Methodology

Counterparty Insight publishes the technical methodology and back-testing for its quantitative risk-rating system so that customers, their model-risk-management teams, and external reviewers can validate the approach independently. All documents below are versioned and updated as methodology changes.

• PD Model Methodology

  Full technical methodology for the Counterparty Insight probability-of-default model: feature set, training corpus (17 years of FDIC failure history), XGBoost calibration, 17-letter scale, and ongoing PSI monitoring.

• Bank Risk Rating Methodology

  Full technical methodology for the Counterparty Insight CAMELS-based Risk Rating: 17 ratios across six components, asset-size peer groups, frozen pooled percentile thresholds, machine-learned weight refinement with shrinkage to an analyst prior, large-bank weight override, and hard-floor qualitative adjustments.

• PD Model vs Agency Ratings (Back-Test)

  Back-test of Counterparty Insight's probability-of-default model against agency ratings (S&P, Moody's, Fitch) on the population of bank holding companies that have both an agency rating and an FDIC-supervised bank subsidiary.

• PD Model Letter Distribution Explainer

  Why the PD model places ~76% of FDIC-insured banks in the A+/A/A- tiers. Structural reasons rooted in the population's low base failure rate (0.1 to 0.5% historical) versus corporate bond defaults (0.5 to 5%), and how the master scale bands are calibrated.