Marking-Aware Sequential VaR Recalibration for Standardized Option Books

TL;DR

This paper introduces a marking-aware sequential VaR recalibration method that improves risk estimation accuracy for option books by directly targeting normalized book-level loss and using only past residuals, outperforming traditional approaches.

Contribution

It proposes a novel VaR recalibration framework that accounts for operational choices and directly targets book-level loss, enhancing risk control in options trading.

Findings

Sequential VaR recalibration achieves near-target exceedance rates.

Method outperforms benchmarks with lower violations and pinball loss.

Robust across various market conditions and model configurations.

Abstract

Daily Value-at-Risk (VaR) for option books requires more than an accurate quantile forecast. It first requires a precise definition of the loss target. Before any model is evaluated, the protocol must fix the book construction rule, the marking rule for the next day, the loss scale, and the information set available at forecast time. Common pipelines instead apply VaR methods to underlying returns or preconstructed book loss series, leaving these operational choices outside the statistical target. We propose a marking-aware sequential VaR recalibration framework that targets normalized book-level loss directly, restricts the forecast state to information available at forecast time, and recalibrates an upper tail VaR using only past forecast residuals. In out-of-sample evaluation on S\&P 500 index (SPX) and QQQ exchange-traded fund (ETF) options, the reference VaR undercovers all three books in both markets. Sequential VaR recalibration moves exceedance rates close to the target and delivers the best aggregate performance across books, with the lowest average violation, lowest pinball loss, and smallest maximum exceedance over rolling 50 trading day windows among the evaluated methods. Robustness checks preserve the same conclusion under strict direct marking, stricter book selection screens, and removal of the VaR floor. The result is also stable across alternative quantile learners, residual recalibration windows, and decay rates. These findings support marking-aware sequential VaR recalibration as a leakage-safe risk control layer for option-book VaR under realistic quote and marking frictions.