Daily Summary for 2026-11-09
daily
Transcript
Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.
Ruth: It's the ninth of November, twenty twenty-six, and this is the day's research.
Warren: 15 new papers came out today.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We'll take the day in one pass, then pull out the papers we're staying with.
Warren: Start us off with AIDA: An AI Outage Registry with Evidence from On-Chain MEV.
The day's research: Ruth: Generative AI lowers software writing costs but lacks systematic failure records.
Warren: AIDA tracks five thousand eight hundred sixty-eight outages across thirty providers from twenty twenty five to twenty twenty six.
Iris: That registry pairs outage data with maximal extractable value transactions on Ethereum and Base Layer two.
Ruth: This shows code availability expands where public state is enough, but trade execution remains concentrated.
Warren: Beyond Anonymous Order Flow uses persistent wallet histories to predict short-term returns based on wallet scores.
Iris: Those scores correlate with markouts across windows, and top-decile wallets significantly boost prediction performance in certain markets.
Ruth: PanelGP shows firm characteristics and economic conditions determine factor loadings for asset returns using Gaussian processes.
Warren: So individual wallet signals matter, but broader economic models still provide a robust framework for understanding asset risk.
Iris: The most significant finding addresses return drag and regime whipsaws in sector rotation.
Ruth: Symmetric variance penalizes upside too harshly, while unconstrained hidden Markov models cause excessive instability through rapid switching.
Warren: Asymmetric Downside-Risk Nested Clustered Optimization uses a downside semi-covariance matrix to address this issue.
Iris: This approach elevated the compound annual growth rate from ten point six five percent to thirteen point four one percent over six hundred fifty months.
Ruth: Simultaneously, this framework cut regime whipsaws by seventy-five percent.
Warren: That builds on volatility response work showing asymmetry is positive for cryptocurrencies, matching the leverage effect.
Iris: The Information-Theoretic Model suggests a proxy for growth-optimal portfolios follows a Student t distribution with degrees of freedom slightly less than four.
Ruth: Return Decomposition in the Brazilian Market found the small and medium-sized business factor showed high significance despite momentum not being significant.
Warren: That suggests value criteria tilt selections toward quality firms, contrasting with structural issues targeted by Asym-Sticky-NCO.
Lucky paper: 2611.01656: Ruth: So we're looking at AIDA: An AI Outage Registry with Evidence from On-Chain MEV. This paper is mapping how generative AI tools are affecting trading infrastructure by tracking outages and seeing how that ripples through the system.
Warren: It's showing five thousand eight hundred sixty-eight deduplicated outage incidents across thirty providers from twenty twenty five to twenty twenty six, paired with maximal extractable value transactions on Ethereum and Base Layer two.
Iris: That archive includes verified maximal extractable value transactions by eight thousand five hundred sixteen Ethereum addresses, their contract creations, and a two-two month panel of one point zero nine nine million five hundred ninety-three thousand one hundred ninety-one swaps and twenty-nine thousand one hundred twenty private executors.
Ruth: The researchers found that the entry pattern shows that the expanding base of MEV developers enters exactly the strategies where code and public state suffice, which is consistent with cheaper software lowering the building barrier.
Iris: Specifically, active arbitrage and liquidation addresses expanded markedly between the fourth quarter of twenty twenty four and the third quarter of twenty twenty six, while sandwich attackers contracted.
Warren: Execution remained concentrated among leading operators because the top fifty addresses execute about two thirds of verified trades.
Ruth: But what about when things go wrong? The paper looks at how AI outages stall new trading code while deployed bots keep trading across both chains.
Iris: On Ethereum, addresses releasing a new executor contract decline sharply on the day of a multi-hour disruption and rebound within days, which suggests developers rely on AI coding services for off-chain engineering but live execution runs autonomously when services fail.
Warren: That means the development pipeline is separate from the live execution bots during an outage.
Ruth: So what's the actual implication for someone just listening to this show? It suggests that while cheaper code lets you build more strategies, the real money keeps flowing through a small group of established operators even when new tools fail.
Iris: The study notes that private order flow, latency and capital still determine who wins execution, even as cheaper code lowers the barrier to building MEV contracts.
Warren: That confirms that the infrastructure layer is still heavily dependent on those few key addresses for actual trade execution.
