jkurnia

We actually explored that before we met Bayes Impact. The Kaggle team thought a competition would be overkill for our data set (about 8000 loans).

That's an interesting question. Thus far our market selection has been driven more by practical constraints than sophisticated analysis. We've focused on countries that have well-developed electronic payment systems (such as M-PESA in Kenya). We also try to target places where average incomes are low enough that loan amounts that seem small in wealthy countries can have life-changing impact.

I'm the founder of Zidisha, and I can't thank the Bayes Impact team enough for this project. Like most small nonprofits, we don't have the resources to hire in-house data scientists. Bayes Impact brought data science expertise within our reach for the first time, and they've had a transformative impact on Zidisha.

A good argument can be made either way, but I doubt we'd ever use a proofreader as so much of our identity is in keeping the lender-borrower communication direct and authentic.

That said, we do allow our users to post translations of non-English content, while leaving the untranslated versions accessible - you may view an example here: https://www.zidisha.org/microfinance/loan/SALIMATA2/9512.htm.... We also sometimes edit the titles of loan projects for clarity.

I expect the default rate for loans being issued today will be less than 10%, because some of the amounts still outstanding will likely be repaid late. 90% is the portion of due amounts that has already been repaid (for 2014 loans).

That said, we are continuously improving our lending model in response to experience, and the repayment rate is increasing with each new cohort of loans we disburse.

Bayes Impact (another YC nonprofit which connects data scientists to nonprofits) is developing a credit risk prediction algorithm for use in vetting Zidisha loan applicants. Once that is deployed, we expect it will improve repayment rates substantially.

Companies like Kabbage (in the US) and Lenddo (Philippines, Colombia) use social graph data pretty extensively to predict credit risk. At Zidisha thus far, we're only using Facebook data to verify online identities of the applicants.

We actually don't set any interest rate. Zidisha charges a flat 5% fee per year the loan is held (for example a $100 loan held for three months would cost $1.25). In addition to that, we allow each lender to choose the interest to receive, if any - up to a maximum cap of 15%. If the loan is oversubscribed, the portions with the lowest interest are retained. The average rate chosen by lenders is around 5.5% per year, but there's a lot of variation, as borrowers with the longest positive repayment histories tend to attract lower interest rates.

Yes, this is true direct lending. In fact it's more direct than Kiva Zip, because the latter uses local organizations called "trustees" to help vet borrowers, and the trustees frequently post content in place of the borrower, especially outside the US. Zidisha does not use trustees or any other intermediary, and the content in the loan profile pages is written directly by the borrowers. You can view a sampling of the sort of comments our borrowers post here: https://www.zidisha.org/microfinance/testimonials.html.

Our best recourse against default is prevention, as it is very hard to collect a loan once it has become delinquent. Most of our credit risk protection happens at the application stage: things like a referral program that incentivizes borrowers to invite only trustworthy acquaintances to join Zidisha, machine learning tools to detect fraudulent accounts, and a credit limit progression system that incentivizes members to maintain high on-time repayment rates.

We spend a lot of time on fraud protection, but the most frequent cause of default at Zidisha is simple financial hardship. We're lending to a demographic whose income is not only low but also erratic, and lacks substantial savings, insurance or government safety nets. A big part of our credit risk protection is recognizing that reality and accommodating it through mechanisms like allowing borrowers to choose their own repayment schedules and adjust their installment amounts up or down to adapt to their financial situation.

The 90% figure is based on the amount of money repaid. More precisely, it's the amount that has been repaid for loans disbursed in 2014, divided by the amount that is due to have been repaid.

I'm Julia, the founder of Zidisha. I'd be happy to respond to questions and comments.

We can receive SMS replies at the virtual Twilio number we use for automated notifications. We have an amazing virtual team of volunteers and interns who reply to those messages in various languages - we actually end up doing more support for borrowers via SMS than email.

Most Zidisha borrowers own feature phones, but not computers. When lenders post comments on their profiles, we send the comments as an SMS text to the borrowers' phones. The borrowers can then opt to visit a cybercafe to post a response.

I'm Julia, founder of Zidisha. I'd be happy to respond to questions and comments.

