above shows our repayment rates by industry. One of the reasons we are so focused on farmers on the Kiva Zip team is that they have a very high repayment rate – 96.2% based on 203 loans.
(H) Startups are a risky business
Unsurprisingly, time in business is a significant predictor of repayment rate. Risk-sensitive lenders should avoid pre-revenue startups, and lend to more established businesses.
(I) Lower income borrowers are higher risk
As noted above, we see a direct trade-off between risk management and social impact. Borrowers with a household income less than $40,000 have a repayment rate of 86.5%, compared to 91.0% for borrowers whose household income is more than $40,000.
(J) Borrowers that voluntarily make loans tend to repay
320 borrowers have voluntarily made Kiva Zip loans over the last four years. These borrowers have a repayment rate of 93.4%, compared to 87.6% for the 1,289 borrowers that have not made a loan.
This finding was the main reason for us recently instituting the policy by which we now require borrowers to make a $25 loan before they get a loan.
(K) Breakdown by state
The repayment rate in California, where we have made the most loans, is 93% – significantly above the national average.
(L) Top trustees!
Lastly, a shout out to these 10 Trustees, each of whom have had more than $20,000 of repayments come due on their loans, and maintained a 100% repayment rate.
We hope you found this overview of Kiva Zip’s risk management philosophy, evolution, analytics and insights interesting. One of our firm values at Kiva is Transparency, and this blog post is hopefully an example of how we aspire to that value. But another reason for writing this post is that, while we are encouraged by our recent growth, as a non-profit we will probably always remain small compared to the trillions of dollars of loans that are moved every year by the world’s big banks. But perhaps the way in which we might have the most impact on the world’s financial system is to share insights like these, in the hope that bigger, for-profit lending institutions might take some of the lessons that we are learning onboard.
We believe that the social underwriting model we have developed, and are iterating on, is very powerful, and can be very “financially including” – at a time when millions of Americans (and billions of low-income people around the world) are unbanked or underbanked. From our data, it seems that if banks could embrace some of the character-based lending principles of social underwriting, they could reach more customers, more cheaply, while maintaining a robust repayment rate.