Online Lenders are doing a good job of identifying the frauds (even without hard income verification)

Default Rate for Verified vs Non Verified Income - Lending Club

Recently there was an article in Bloomberg (Article Link) that talked about how online lenders are not always verifying the basic borrower information like Income. Subsequently it got picked up by websites like Yahoo & Business Insider (Article Link) generating some sensational headlines.

Though both articles did mention that sites like Lending Club use various indicatiors and some Machine Learning models to determine whether they need a manual verification of the income, but that detail was hidden. We at Croudify have been analyzing the loan data for more than 2 years and wanted to shed some light on the article and show that while the headline is true the devil is in the details and actually the platforms have been doing a very good job in identifying the fraud.

% of Loans with Non verified Incomes

Our First step in analysis consisted of identifying how many loans for which Income has not been verified by Platforms. For our analysis we chose Lending Club data. If you see the graph below you will see that in early years (2008) nearly 85% if the loans were originated without Income Verification. This can be attributed to lack of resources and originating business practices. Off late (since 2011) nearly 1/3rd (33%) of loans had that status. So the trend is not particularly new from a historical  perspective. Actually if you look historically as platforms have grown they are verifying more incomes and not less as the headline would have you believe.

 

Performance of Non Verified Loans

Once we had concluded that the non-verified loans are not growing as percent of population the logical next step was to check if these loans are performing worse than before. Is there a possibility that the loans without income verification have deteriorated over time and hence the red flag.

Here the analysis showed us a completely opposite result. The loans for which Lending Club had decided not to verify the income has performed better historically and continue to outperform the overall population. This points to a very important finding that the preliminary indicators that Lending Club is using in identifying the fraudulent behavior is not only working it is working great and is providing a performance lift to loans.

This is a very important discovery and points that what was true in old world with traditional lending is no longer true with the new world with Online lending and advance Machine Learning models and we as investors need to change our thinking accordingly.

8 thoughts on “Online Lenders are doing a good job of identifying the frauds (even without hard income verification)

  1. Thanks for pulling this together Abhishek.
    Isn’t this simply saying that income verification is done for higher risk applicants and they default more? Not sure you can draw too many other conclusions or are you looking at matched samples?

    1. Thanks Charlie for the comment. So first I wanted to rebut the article that made it appear like the increasing defaults are due to lack of income verification. Second, we have done comparison across terms and ratings and for most of the buckets since 2014 the statement holds. For earlier vintages in 2010 and 2011 it was not true for some buckets but LC models should have evolved since then.

    1. Brian, the data on Lending Club is similar and not contradictory. The first graph in the post corresponds to the Graph “Portions of Loans issues that are not verified”. It shows verified income and source verified, so not verified is is 1 – sum of both verified. For 2016 the sum of both verified is 69.19% so unverified is that comes 30.81% same as the one represented in the first graph.

      Similarly of you look at charge-offs by vintage for 2016 the numbers are around 2-3% for Verified and nearly 1% for non verified, this gives a combined default rate of around 2% and 1% for non verified as represented in the second graph.

Leave a Reply

Your email address will not be published. Required fields are marked *