A portfolio credit-risk monitoring suite on real loans: vintage performance, delinquency, model drift, concentration, and a RAG early-warning dashboard, assembled into a committee-ready management-information pack.
This project is the oversight and reporting side of credit risk. It takes a live loan book and produces the monitoring a risk function runs and the management information it presents to its committees: how each origination vintage is performing, where the book sits on delinquency and impairment, whether the scorecard is still stable, how concentrated the exposure is, and a single early-warning dashboard that rolls it up with red, amber and green flags. The committee pack is in MI_PACK.md.
The book performs on a cash basis (low delinquency, well diversified) but two forward-looking signals warrant attention: origination quality is deteriorating and the scorecard's population is drifting.
| Indicator | Reading | Status |
|---|---|---|
| Vintage 12-month default trend | 6.8% latest mature vintage, up from 4.7% | RED |
| Live-book delinquency rate | 3.1% of active EAD | GREEN |
| Impaired (30+ days past due) | 2.6% of active EAD | GREEN |
| Scorecard population drift (PSI) | 0.13 vs baseline | AMBER |
| Geographic concentration (HHI) | 0.051, top-5 states 41% | GREEN |
| Product concentration | 58% of EAD in one purpose | AMBER |
Portfolio: 907,904 active loans, $9.51bn exposure at default.
flowchart LR
A["Live loan book<br/>908k loans, $9.51bn"] --> B["Vintage cohort performance"]
A --> C["Delinquency and impairment"]
A --> D["Scorecard drift (PSI)"]
A --> E["Concentration (HHI)"]
B --> F["Early-warning dashboard<br/>RAG"]
C --> F
D --> F
E --> F
F --> G["MI pack<br/>for the risk committee"]
- Vintage cohort performance. Cumulative default by months on book for each origination cohort, the standard way to see whether newer lending is performing worse than older lending.
- Delinquency and impairment. The live book by arrears bucket, and the share of exposure that is delinquent or impaired.
- Scorecard drift. Population Stability Index of the model's score distribution over time, against thresholds, so model decay is caught early.
- Concentration. Herfindahl-Hirschman indices and top-exposure shares by geography and product.
- Early-warning dashboard. A RAG roll-up of the above, the page a committee reads first.
Vintage cohort curves. 
The 2015 and 2016 cohorts sit above the 2012 to 2014 cohorts at every point in their life. Newer lending is defaulting faster.
12-month default by vintage. 
The early-default rate has risen from 4.7% (2013) to 6.8% (2016), about a third higher than the early-vintage average. The latest bar is greyed because that vintage is only partly observed.
Delinquency profile. 
Exposure by arrears bucket. The book is overwhelmingly current; delinquency is 3.1% of active exposure.
Scorecard drift. 
Population Stability Index by vintage against the development baseline. It has climbed into the moderate-shift band, a signal to review the model.
Concentration. 
Top-10 states by exposure with the cumulative share. With an HHI of 0.051 the book is well-diversified geographically.
Early-warning dashboard. 
The RAG roll-up: two reds and ambers to watch, the rest green. This is the committee's first page.
Risk-band default, early vs recent. 
Default rate by model risk decile for early versus recent vintages. Recent vintages default more at the same modelled risk, the calibration drift behind the dashboard's amber flags.
This is the monitoring and management-information half of a credit-risk function: vintage and delinquency monitoring, model surveillance, concentration and large-exposure oversight, and committee reporting with clear escalation. The deliverable is written as an MI pack with a RAG dashboard and recommended actions, which is the form this reporting takes in practice. For an auto or captive-finance book the same pack would add daily single-counterparty large-exposure monitoring on the dealer side, a direct extension of the concentration analysis here.
pip install -r requirements.txt
python analysis.py # builds all monitoring metrics, figures, and results.json.
├── analysis.py # vintage / delinquency / drift / concentration / EWI pipeline
├── credit_risk_monitoring.ipynb # the same pipeline as an executed notebook
├── MI_PACK.md # the committee-ready management-information pack (read this)
├── results.json # all metrics
├── figures/ # seven charts
├── data/
│ └── exposure_by_state.csv # exposure and share by state
├── requirements.txt
├── LICENSE
└── README.md
The underlying Lending Club loan data is publicly available and used here for a non-commercial portfolio project. The code is released under the MIT License (see LICENSE).