I think domestic P2P platforms are a big untapped opportunity in developing countries, especially large ones. They are still rare; Fairplace in Brazil is a notable exception.

An international platform like Zidisha can offer a different value proposition, in that 1) we enable arbitrage in purchasing power parity between wealthy and developing countries (the price of dinner at an upscale restaurant in America could finance the startup of an entire restaurant in Kenya), and 2) our users value the chance to connect with an ordinary person on the other side of the world, learning about their culture, socioeconomic realities, etc.

Most loans are held for less than one year, so the fee ends up being less than 5% of the amount raised.

This is intended as a service fee, not interest. It is set at a level that, along with registration fees and lender tips, just covers our operating costs. (For example, in the first quarter of 2014 our average operating costs were $8,225 per month. Our average monthly revenues were $8,507.)

It might be more precise to structure the service fee as a fixed percentage of the loan amount regardless of the term. We opted to structure it as an annual rate mainly for simplicity: it can be added to lender interest so that loan applicants can see the full cost of the loan in a single rate, without any other unexpected fees.

We are intended as a platform for philanthropy. Any interest returns wouldn't be enough to compensate for risk.

a) what collateral do you accept and how do you value the collateral?

We don't accept any collateral, because 1) most of our borrowers are low-income individuals just entering the working population, and don't have any substantial assets, and 2) a collateral-based credit risk mitigation system would be too expensive relative to the size of the loans (the starting loan amount is $50 - $150).

b) do you consider country risk ? and how do you quantify individual's risk profile ? have you developed a credit risk model ?

Currently, once a borrower is admitted to Zidisha, lenders alone judge country and credit risk. However, we're currently working on a data-driven credit risk model to complement our other risk mitigation measures.

c) would you consider offloading credit risk by partially securitizing your loans?

Not in the near term, because our lending model is too new and changes too rapidly for credit risk to be easily quantifiable. In addition, we'd be concerned about moral hazard.

d) Countries where you operate in have a serious risk of document fraud. can you share stories of fraud mitigation ?

One of the major shifts in our lending approach has been a move away from document-based vetting. For example, we used to require local community leaders (religious officials, school principals etc.) to sign recommendation forms for loan applicants. We found that fraudulent applicants would forge these, even having false organizational stamps manufactured for this purpose. It was possible to mitigate this by telephoning the officials who signed the forms and asking them to verify their institutions' addresses or other hard-to-remember information, but it proved too difficult for our all-volunteer staff to cope with the erratic phone networks and linguistic barriers once we started processing thousands of applications.

We never did figure out a way to mitigate document fraud at scale. Instead, we developed other verification methods that were more difficult to falsify, and reduced the initial loan amount, until the costs of getting a fraudulent application through our vetting system outweighed the expected gains.

f) finally, what kind of lenders do you seek ? Are you looking for altruistic people or people seeking to spread their investment risk ?

We are looking for altruistic people, who want to help people in the world's poorest places without dependency-creating handouts or exorbitant interest rates. Zidisha at present is not viable as a purely commercial investment.

e) Also, do you realize that you have the opportunity to become the Equifax of third-world countries ?

Yes, that's an interesting possibility, though I've not yet heard of Zidisha being used in this way. Most people who can borrow with Zidisha do not seek credit elsewhere.

We did consider adding bitcoin support, but opted to refrain for now because:

- we did not identify a straightforward way to convert bitcoins to cash in borrower currencies

- it seemed from our informal research that supporting bitcoin has not resulted in many new supporters for most nonprofits that have tried it

- supporting another payment method would add overhead weight to our organization

- so few of our lenders use bitcoin that we estimated the savings in transfer fees wouldn't be worth introducing more complication in our payment options

We're keeping an eye on bitcoin and may begin supporting it if these conditions change.

Traditional microlending operations are expensive. For example, the average microfinance organization needs to charge borrowers 35% interest for capital borrowed at zero interest from Kiva.org, to cover its overhead costs. (Source: https://en.wikipedia.org/wiki/Kiva_(organization)#Current_in...)