Lucky paper: 2611.07120: Ruth: Okay, we're moving into paper number two of five today with us, and this one is called A Rough Instantaneous Squared VIX Model for Term Structure and Option Pricing by Tianfeng Yang, Wen Su, and Blanka Horvath.
Warren: It proposes a rough instantaneous squared VIX model using fractional memory into the conventional Markovian framework to derive an explicit term structure.
Iris: That introduces a fractional kernel K of t equals t alpha minus one over Gamma of alpha into the dynamics.
Ruth: So, it leads to an explicit VIX term structure governed by a Mittag-Leffler decay function and a semi-closed form pricing formula for VIX options.
Warren: The model is defined by the process Vt equals V0 plus Z t zero n K of t minus s over kappa of theta s minus Vs ds plus sigmaV p Vs dB1s, and theta t equals theta zero plus sigma theta Z t zero p theta s dB2s.
Iris: Those parameters have to satisfy kappa, mu y greater than zero and alpha between one over two and one.
Ruth: The terminal squared VIX is represented by Equation eleven which shows that the time-zero squared VIX curve moves monotonically from V0 towards theta zero.
Warren: The diffusion and jump parameters don't enter this curve, but they affect the distribution of future VIX through the history term in Equation eleven.
Iris: They are focusing on how the squared VIX is represented as tau minus one E Q t R t plus tau t ds.
Ruth: The option pricing is achieved through a Fourier-Laplace transform, and Theorem three point two guarantees that the VIX transform is finite on every vertical contour u equals eta plus i xi with eta between zero and eta bar T tau zero.
Warren: The resulting option value comes from solving one nonlinear Volterra equation for each transform argument and evaluating a one dimensional numerical integral.
Iris: Empirically, they used daily VIX term structure observations and quotes from January two thousand eighteen to June two thousand twenty-five to validate the RISVIX model.
Ruth: They found that RISVIX improves the fit of the VIX term structure, with larger gains for its slope and curvature during periods of elevated volatility.
Warren: It also improves out-of-sample VIX option pricing, reducing the mean squared error by thirteen point zero six percent and the mean absolute error by six point three three percent.
Iris: Those gains are larger for shorter maturities and in high VIX states, which is interesting given the model's focus on those regimes.
Ruth: Model comparison showed RISVIX achieves the lowest mean squared error and mean absolute error among five models tested for both periods.
Warren: Relative to LZZ, the full sample mean squared error reductions are ten point four two percent for VIX levels, seventeen point two one percent for slopes, and thirteen point one eight percent for curvature.
Iris: The option pricing results show a similar pattern across those comparisons.
Ruth: The paper concludes that RISVIX improves term-structure fit, particularly during periods of market stress like January to June two thousand twenty where the reductions increase to forty-six point five nine percent for curvature.
Warren: With global parameters fixed at their training sample estimates and date-specific initial states updated from the contemporaneous VIX term structure, RISVIX reduces out-of-sample mean squared error by thirteen point zero six percent and mean absolute error by six point three three percent again.
Iris: The gains are larger for shorter maturity options and high VIX states, which suggests where this model is most effective right now.
Ruth: Future work could extend RISVIX to a joint SPX minus VIX framework to impose consistency across equity and volatility markets.
Warren: That would test how well the fractional memory handles that cross-market synchronization.
Lucky paper: 2611.03022: Ruth: Alright, we’re looking at the paper from Cornell about interpretable allocation and model-predictive rebalancing of equity portfolios, tested on real and synthetic markets.
Warren: It tests three core challenges: stock selection, timing trades, and strategy robustness beyond observed market conditions.
Iris: They developed a system testing across three hundred seventy-three US equities using data calibrated from two thousand fourteen to two thousand twenty-four and trading from January two thousand twenty-five to September twenty twenty six.
Ruth: The system incorporates various stock selection methods, like score-based methods or bandit approaches, and seven distinct weighting schemes.
Warren: They also use four consumer-theory utilities and three standard portfolios when setting those weights.
Iris: Empirically, they found that the stocks chosen impacted terminal wealth more than the weighting rule itself.
Ruth: That’s interesting because it means even if you have a bad weighting scheme, picking the right stocks can win.
Warren: And they showed early re-planning mechanisms had a positive effect by reducing median worst drawdowns in most combinations compared to fixed calendar strategies.
