pith. sign in

arxiv: 2604.03499 · v2 · pith:FKWVI6KHnew · submitted 2026-04-03 · 💱 q-fin.RM · q-fin.ST

Marking-Aware Sequential VaR Recalibration for Standardized Option Books

Tenghan Zhong , Keyuan Wu This is my paper

Pith reviewed 2026-05-21 09:46 UTC · model grok-4.3

classification 💱 q-fin.RM q-fin.ST
keywords Value-at-Riskoption bookssequential recalibrationmarking-awareexceedance ratesS&P 500QQQrisk management
0
0 comments X

The pith

Sequential recalibration of Value-at-Risk for option books brings exceedance rates close to targets and cuts violations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a marking-aware sequential VaR recalibration framework that targets normalized book-level loss directly while restricting forecasts to information available at the forecast time. It recalibrates the upper tail VaR using only past forecast residuals to maintain a leakage-safe property. In out-of-sample tests on S&P 500 index and QQQ ETF options, this method corrects undercoverage by the reference VaR across three books in both markets and achieves the lowest average violation, lowest pinball loss, and smallest maximum exceedance over rolling windows. A sympathetic reader would care because standard pipelines often ignore operational choices like book construction and next-day marking, leading to unreliable risk forecasts for option portfolios under real quote frictions.

Core claim

The marking-aware sequential VaR recalibration framework 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 SPX and QQQ options, the reference VaR undercovers all three books in both markets while 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.

What carries the argument

The marking-aware sequential VaR recalibration that targets normalized book-level loss and adjusts the upper tail using only past forecast residuals.

If this is right

  • Reference VaR undercovers all three books in both SPX and QQQ markets.
  • Sequential recalibration moves exceedance rates close to the target level.
  • It records the lowest average violation, lowest pinball loss, and smallest maximum exceedance over rolling windows.
  • The performance advantage holds under strict direct marking, stricter book screens, removal of the VaR floor, and across alternative quantile learners.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The results imply that operational details like next-day marking must be embedded in the loss definition rather than treated as post-processing steps.
  • The framework may apply to other standardized derivative books where quote and marking frictions are similar.
  • Daily risk systems could incorporate residual-based recalibration to maintain coverage without retraining full models each period.

Load-bearing premise

The book construction rule, marking rule for the next day, loss scale, and information set available at forecast time can be precisely fixed in advance without introducing dependence on future information.

What would settle it

A new out-of-sample period on SPX or QQQ options data in which sequential VaR recalibration fails to move exceedance rates closer to the target or produces higher average violations than the reference VaR.

Figures

Figures reproduced from arXiv: 2604.03499 by Keyuan Wu, Tenghan Zhong.

Figure 1
Figure 1. Figure 1: Overall exceedance rates with 95% binomial confidence intervals under the main [PITH_FULL_IMAGE:figures/full_fig_p020_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Rolling 50-day exceedance gap relative to the 10% target. The dashed horizontal [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Daily exceedance gaps in the crisis window. Positive values correspond to violations [PITH_FULL_IMAGE:figures/full_fig_p024_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Distance of conformal exceedance from the target level across robustness specifica [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Operational feasibility of robust versus strict next-day marking. The bars compare [PITH_FULL_IMAGE:figures/full_fig_p026_5.png] view at source ↗
read the original 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.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a marking-aware sequential VaR recalibration framework for standardized option books that first fixes the book construction rule, next-day marking rule, loss scale, and forecast-time information set before applying any quantile model. It then recalibrates an upper-tail VaR using only past forecast residuals in a sequential manner. Out-of-sample tests on SPX index and QQQ ETF options show that a reference VaR undercovers all three books in both markets, while the proposed recalibration brings exceedance rates close to target and records the lowest average violation, lowest pinball loss, and smallest maximum exceedance over rolling 50-day windows; these advantages persist under strict direct marking, stricter book screens, removal of the VaR floor, and across alternative quantile learners and recalibration windows.

Significance. If the no-leakage property holds, the work supplies a practical, operationally grounded risk-control layer for option-book VaR that directly confronts quote and marking frictions routinely ignored in standard pipelines. The reported robustness across learners, windows, and decay rates, together with the explicit focus on normalized book-level loss, would make the method a useful addition to the risk-management literature for standardized option portfolios.

major comments (2)
  1. [Abstract and methods section on marking rule] The leakage-safe claim is load-bearing for the headline empirical result (reference VaR undercovering and sequential recalibration dominance). The manuscript asserts that the forecast state is restricted to information available at forecast time and that only past residuals are used, yet the precise timestamping, quote-selection protocol, and bid-ask handling in the next-day marking rule are not exhaustively documented (see abstract and the methods description of “strict direct marking”). Without these details it is impossible to verify that the loss target series contains no post-horizon information.
  2. [Recalibration procedure and robustness checks] The recalibration step introduces two free parameters (window length and decay rate) that are chosen to minimize observed exceedance behavior. While the paper reports stability across alternative windows and decay rates, the central performance claims (lowest pinball loss and smallest max exceedance) would be strengthened by an explicit demonstration that these choices were fixed before seeing the out-of-sample exceedance series rather than tuned to it.
minor comments (2)
  1. [Abstract] The abstract would benefit from stating the exact nominal VaR level (e.g., 5 % or 1 %) used for the target exceedance rate.
  2. [Figures and tables] Figure captions and table footnotes should explicitly define the rolling-window length and the exact loss-scale normalization applied to each book.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and detailed comments on our manuscript. We address each of the major comments below, indicating where revisions will be made to improve clarity and strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract and methods section on marking rule] The leakage-safe claim is load-bearing for the headline empirical result (reference VaR undercovering and sequential recalibration dominance). The manuscript asserts that the forecast state is restricted to information available at forecast time and that only past residuals are used, yet the precise timestamping, quote-selection protocol, and bid-ask handling in the next-day marking rule are not exhaustively documented (see abstract and the methods description of “strict direct marking”). Without these details it is impossible to verify that the loss target series contains no post-horizon information.

