Proxy-Reliance Control in Conformal Recalibration of One-Sided Value-at-Risk

Battle Count: 77

Highly relevant for quantitative risk management in trading desks and asset managers. The framework addresses a practical design question: how much to trust a volatility proxy when recalibrating VaR forecasts. Directly applicable to portfolio risk monitoring, regulatory capital assessment, and stressed-state tail control. The rolling out-of-sample design and ETF panel evidence make findings immediately actionable for practitioners managing equity, commodity, and fixed-income exposures. However, it is a recalibration layer rather than a trading signal generator.

Moderate complexity. The core conformal recalibration with proxy-reliance parameter is straightforward (scaling residuals by v_t^ρ and taking empirical quantiles). However, the full pipeline requires: (1) composite volatility proxy construction from three components with normalization, (2) rolling window management with multiple sub-windows, (3) nested selection of ρ via global-average or stress-aware selectors, (4) regime-aware extension with monotone tuple search, (5) multiple baseline forecaster implementations (HS, FHS, QR, GPQ, GARCH-t, GJR-GARCH-t), and (6) comprehensive backtesting with three statistical tests. The GARCH-style proxy requires expanding-window estimation with fallback rules.

The paper provides detailed implementation specifications including rolling window sizes (504 training, 252 calibration-selection, 126 final calibration), candidate grids for ρ, composite proxy construction formulas, and baseline forecaster definitions. However, code is only available upon request during review, and a public replication repository is promised only after the review process. Data sources (Twelve Data API, CBOE VIX) are publicly available but require API access.