Zidisha's overhead costs average about 10% of the value of the loans we facilitate. The costs are fully covered by borrower fees and optional lender tips. Borrower fees include a lifetime membership fee of $12 paid when the first loan is disbursed, and a flat fee of 5% of each loan funded per year the loan is held.

We don't at the moment, because our programming resources are limited and we have been prioritizing development of our lending platform. We will likely be able to make our data available in an API within a year.

You can view the data used to generate the repayment graph here: https://www.zidisha.org/index.php?p=114

When we first open a lending program in a new country, a local volunteer or visiting international intern typically finds an initial group of borrowers by making presentations at NGOs, pitching Zidisha to local cybercafe owners, etc. After that, growth happens almost exclusively by word of mouth.

Most in-country volunteers (we call them "Volunteer Mentors") are also Zidisha borrowers. They learn about the mentoring program via a note in the borrower account interface of our website, and apply via email.

I'm Julia, the founder of Zidisha. I'd be happy to answer questions here.

"The microlending websites you mentioned have much higher repayment rates, so that there's a much smaller amount of portfolio actually at risk."

There is a big difference between the repayment rates for ended, and portfolio at risk for currently outstanding loans. In order to have a complete picture, lenders should know the performance of currently outstanding loans as well as ended ones. Zidisha is the only microlending website I know of that provides statistics on the performance of the currently outstanding loan portfolio.

Even our statistics for ended loans are not defined in the same way as other websites. For example, I understand that Kiva expresses its repayment rate as 100% minus the writeoff rate, including as repaid loans amounts not yet repaid, but not yet written off. (http://www.kivafriends.org/index.php?topic=5791.0)

More importantly, I understand that Kiva allows field partners to cover repayments not actually made by the borrowers and report these amounts as paid on the Kiva website. (http://blog.givewell.org/2009/10/13/kiva-repayment-data/) These amounts are repaid in that lenders do receive the funds back, and I am not accusing Kiva of dishonesty. But the fact remains that substantially less than 98.95% of Kiva loans are repaid by the borrowers themselves.

I have lived with many low-income households in developing countries, and in my experience, cash availability of households without formal jobs, insurance, savings or social safety nets is not only limited, but highly irregular. It is simply not realistic to provide people under these circumstances enough capital to change their lives, and expect its repayment to obey a predetermined schedule with 98.9 percent accuracy. That is why I am skeptical of such high published repayment rates.

Zidisha's published rates are lower, but they do accurately reflect the repayment performance of the borrowers themselves. Comparing them with the statistics of intermediated microlending websites is misleading, because they are not measuring the same thing.

I have not managed to find information on how Kiva Zip's repayment rate is defined, so cannot comment on it.

Thank you for responding here. I tried to post a response in your blog, but my comment was removed or not yet published.

Thank you for letting me know. We'll aim to repair that as soon as possible.

Thanks for this advice. I've edited my original comment to focus on the content of the article.

I agree that the way we present information about rescheduled loans can be improved. We do not consider them to be in arrears, because at Zidisha borrowers have the flexibility to reschedule loans to repay either later or earlier than planned, and rescheduling does not necessarily indicate greater likelihood of default.

When web development resources allow, we aim to improve our Statistics page to use a single metric to encompass both ended and current loans, and allow users to filter the data by date disbursed.

I would agree with these statements: Zidisha is a platform for philanthropy.

We're optimized for giving disadvantaged entrepreneurs in the world's poorest places the chance to improve their lives, not for storing financial assets. I think we make that distinction abundantly clear to prospective lenders. See for example our first FAQ for lenders, "Is Zidisha a safe place to store my savings?" (https://www.zidisha.org/microfinance/faq.html), or the first paragraph of our Trust and Security page (https://www.zidisha.org/index.php?p=123).

That said, repayment rates at Zidisha are increasing steadily as we continuously improve our lending model in response to experience. The 18% writeoff rate includes mostly loans that were issued over a year ago (which have had time to mature).

Our credit risk control mechanisms have improved dramatically since that time, such that I think it's reasonable to expect that the value of lending funds can be maintained over time without erosion for the loans that are being issued today. That said, Zidisha is still not a safe place to store funds one cannot afford to lose, and we do not market it as such.

