Credit Risk Lab — LendingClub Loans

Credit-risk report on 2.26 million LendingClub consumer loans (2007–2018): what was lent, what came back and what was lost, with a probability-of-default model and expected loss on the live book.

Preview

Highlights

The naive metric says grades D–G lost money; measuring matured loans only flips that conclusion
Built and validated as code; not yet opened in Power BI Desktop

Problem

A lender needed to know whether higher interest rates on riskier borrowers paid for their extra defaults, which borrowers fail most, and how much loss the loans still open are likely to bring.

Approach

DuckDB star schema over the 1.2 GB loan tape with 16 data tests, including a cash bridge that adds up to net profit
Default, loss and return measured on matured loans only, because young loans that default resolve early and make recent years look worse than they are
Gradient-boosted default model using only what was known at origination, tested on a later year it never saw
Six report pages with synced Year / Tier / Term slicers, KPI cards, default-rate heatmaps, a tile-grid US map and measure-driven colours

Stack

Power BIDAXTMDLPBIRDuckDBscikit-learnPython

Outcome

$34.0bn lent; on matured loans 14.7% defaulted and lenders still kept 9.6% of principal as profit
Every grade made money: the riskiest grades earned about 15% lifetime return vs 7.4% for the safest, while recoveries after default stay near 9% for all grades
Default model: AUC 0.691 out-of-time; its riskiest 10% of loans hold 21% of all defaults
Model: 13 tables, 59 measures; 6 report pages; both validators pass