Seven dimensions, all visible
A composite score is available, but the constituents stay in view. Each dimension is separately measured, separately explained and separately challengeable:
- Distress probability
- Probability of insolvency or another defined adverse event within 12 months
- Financial resilience
- Liquidity, leverage, profitability, asset coverage and deterioration
- Filing behaviour
- Lateness, accounting-period changes, missing or inconsistent filings
- Governance stability
- Director, PSC, address and ownership changes
- Network risk
- Adverse outcomes among connected entities and directors
- Security position
- Charges, charge concentration and recent secured borrowing
- Data confidence
- Freshness, completeness and reliability of the available evidence
What public data cannot see
Public records support a useful distress model. They do not reveal what only lenders know:
- 30-, 60- and 90-day arrears
- Payment performance
- Covenant breaches
- Recoveries and lender losses
- Declined applications
- Fraud discovered privately
- Personal-guarantee performance
So we separate public distress risk, built entirely on reusable public evidence, from commercial performance risk, which will only exist where lenders contribute outcomes under proper governance. Any product that blurs the two is selling you confidence it doesn’t have.
What rigour means here
State of the art in credit is not the biggest model. It is discipline:
- Strictly point-in-time training data, with no leakage from later filings
- Temporal validation splits, never random ones
- Calibration by sector, company age and size
- Published model cards, benchmarks and known limits
- Independent validation and drift monitoring
Status
Creditbase Risk is in research. Nothing here replaces lender judgement, established bureau data or your own due diligence, and we will keep saying so, because that honesty is the product.