Dear Modern Microcredit,

I am surprised that you find Zidisha to be insufficiently transparent. Zidisha has room for improvement in many respects, but we've always considered transparency to be a strong point of our platform.

We have more detailed statistics on repayment rates, trends in repayment performance over time, and risk levels in currently outstanding loans than any other microlending website I know. For example, Kiva, Kiva Zip, United Prosperity, and other well-known microlending websites do not even publish Portfolio At Risk statistics.

Since Zidisha is a direct P2P model, our repayment rate is not inflated by "phantom payments" reported by local intermediary partners. (http://blog.givewell.org/2009/10/13/kiva-repayment-data/) When we report a repayment as having been made, that means the borrower truly did make that payment.

We make it very clear to prospective lenders that lending through Zidisha is risky. Here are some excerpts from our website:

"Zidisha is a low-cost nonprofit community managed by volunteers. Our purpose is to connect individuals in marginalized countries with the chance to improve their lives through affordable loans. We are not a financial institution or an investment service, and Zidisha is not a safe place to store your savings or financial assets. We recommend you treat lending through Zidisha as a philanthropic activity, and that you not lend any funds that you cannot afford to lose." (first paragraph, Trust and Security, https://www.zidisha.org/index.php?p=123)

"Is Zidisha a safe place to store my savings? No. Zidisha is a platform for philanthropy, not a replacement for traditional savings accounts or financial investments. Zidisha loans are highly risky, and direct person-to-person lending across international borders is such a new endeavor that expected losses cannot be predicted with sufficient accuracy to make Zidisha loans a safe way to store funds you cannot afford to lose. We recommend that you treat Zidisha loans simply as a way to help others, and that you lend no more than you would normally be prepared to give away to charity." (Lender FAQ number one, https://www.zidisha.org/microfinance/faq.html)

"The principal purpose of Zidisha's lenders in funding loans is to improve the lives of these entrepreneurs, and not to make a profitable financial investment. Lending to entrepreneurs through Zidisha involves a meaningful risk of loss of your principal lent. In addition, you are very likely to lose money in real terms on any loan you fund, even if it is repaid by the borrower, as a result of currency exchange rate changes, PayPal fees applied to uploading credit to your lender account, inflation and other factors affecting the value of principal and interest payments made by a borrower. Accordingly, we recommend that you consider lending through Zidisha to be a philanthropic activity, and you should not lend more money through Zidisha than you are prepared to lose." (Terms of Use, https://www.zidisha.org/index.php?p=5)

More generally, our website code is open source (https://github.com/Zidisha/zidisha), and conversations among our volunteer staff take place in public in Zidisha's forum. (https://www.zidisha.org/forum/categories/volunteer-conversat...). Our operational procedure manual is published on the internet for anyone to see. (https://sites.google.com/a/zidisha.org/zidisha-staff/)

If anything, we are transparent to an extreme.

I do agree that Zidisha loans are not viable as a purely financial investment. Financial investment is not the purpose of our platform. And that is exactly what we are communicating to prospective lenders.

Best regards,

Julia Kurnia Director, Zidisha

We do offer access to the data that makes up the graph depicting trends in on-time repayment rates at Zidisha: https://www.zidisha.org/index.php?p=114

It generates a simple table with the repayment performance of each of the 6000+ loans that have been funded through our platform, with links to the individual loan profile pages.

We'd definitely like to offer an API similar to Kiva's, when web development resources allow.

It is true that 74% of loans whose final repayment due date has passed have been repaid. We expect the ultimate repayment rate of those loans will be higher than that, because many loans are repaid after their due dates.

In addition, most of the loans covered by the historical repayment rate displayed in our website were issued over a year ago. Our repayment performance has improved substantially in the past year, such that risk level of loans currently being issued is lower than our historical repayment rate suggests.

Zidisha is still not a safe place to store savings or a lucrative investment opportunity, but I think it's reasonable to expect that the value of lending funds can be preserved over the long term, while offering financing opportunities at dramatically lower cost to the borrowers. As a nonprofit philanthropic platform, that is Zidisha's purpose.