Skip to content

2025 Q4 Quarterly Research Archive

Records scoring at least 40 within the primary scope pass rule review and are published without additional manual review. This page does not validate author claims or provide investment advice.

  • Coverage: 2025-10-01 to 2025-12-31
  • Passed rule review: 16
  • Sources: 9

Topic distribution

Domains

  • volatility: 7
  • option returns: 5
  • microstructure: 5
  • hedging exposure risk: 3
  • portfolio construction risk transfer: 3
  • execution costs: 2

Methods

  • financial ml: 7
  • research methods: 7

Facets

  • instrument vix options: 2
  • instrument single stock options: 2
  • instrument etf options: 1
  • horizon 0dte: 1
  • exposure gamma: 1
  • exposure vega: 1

Passed rule review

Overnight Reversals of Implied Higher Moments and Their Put‐Call Spreads

  • Published: 2025-12-29
  • Source: Journal of Futures Markets
  • Publication status: peer_reviewed
  • Original source: Open original source

Why it matters

This paper extends overnight reversal analysis to implied higher moments and put-call spreads, affecting dimensional risk-characteristic modeling, though only abstract-level independence and asymmetry claims are available. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5)

Main author claims

  • The authors claim overnight reversals extend beyond spot and implied volatility to implied higher moments and associated put-call spreads. (abstract:S1, abstract:S2)
  • They claim reversals are largely independent across variables with limited spillover, asymmetric for underlying returns and implied volatility but not for higher moments, and driven in spreads by both call- and put-implied moments. (abstract:S2, abstract:S3, abstract:S4, abstract:S5)

Data, method, or discussion scope

Scope includes phenomenon statements, stated independence, asymmetry, and multidimensionality; no sample windows, error diagnostics, or execution implications are provided. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5)

Main limitations

Significance and independence are asserted without explicit statistical controls; robustness across maturities, position constraints, and liquidity differences is not shown. (abstract:S2, abstract:S5, abstract:S1, abstract:S3)

Inferring Latent Market Forces: Evaluating LLM Detection of Gamma Exposure Patterns via Obfuscation Testing

  • Published: 2025-12-27
  • Source: arXiv Quantitative Finance History
  • Publication status: preprint
  • Original source: Open original source

Why it matters

This preprint proposes obfuscation testing to distinguish responses to structural patterns from temporal association, which matters for model governance and prompting design; the reported detection rates do not by themselves validate causal reasoning. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6)

Main author claims

  • The authors claim to introduce obfuscation testing to validate whether LLMs detect structural market patterns via causal reasoning rather than temporal association. (abstract:S1)
  • They report 71.5% detection on unbiased prompts and stable 91.2% under certain profit variations, with 100% when regime labels are included, arguing recognition of structure over profitable patterns. (abstract:S2, abstract:S4, abstract:S5)

Data, method, or discussion scope

Reviewable scope includes the framework, dataset coverage, and reported detection rates; prompt specifications, statistical significance, and computational cost are not provided. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6)

Main limitations

Coverage and period are specific, and external-period transfer is not shown; using profit variation stability as structural evidence may still conflate with sample preference or construction effects. (abstract:S2, abstract:S4, abstract:S6, abstract:S1)

Why SIG Tells Traders Not to Hedge! - Ex-SIG Trader and Moontower Founder, Kris Abdelmessih

  • Published: 2025-12-27
  • Source: Odds on Open
  • Publication status: unknown
  • Original source: Open original source

Why it matters

The interview frames hedging in options market making as a portfolio-level trade-off among expected value, residual exposure, and P&L variance, making it useful for questioning mechanical trade-by-trade hedging assumptions. (description:S2, description:S3, description:S4)

Main author claims

  • The authors report: The guest says SIG often did not immediately hedge positive-expected-value trades unless residual exposure distorted expected value or created excessive P&L variance. (description:S2, description:S3, description:S4)
  • The authors report: The publisher says this centralized risk approach helped retain options market-making edge, tighten spreads, and avoid mechanical hedging. (description:S4, description:S5)