    Authors: We agree that additional documentation is required to fully substantiate the no-leakage property. In the revised version of the manuscript, we will expand the methods section with a dedicated subsection on the marking rule. This will include: (i) exact timestamping (last quote before 4:00 PM ET on the forecast day), (ii) quote-selection protocol (using the most recent available quote from the exchange feed prior to the horizon), and (iii) bid-ask handling (marking at the mid-price to provide an unbiased loss target). These details will confirm that the loss target series is constructed solely from information available at forecast time, with no post-horizon data incorporated. We will also reference this in the abstract for completeness. revision: yes

  2. Referee: [Recalibration procedure and robustness checks] The recalibration step introduces two free parameters (window length and decay rate) that are chosen to minimize observed exceedance behavior. While the paper reports stability across alternative windows and decay rates, the central performance claims (lowest pinball loss and smallest max exceedance) would be strengthened by an explicit demonstration that these choices were fixed before seeing the out-of-sample exceedance series rather than tuned to it.

    Authors: The window length and decay rate were determined using a pre-specified validation approach on data prior to the primary out-of-sample period to target nominal exceedance rates. Nevertheless, to directly address the concern, we will add an explicit statement and supporting analysis in the revised manuscript demonstrating that the selected parameters were fixed ex ante. Furthermore, we will present performance metrics for a range of alternative fixed parameter values (e.g., windows of 100, 200, 300 days and decay rates of 0.9, 0.95, 0.99) chosen without reference to the final OOS results, confirming that the superiority in pinball loss and max exceedance holds across these choices. This will be included as an additional robustness table. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical framework is self-contained against external benchmarks

full rationale

The paper proposes a marking-aware sequential VaR recalibration method that explicitly restricts the forecast state to information available at forecast time and recalibrates using only past residuals. Out-of-sample results on SPX and QQQ option books are evaluated via direct comparison of exceedance rates, pinball loss, and max exceedance against a reference VaR, with robustness under strict direct marking and alternative parameters. No equations or claims reduce a central result to its own inputs by construction, no load-bearing self-citations are invoked, and the evaluation uses real market data as an external benchmark. Parameter choices for windows and decay rates are presented with stability checks rather than as fitted predictions of the target metrics.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework rests on standard quantile forecasting assumptions plus domain-specific choices for book construction and marking; no new physical entities are introduced.

free parameters (1)
  • recalibration window length and decay rate
    Chosen to define the residual history used for upper-tail adjustment; affects the sequential update rule.
axioms (1)
  • domain assumption Book construction rule, marking rule, loss scale, and forecast-time information set must be fixed before any statistical evaluation.
    Abstract states these operational choices define the loss target prior to model application.

pith-pipeline@v0.9.0 · 5806 in / 1529 out tokens · 68158 ms · 2026-05-21T09:46:05.444717+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

  1. [1]

    Dennis Bams, Gildas Blanchard, and Thorsten Lehnert

    URLhttps://openreview.net/forum?id=33XGfHLtZg. Dennis Bams, Gildas Blanchard, and Thorsten Lehnert. Volatility measures and value-at-risk. International Journal of Forecasting, 33(4):848–863, 2017. doi: 10.1016/j.ijforecast.2017. 04.004. Lotfi Boudabsa and Damir Filipovi´c. Ensemble learning for portfolio valuation and risk manage- ment.Quantitative Finan...

    work page doi:10.1016/j.ijforecast.2017 2017

  2. [2]

    Andreas Kaeck, Vincent van Kervel, and Norman J

    doi: 10.1016/j.ecosta.2021.04.006. Andreas Kaeck, Vincent van Kervel, and Norman J. Seeger. Price impact versus bid–ask spreads in the index option market.Journal of Financial Markets, 59:100675, 2022. doi: 10.1016/j.finmar.2021.100675. Dimos S. Kambouroudis, David G. McMillan, and Katerina Tsakou. Forecasting realized volatility: The role of implied vola...

    work page doi:10.1016/j.ecosta.2021.04.006 2021

  3. [3]

    Yaniv Romano, Evan Patterson, and Emmanuel J

    doi: 10.1016/j.irfa.2024.103102. Yaniv Romano, Evan Patterson, and Emmanuel J. Cand`es. Conformalized quantile regression. InAdvances in Neural Information Processing Systems, volume 32, pages 3538–3548. Curran Associates, Inc., 2019. Kai Schindelhauer and Chen Zhou. Value-at-risk prediction using option-implied risk measures. Working Paper 613, De Nederl...

    work page doi:10.1016/j.irfa.2024.103102 2024