モデルリスク管理キット

Model definition, roles, inventory fields, tiering matrix, development, risk-based validation, monitoring, overlays, vendor models, GenAI scope, issues, reporting

Purpose and use, design options, data and default definition, logistic PD method, calibration and MoC, Gini, PSI and binomial tests, limitations, implementation, monitoring

Independent model validation: soundness, data, outcomes and back-testing, benchmarking, stress tests, findings, conclusion

Hypothetical and actual P&L, exception list and root causes, zone, Kupiec and Christoffersen tests, multiplier effect, desk-level backtesting and PLA, actions

Model details, intended and out-of-scope uses, factors, metrics, data, disaggregated results, ethics, caveats

Bank stress testing programme: governance, test inventory, scenario design, macro variables, methods by risk, results, actions, reverse stress test, use

ECL methodology: scope, governance, segmentation, default, staging and SICR, PD x LGD x EAD x discount example, scenarios and weights, overlays, CECL

Production ML monitoring: performance, drift (PSI), data quality, fairness, latency, alerts, retraining triggers