Data, method, or discussion scope

The material is only the publisher's description of one practitioner interview; it also covers historical trading floors, order-flow toxicity, trader training, and inventory, vega, and volatility-regime risk in natural-gas options. (description:S4, description:S5)

Main limitations

The supplied material contains no transcript, position data, comparison sample, or reproducible model; claims about edge, spreads, and competitive advantage come from the promotional episode description and are not independently verified. (description:S1, description:S3, description:S5)

SIG Director: How Susquehanna Trains Top Traders with Poker

  • Published: 2025-12-24
  • Source: Odds on Open
  • Publication status: unknown
  • Original source: Open original source

Why it matters

The episode places probabilistic reasoning, Bayesian updating, order-flow interpretation, and team communication in one training framework, offering context on how a market maker develops real-time decision skills. (description:S2, description:S3, description:S4)

Main author claims

  • The authors report: The guest says SIG trains traders with poker, probabilistic games, and decision-making under uncertainty, emphasizing Bayesian updating and asymmetric information in live-market contexts. (description:S2, description:S3)
  • The authors report: The guest says strong traders need humility, truth-seeking, and the ability to update beliefs quickly when new information arrives, in addition to technical skill. (description:S4, description:S5)
  • The authors report: The publisher describes trading systems, collaboration, parameter tuning, and model monitoring as important components of modern quantitative-trading performance. (description:S6)

Data, method, or discussion scope

The material is a publisher description covering SIG recruiting, training, desk collaboration, market making, prediction markets, and risk transfer, rather than an empirical study of training effectiveness. (description:S2, description:S5, description:S6)

Main limitations

The supplied material provides no curriculum, trainee sample, performance metric, comparison group, or full transcript, so it cannot establish whether poker training causally improves trading outcomes. (description:S1, description:S2, description:S6)

Unified GARCH-Recurrent Neural Network in Financial Volatility Forecasting

  • Published: 2025-11-21
  • Source: arXiv Quantitative Finance History
  • Publication status: preprint
  • Original source: Open original source

Why it matters

This preprint embeds GARCH dynamics into GRU/LSTM gates and claims outperformance across several metrics plus efficiency gains, making it relevant to volatility-model design and interpretability trade-offs; the benchmark specifics remain unverified. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6, abstract:S7, abstract:S8)

Main author claims

  • The authors claim hybrid GARCH-GRU and GARCH-LSTM architectures that incorporate GARCH(1,1) updates inside recurrent gates. (abstract:S1, abstract:S2, abstract:S3)
  • They claim consistent outperformance over classical GARCH, pipeline hybrids, and Transformer baselines, and state GARCH-GRU trains faster while remaining robust in normal and turbulent periods, including 99% VaR diagnostics. (abstract:S4, abstract:S5, abstract:S6, abstract:S7)

Data, method, or discussion scope

Scope includes architecture claims, listed metrics, and stress performance claims, but lacks sample sizes, baseline implementations, resource budgets, and outlier handling details. (abstract:S1, abstract:S2, abstract:S4, abstract:S5, abstract:S6, abstract:S7, abstract:S3)

Main limitations

Speed gains and robustness are not anchored to hardware and confidence intervals, and interpretability of parameters across assets is not operationalized. (abstract:S5, abstract:S7, abstract:S3, abstract:S6)

"Hands-On AI Trading with Python, QuantConnect, and AWS" with Jiri Pik, Ernest Chan, and Jared Broad

  • Published: 2025-11-14
  • Source: Quantopian Webinars
  • Publication status: unknown
  • Original source: Open original source

Why it matters

This item is a webinar description about AI trading education and tools; it is relevant to educational resource context, not empirical validation. Value lies in clarifying tooling and data-access assumptions rather than research claims. (description:S5, description:S7, description:S8, description:S10, description:S11, description:S24, description:S27, description:S28)

