Risk Governance Through Long-Term Risk Modelling: An Enhanced Filtered Historical Simulation Approach for Financial Institutions

Financial institutions must produce coherent tail-risk measures across multiple regulatory horizons-from short-term market-risk monitoring to longer-term capital and solvency assessments-under stringent model-risk governance expectations. We develop a governanceoriented framework within the historical-simulation family that preserves empirical crosssectional dependence through multivariate resampling while remaining operationally scalable for large portfolios. We compare standard historical bootstrap engines with filtered historical simulation architectures and introduce LT-FHS, a long-horizon extension that constrains cumulative shocks using historically observed levels and volatility-conditioned dynamic buffers. The empirical assessment combines portfolio-level out-of-sample backtesting at market-risk horizons with a risk-factor-level tail-quantile accuracy exercise that benchmarks simulated tails against an empirical resampling proxy across horizons. Covering interest-rate, creditspread, and equity risk classes, filtered architectures are consistently competitive at marketrisk horizons, while LT-FHS delivers the best long-horizon tail-quantile accuracy and the closest match to empirical tail benchmarks, with gains concentrated in stressed regimes. These results provide a practical, regulator-aligned playbook for selecting scenario engines over different holding periods in banking and insurance internal-model settings.