Marking-Aware Sequential VaR Recalibration for Standardized Option Books

Battle Count: 72

Highly relevant for quantitative risk management desks dealing with option books. The paper addresses a practical gap: standard VaR backtests applied to underlying returns or preconstructed loss series ignore the operational reality of contract-level marking, liquidity screens, and book construction. The marking-aware framework and sequential recalibration layer are directly applicable to derivatives trading desks, market risk teams, and regulatory compliance. The leakage-safe design and dual-market validation (SPX index options and QQQ ETF options) make findings broadly transferable. However, the focus is on risk measurement and backtesting rather than alpha generation or trading strategy development.

Moderate to high complexity. Requires: (1) option chain data with bid/ask/midpoint/IV/open interest/volume screens; (2) deterministic book construction rules for three standardized books; (3) next-day marking hierarchy with direct marks, interpolation, and nearest-neighbor fallback; (4) rolling LightGBM quantile regression with 79 features and 252-day window; (5) exponentially weighted residual quantile recalibration layer; (6) comprehensive backtesting suite (Kupiec, Christoffersen, pinball loss, rolling exceedance). The marking hierarchy and book construction logic require careful implementation. The recalibration layer itself is straightforward (weighted empirical quantile of past residuals). Code is provided on GitHub.

Code repository is publicly available at GitHub. Data source (OptionMetrics IvyDB US) is commercial and requires subscription. The methodology is fully specified with explicit equations, parameter choices (W=126, eta=0.01, 252-day rolling window, 79 features), and detailed marking hierarchy. Robustness checks across learners, windows, decay rates, and target probabilities are reported. However, the commercial data source limits full independent replication.