Main author claims

  • The authors report: The description states the session covers AI/ML methods from classical models to deep learning, LLMs, and reinforcement learning, and demonstrates building testable trading strategies. (description:S5, description:S6, description:S10)
  • The authors report: It highlights full code and examples on GitHub and Jupyter resources while explicitly disclaiming that content is informational only and not investment advice, with no guarantee of completeness. (description:S8, description:S11, description:S24, description:S25, description:S26, description:S27)

Data, method, or discussion scope

Reviewable scope is instructional scope and disclaimer text, without quantified model-performance evidence or reproducible experimental outcomes. (description:S5, description:S6, description:S8, description:S9, description:S10, description:S11, description:S24, description:S25, description:S26, description:S27, description:S28)

Main limitations

As marketing text, it cannot substitute for verifiable empirical claims; assertions about practical applicability are not linked to licensing, data-access, and evaluation assumptions. (description:S5, description:S6, description:S10, description:S24, description:S25)

Volatility's Heartbeat, the AI Boom, and MJ’s Bulls with Equity Armor’s Brian Stutland

  • Published: 2025-11-06
  • Source: The Derivative by RCM Alternatives
  • Publication status: unknown
  • Original source: Open original source

Why it matters

This is episode metadata for a podcast, not a primary study. Its value is limited to source context and risk communication constraints when treating it as a non-structured information stream. (description:S1, description:S2, description:S3, description:S4, description:S6, description:S10, description:S11)

Main author claims

  • The authors report: The description states the episode covers volatility, AI-driven market shifts, and option-strategy perspectives. (description:S1, description:S3, description:S4)
  • The authors report: It also explicitly includes information-only framing, statement-of-opinion language, and non-investment-advice/high-risk disclaimers. (description:S6, description:S7, description:S8, description:S9, description:S10)

Data, method, or discussion scope

The scope is limited to descriptive metadata and disclaimers; there are no quantitative datasets, statistical outputs, or reproducible methods. (description:S1, description:S2, description:S3, description:S4, description:S6, description:S7, description:S9, description:S10)

Main limitations

This is not evidence for empirical claims because no methods or metrics are provided, and viewpoints may be presenter- and guest-dependent. (description:S1, description:S11, description:S4)

Christina Qi Started a Hedge Fund From Her Dorm Room. Now, Top Trading Firms Now Buy Her Data.

  • Published: 2025-11-05
  • Source: Odds on Open
  • Publication status: unknown
  • Original source: Open original source

Why it matters

The interview connects clean tick and order-book data, licensing, backtesting infrastructure, and the limits of ML use, making it relevant to whether research data can support execution and reproducibility. (description:S3, description:S4, description:S5, description:S6)

Main author claims

  • The authors report: The guest says clean order-book and tick data matter for robust strategies, faster research, reliable execution, and scalable quantitative workflows. (description:S3)
  • The authors report: The guest says trading tools were easier to access in 2025 while alpha was harder, requiring high-quality data, disciplined backtesting, and risk management. (description:S4)
  • The authors report: The guest recommends mastering market structure and building real strategies before applying machine learning where it adds genuine incremental value. (description:S5)

Data, method, or discussion scope

The publisher description covers Domeyard, Databento, HFT, market-data licensing, fundraising, product distribution, and quant career advice; it is an entrepreneur interview, not an independent comparative evaluation of data vendors. (description:S2, description:S3, description:S6)

Main limitations

Claims about customer adoption, data quality, cost, Sharpe ratios, and competitive advantage have no benchmark, sample definition, or independent validation in the supplied material, which also lacks a full transcript. (description:S2, description:S3, description:S6)

Determinants of Price Discovery in Option Markets: An Interpretable Machine Learning Perspective

  • Published: 2025-10-30
  • Source: Journal of Futures Markets
  • Publication status: peer_reviewed
  • Original source: Open original source

