Changes to admittance for new borrowers

Discussion in 'Announcements' started by Julia Kurnia, Nov 4, 2014.

  1. Julia Kurnia

    Julia Kurnia Director, Zidisha Zidisha Staff

    Starting this month, we will again begin admitting a limited number of new borrowers to Zidisha. Places for new borrowers will now be allocated on the basis of predicted credit risk, as estimated by a data science algorithm supplemented by manual application reviews.

    For active borrowers, this means that as long as you repay on time and post quality applications, you can expect to continue to be able to raise loans in gradually increasing amounts.

    For lenders, the intent is that fewer of your bids will expire, and repayment performance of new loans will continue to improve.

    For new applicants, there is no change to fees or loan sizes, but admittance is more selective now, and fulfilling the criteria and submitting an application does not guarantee acceptance. We will do our best to make the application form simple, and only ask for the information we need to make an accurate admittance decision. We will also make a timely up or down decision (usually within a couple days) so that you can move forward with your investment plans.

    Why the change?

    Up until recently, we allowed every eligible applicant who passed a background check to join Zidisha and post loan applications for funding. This approach made sense in our early years, because the volume of fundraising applications roughly matched the supply of lender capital. Almost all applications were funded, and lenders' bids rarely expired.

    Last year, we transitioned to an entirely online application process, and began using web-based fraud protection technologies in place of manual documentation checks to screen applications. This change improved repayment rates, while making it easier for borrowers to apply.

    Now many of the borrowers from that time have repaid their initial loans and qualified for larger ones. At the same time, growth in new member applications is continuing to accelerate. Lender funding is growing steadily, but it is likely that borrower demand will continue to outpace it for the forseeable future. If the past is any indication, any growth in lending capital will cause even greater growth in demand from borrowers.

    Though lender bidding has more than doubled this year, the volume of successfully funded loans has grown much more slowly. The reason is that a large percentage of bids are now expiring unfunded. These graphs, produced by a data scientist who has been analyzing our lending data, illustrate these trends:
    bidFate (3).png


    The data also indicate that a high expiration rate decreases the total funding available to our borrowers, because many expired bids are not re-lent soon or at all:

    impact on funding.png

    These findings suggest that reducing the volume of fundraising applications would not only improve individual borrowers' chances of being funded, but would actually increase the total volume of loans successfully funded at Zidisha.

    A more difficult decision is how to limit the volume of fundraising applications. Some options we considered were:
    • Adopting heavier application requirements, perhaps by re-introducing scanned documents and required endorsements. This is closer to how traditional microfinance programs screen applicants, and has sometimes been suggested by our members. It sounds like a good idea. However, our experience with this approach in the past was that that it did not provide optimal fraud protection, encouraged applicants to have others register on their behalf, and was very difficult for an organization without local offices to police at scale.
    • Raising prices for borrowers - either by increasing Zidisha fees, or by offering more interest to lenders. If we were a for-profit business, this would be the obvious solution. As a nonprofit, our concern is that it would undermine our mission of making finance affordable and ensuring the entrepreneurs retain enough profit to benefit from the loans. A stronger argument could be made for offering more interest to lenders if this could be shown to make it easier for borrowers to raise loans, but when we have experimented with higher rates in the past, it seems to have had the opposite effect: more lenders were turned off by the high rates charged to borrowers than were attracted by the increased financial returns.
    • Decreasing loan sizes. We have already done this to some extent, and now seem to be near an optimal balance, where most loan sizes are just large enough to make a substantial impact in people's lives and incentivize continued responsible repayment.
    • Investing more of our resources in marketing in an effort to increase lending capital availability. At this point, we believe the best long-term investment of our limited staff time and funding is in further improving our platform for our current members, rather than marketing. Also, increasing lending capital alone is unlikely to be a sufficient solution, because any increase in loan funding causes an even greater growth in borrower demand.
    We were fortunate to have a successful fundraising campaign at the end of our participation in Y Combinator this year. We have been using part of the funding to improve our website. Another investment we made was to have a team of data scientists examine our loan and repayment data over the past five years, and identify patterns that correlate with successful repayment. Over the past four months, the data scientists have developed a credit risk prediction algorithm similar to those used by modern financial institutions (though rarely by microfinance organizations).

    While it is impossible to predict with full certainty whether a new applicant will repay loans, tests suggest that using the credit risk algorithm in admittance decisions will reduce non-payment substantially. We will track new loan repayment rates carefully as we begin to use it, so that we can make any necessary adjustments to ensure continued performance improvements.

    The goal is a respectful and impartial admittance process for new applicants, which keeps funding rates high for borrowers and repayment rates high for lenders.
    Last edited: Nov 4, 2014
  2. JimVandegriff

    JimVandegriff Gold Member

    Julia, can you share some of the components of the credit risk algorithm that relate to credit repayment? Thanks for your work. Jim
  3. Julia Kurnia

    Julia Kurnia Director, Zidisha Zidisha Staff

    Hi Jim,

    Thanks for your comment. I'd like to share more about how the algorithm works, but the specific factors it uses need to stay confidential to prevent anyone with dishonest intentions from gaming the application process. I can say that it does not take into account gender, age, family status or any other factors that should not be used as a basis for differentiation among applicants.

    If you'd like to learn more about the data science project itself, here is a really informative article published by one of the Bayes Impact team:
    Oscar Alochi likes this.
  4. JimVandegriff

    JimVandegriff Gold Member

    Thanks, Julia. Quite an interesting article. I appreciate the post, and it helps me understand the process to a greater degree. Jim
  5. Nanette

    Nanette Forum Member

    Julia, this is such an exciting, smart new approach! I read the paper/blog post over at Bayes Impact, I admire the hard work that is put into this approach by yourself and the Bayes Impact team.

    Oscar Alochi likes this.
  6. julidaze

    julidaze Silver Member

    thanx for the updates

Share This Page