Lending Club – $2.4BN improvement

By analysing the defunct loan book of US-based Lending Club, provided to us through the FCA Digital Sandbox, Finexos was able to obtain some significant results using our ML-driven credit risk algorithm.

This data set, while only enabling us to utilise 4 of the algorithm’s 240+ input factors, was nevertheless valuable in demonstrating the efficacy of the platform because of its size, completeness, and fit with our target segments.

On this basis, we plugged the results into our risk engine and looked at the metrics that we could have provided to Lending club, enabling them to make far more informed origination decisions, over the multi-year period.

Understanding these results in detail:

Defaults Identified

From the 890,000 loans made by the lender, 28% of the total lending capital went towards bad loans. The algorithm would have identified 58% of those bad loans.

Good loans identified

Improved identification of bad loans does not affect the performance of the overall book if, at the same time, you disqualify loans that would have otherwise been successfully repaid; it is just as important not to disqualify good loans as it is to correctly identify bad loans. From the 890,000 loans made by the lender, 78% of the total lending capital went towards loans that were successfully repaid. The algorithm would have correctly identified 96% of these loans as a good credit risk. In other words, the algorithm would only have said no to 4% of the good loans that should have been originated by the lender.

Losses Avoided & Capital Saved

The total amount of loan losses that would have been saved by using Finexos’ analysis to inform lending decisions would be 17% of the total lending capital, or an aggregate saving of USD $2.4B.