Why it matters

The paper studies information leadership in SSE50 ETF options and ranks features with interpretable ML. This informs signal interpretation and feature priority, but feature importance does not by itself establish a causal mechanism. (abstract:S1, abstract:S2, abstract:S3, abstract:S4)

Main author claims

  • The authors claim option markets show more information leadership than the underlying market, measured using 1-second Information Leadership Share. (abstract:S1, abstract:S2)
  • They further claim trading cost is the leading explanatory feature, followed by leverage, market-maker risk, and speculation, with robustness checks for feature importance stability. (abstract:S3, abstract:S4)

Data, method, or discussion scope

Scope covers abstract definitions of information leadership, feature ranking, and mention of robustness checks; no sample-window details, significance levels, or microstructure noise handling are provided. (abstract:S1, abstract:S2, abstract:S3, abstract:S4)

Main limitations

Claims of stronger information discovery are not accompanied by explicit boundaries for liquidity regimes, rule changes, or label mismatch effects. (abstract:S2, abstract:S3, abstract:S4)

Jump risk premia in the presence of clustered jumps

  • Published: 2025-10-24
  • Source: arXiv Quantitative Finance History
  • Publication status: preprint
  • Original source: Open original source

Why it matters

This preprint introduces clustered jumps with sign-specific jump risk premia in option pricing, relevant to dynamic skew behavior and sentiment-driven pricing adaptation, though only abstract-level evidence is provided. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6)

Main author claims

  • The authors claim a bivariate Hawkes process for clustered jumps captures self- and cross-excitation of positive and negative jumps, producing time-varying skewness and skews of either sign. (abstract:S1, abstract:S2)
  • They further claim inferred positive and negative jump premia, identified from options data, show predictive power for BTC futures carry cost and delta-hedged option-strategy performance. (abstract:S4, abstract:S5, abstract:S6)

Data, method, or discussion scope

The scope includes abstract-level model mechanics and BTC empirical claims, with no jump-threshold selection, estimation uncertainty, or explicit OOS test diagnostics provided. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S6)

Main limitations

Empirical evidence is centered on BTC, limiting direct generalization, and claims of skew dynamics and predictive power are not tied to explicit significance standards or OOS loss definitions. (abstract:S3, abstract:S6, abstract:S4)

Fusing Narrative Semantics for Financial Volatility Forecasting

  • Published: 2025-10-23
  • Source: arXiv Quantitative Finance History
  • Publication status: preprint
  • Original source: Open original source

Why it matters

This work unifies structured features with unstructured news for volatility forecasting and claims to mitigate look-ahead bias, making it relevant to multimodal inputs and point-in-time data controls; however, only abstract-level claims are available. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5)

Main author claims

  • The authors claim to propose M2VN, a framework combining time-series features with textual news data for volatility forecasting. (abstract:S1)
  • They claim it addresses fusion/alignment of numerical and text modalities, uses an auxiliary alignment loss, and uses point-in-time Time Machine GPT embeddings to mitigate look-ahead bias. (abstract:S2, abstract:S3, abstract:S4)
  • The authors claim that M2VN consistently outperforms the existing baselines in their experiments. (abstract:S5)

Data, method, or discussion scope

The reviewable scope is confined to abstract statements about architecture goals and auxiliary-loss design; it also contains a generic performance claim without baseline definitions or fold-level error decomposition. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5)

Main limitations

Baseline set, significance criteria, and text-noise handling are unspecified; the point-in-time embedding pipeline lacks explicit slicing rules, leaving availability and update assumptions unclear. (abstract:S2, abstract:S3, abstract:S5)

VIX Option Pricing With Detected Jumps

  • Published: 2025-10-16
  • Source: Journal of Futures Markets
  • Publication status: peer_reviewed
  • Original source: Open original source

Why it matters

The paper claims direct modeling of VIX jumps for option pricing and suggests performance gains over standard RV/Heston-Nandi approaches; this is relevant to jump-aware pricing and monitoring design, though details are absent. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5)

