Skip to content

Latest commit

 

History

History
executable file
·
29 lines (20 loc) · 2.19 KB

Results_Summary.md

File metadata and controls

executable file
·
29 lines (20 loc) · 2.19 KB
Raw

Results Summary

Original Loan Portfolio

Note: despite good loans outnumbering bad loans by roughly 4:1, the losses from the bad loans wiped out 61.3% of the gains from the good loans

  • Total portfolio size 5.9B
  • Gains from good loans ~$912m
  • Losses from bad loans: ~$560m
  • Net Gains: ~$353M
  • Default Rate: 19.2%
  • Annualized return 1.92%

High Level Results - Machine Learning Portfolio

Note: the results are based on extrapolating the results of the test dataset to the entire loan portfolio.

  • Total portfolio size: $3.4M
  • Default rate drops to 11.2%
  • Annualized returns increase from 1.92% to 2.65%
  • Net gains: $275M vs $3.4B in Capital, a reduction in capital outlay of $2.5B
  • $78M fewer gains, but given the significant reduction in capital, capital efficiency improves substantially: the annualized return on the extra $2.5B investment was ~1.1%.

On a capital efficiency basis and default rate basis, the model out performed both potential interventions identified during EDA

  • Only originating loans to rated A, B, C & D per Lending Club's internal grading system. Doing so would only reduce earnings by ~1M, but would reduce capital outlay by ~900M. Annualized returns would grow by 15%, default rates would fall to around 16.2%. While one could make the argument that this approach is better than investing in the loans selected by the model, the higher default rate + reduced capital efficiency introduces a lot of risk for little benefit.

  • Setting an interest rate cap of 13.5%, i.e., only originating loans to customers when the current models set interest rates of 13.5% or lower. This approach would reduce default rates to 12.3%, increase annualized return by 25%, but reduce total returns by 87M, while also requiring about 2.4B less in capital.

The value of this model to a lender would be a function of how they are looking to optimize their portfolio. Namely: if an outlay of an additional 2.5B for annualized gains of less than 1% is worth it to gain an extra ~$80M, then neither the models nor the suggested interventions are valuable. However, if capital efficiency and/or maximizing returns is the lender's focus then this model could prove quite useful to them.