Main author claims

  • The authors claim a direct modeling approach that identifies jumps from high-frequency VIX data and incorporates realized jump and bipower variation into conditional variance and jump intensity dynamics. (abstract:S1, abstract:S2)
  • They claim to derive a closed-form pricing formula and report that the jump-based model consistently outperforms conventional realized variance and Heston-Nandi GARCH benchmarks in-sample and out-of-sample. (abstract:S3, abstract:S4, abstract:S5)

Data, method, or discussion scope

Reviewable scope is abstract-level model chain and relative performance claims; it does not provide jump detection thresholds, exact estimation windows, or pricing error metrics. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5)

Main limitations

The outperformance claim is unqualified by significance/tolerance, and no operational triggers are provided for stability of jump-continuous decomposition under noise-dominant periods. (abstract:S4, abstract:S5, abstract:S2, abstract:S3)

Common Factors in Equity Option Returns

  • Published: 2025-10-11
  • Source: The Review of Financial Studies
  • Publication status: peer_reviewed
  • Original source: Open original source

Why it matters

The study builds a testable benchmark factor structure for delta-hedged equity-option returns and separates common cross-sectional factors from a contract-specific time-series factor, helping assess whether a new option signal is merely known exposure. (full_text:S34, full_text:S39, full_text:S40, full_text:S46, full_text:S47, full_text:S55)

Main author claims

  • The authors report a 1996-2021 benchmark sample of 326,537 firm-month observations, forming 185 call and 185 put characteristic-sorted delta-hedged portfolios; at most four PCA factors capture their main time-series and cross-sectional variation. (full_text:S37, full_text:S38, full_text:S39, full_text:S40, full_text:S75, full_text:S76, full_text:S94, full_text:S459, full_text:S460, full_text:S461)
  • The authors report that a sparse four-factor model using EWOP, the historical-minus-implied volatility gap, cash-to-assets, and volatility-of-volatility approaches the latent model's pricing performance; the latter three span the common cross-section while EWOP mainly captures contract-specific time-series variation. (full_text:S46, full_text:S47, full_text:S472, full_text:S473, full_text:S474, full_text:S476)
  • The authors report that the structure extends to out-of-the-money options and 2,990 portfolios formed on a separate set of 150 characteristics, while the option factors have little relation to traditional stock-return factors. (full_text:S52, full_text:S53, full_text:S468, full_text:S469, full_text:S477, full_text:S478)

Data, method, or discussion scope

The evidence combines OptionMetrics equity options with CRSP, Compustat, and I/B/E/S, studying monthly delta-hedged returns, PCA pricing errors, 120-month rolling out-of-sample errors, ATM/OTM portfolios, and alternative characteristic sorts. It tests factor pricing and return comovement, not an independently replicated trading strategy. (full_text:S75, full_text:S76, full_text:S78, full_text:S103, full_text:S112, full_text:S238, full_text:S239, full_text:S241, full_text:S265, full_text:S468, full_text:S471)

Main limitations

The out-of-sample exercise still uses the same U.S. equity-option history with a 120-month rolling estimation window, not cross-market or live validation; the 150-characteristic portfolios share the same period and data system. Avoiding look-ahead bias requires retaining contracts with zero bids or missing terminal quotes, roughly 25% of which are materially less liquid, so factor-pricing fit is not post-execution-cost return. (full_text:S58, full_text:S59, full_text:S60, full_text:S61, full_text:S238, full_text:S239, full_text:S265, full_text:S468, full_text:S471)

Application of Deep Reinforcement Learning to At-the-Money S&P 500 Options Hedging

  • Published: 2025-10-10
  • Source: arXiv Quantitative Finance History
  • Publication status: preprint
  • Original source: Open original source

Why it matters

The preprint compares DRL with Black–Scholes delta hedging using long-run intraday data and walk-forward evaluation, directly addressing costs, risk penalties, and model dependence. (abstract:S3, abstract:S4, abstract:S5, abstract:S7)

Main author claims

  • The authors state that a TD3 agent using six state variables is compared with Black–Scholes delta hedging over nearly 17 out-of-sample years. (abstract:S2, abstract:S3, abstract:S4, abstract:S5)
  • The authors report better DRL performance in volatile or high-cost settings, deterioration under higher risk penalties, and greater stability with longer volatility-estimation windows. (abstract:S8, abstract:S9, abstract:S10)

Data, method, or discussion scope

The abstract is limited to S&P 500 ATM calls, one six-variable time series, walk-forward training, and several annualized metrics; architecture, uncertainty intervals, cost calibration, and position constraints are absent. (abstract:S1, abstract:S3, abstract:S4, abstract:S6, abstract:S7)

Main limitations

The abstract first says deep Q-learning and then specifies TD3, creating a nomenclature inconsistency; “outperform” remains an author result, with insufficient detail to assess statistical stability or transfer to other underlyings. (abstract:S1, abstract:S2, abstract:S8)

A Bayesian Stochastic Discount Factor for the Cross-Section of Individual Equity Options

  • Published: 2025-10-06
  • Source: Journal of Financial and Quantitative Analysis
  • Publication status: peer_reviewed
  • Original source: Open original source

Why it matters

The paper estimates a Bayesian model averaging SDF for single-stock options and identifies volatility spread, return momentum, and jump risk as frequent factors; this matters for factor governance, though model and prior settings are not given. (abstract:S1, abstract:S2, abstract:S3, abstract:S4)

Main author claims

  • The authors claim a Bayesian model averaging SDF approach that outperforms reduced-form benchmarks for option return anomalies and portfolios in-sample and out-of-sample. (abstract:S1, abstract:S2)
  • They further claim the SDF is dense in characteristics with implied-realized volatility spread, option momentum, and jump risk as the most likely factors, and that it aligns more closely with stock-market SDF than bond-market SDF. (abstract:S3, abstract:S4)

Data, method, or discussion scope

The scope is limited to claimed Bayesian averaging setup and factor inclusion frequencies; posterior thresholds, priors, statistical evaluation, and sample breadth are not included. (abstract:S1, abstract:S2, abstract:S3, abstract:S4)

Main limitations

Claims about likely factors are posterior statements, but the abstract gives no robustness checks for prior sensitivity, re-estimation windows, or changes in the model set. (abstract:S2, abstract:S3, abstract:S4)

Uncertain HAR‐RV Models and Their Extensions: A New Perspective on Forecasting the Volatility of China's Crude Oil Futures

  • Published: 2025-10-02
  • Source: Journal of Futures Markets
  • Publication status: peer_reviewed
  • Original source: Open original source

Why it matters

This paper introduces uncertainty-theoretic extensions to HAR-RV, including uncertain quantiles, potentially expanding volatility forecasting under imprecise data, though deployment-facing details are absent. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6)

Main author claims

  • The authors claim traditional HAR-RV is limited by residual assumptions and feature characterization, motivating uncertain HAR-RV and uncertain quantile HAR-RV models with estimation and proofs. (abstract:S1, abstract:S2, abstract:S3)
  • They claim applying the framework to Chinese crude oil futures yields superior predictive performance across quantiles under randomness tests, OOS evaluations, and robustness checks. (abstract:S4, abstract:S5, abstract:S6)

Data, method, or discussion scope

Scope includes the method motivation, theoretical claims, and abstract-level performance comparison; proof details, test statistics, and parameterization choices are not provided. (abstract:S1, abstract:S2, abstract:S3, abstract:S4, abstract:S5, abstract:S6)

Main limitations

Failure cases and superiority claims lack granular diagnostics and thresholding; implementation details and convergence conditions for uncertainty theory are omitted. (abstract:S5, abstract:S6, abstract:S3)