Iris: In severe simulated declines, even random selections performed well relative to benchmark indices like SPY and the equal-weight universe in the median case.
Ruth: But they did mention that when testing on synthetic markets generated by hidden Markov models with jumps, the advantage from stock selection was less pronounced.
Warren: They observed that individual stock intercepts derived from single-index fits tended to shrink toward the average across all stocks in those synthetic tests.
Iris: That suggests any advantages seen in synthetic tests might come more from the generator's design than genuine real market persistence.
Ruth: The allocation rule used scores based on expected growth rates, modified by absolute beta raised to a power dependent on recent market direction.
Warren: And that whole thing is bounded by a hyperbolic tangent function when setting those scores.
Iris: The rebalancing controller forecasted the holdings' value over a twenty one day horizon using multi-asset geometric Brownian motion and executed trades only when predefined triggers were hit.
Ruth: Those triggers were either when the portfolio value exited its forecast band or when a drawdown limit was breached.
Warren: The system was rigorously tested on real prices and synthetic markets generated by those hidden Markov models featuring jumps linked by market and sector factors.
Iris: Those synthetic markets successfully reproduced the heavy tails observed in individual stocks within that multi-asset environment, which is important for risk modeling.
Ruth: The paper also acknowledged limitations, specifically mentioning survivorship bias and the exclusion of tax modeling when evaluating these systems holistically.
Lucky paper: 2611.04119: Ruth: Okay, we're moving on to the paper titled Multidimensional Views Behind Target-Price Revisions, and this is number four of five this show.
Warren: This research uses a large language model to extract analyst outlooks from research reports as question-based outlook scores.
Iris: So instead of just looking at one target price, the paper breaks down analyst judgments into ten different question-based outlook scores with supporting evidence.
Ruth: It's trying to capture things like expected year on year EPS growth and earnings relative to market consensus.
Warren: The authors used a sample of forty-one thousand six hundred forty single stock equity research reports on US listed companies issued by one global brokerage between January two thousand seventeen and August two thousand twenty-six.
Iris: For each report, they calculate the current target price revision from the published target price and the prior target price.
Ruth: They then convert each report into a vector of ten question-based outlook scores, where each score is on a seven point scale from minus three to plus three.
Warren: These ten dimensions cover financial performance, valuation, shareholder distributions, competitive positioning, industry conditions, regulation, and catalysts.
Iris: The evaluation looked at two tasks: identifying the contemporaneous target price revision and predicting the subsequent revision.
Ruth: The question-based scores representation outperformed both financial dictionary and FinBERT sentiment measures on all three tasks.
Warren: Sector level results showed that the relevance of each outlook dimension depends on the sector, with industry conditions being particularly prominent for upward revisions in materials and downward revisions in industrials.
Iris: This multidimensional view provides a way to inspect textual basis for each assessment, so users can see what kind of analyst sentiment is driving a revision.
Ruth: That sounds like it gives investors something beyond just the final number they see on a screen.
Warren: It does, and the structure allows users to aggregate those scores to examine how reported conditions vary across industries and over time.
Iris: So this paper suggests that using these ten scores gives incremental predictive information beyond what a single target price can convey.
Lucky paper: 2610.15973: Ruth: We’re moving on to paper number five today, Shared AI Decision Interfaces and Credit Allocation in a Simulated Lending Market.
Warren: This one’s about testing whether changing the order of actions alters how much credit gets executed after collateral constraints are set.
Iris: It seems they tested an effect of lever-up-first versus delever-first common-market contrast, finding a T value of one point one two units of initial unused reference borrowing capacity.
Ruth: That sounds like a small difference, but the paper is focused on how shared decision interfaces affect credit execution.
Iris: They did validation under three small open-weight model families and found that the prospectively locked randomization test rejected the sharp null that common-market net debt is invariant to complete-order assignment with a p value of two over eight thousand.
Warren: So, even with this revised interface I one, they saw all twenty four markets increase debt.
Ruth: And where did the gains actually come from in terms of the exploratory accounting split?
Iris: The exploratory accounting split showed that lever-up-first common markets borrowed more and repaid less than delever-first common markets.
Warren: They also narrowed the observed common-market order gap from zero point eight two eight to zero point six eight eight capacity units when using reference valuation.
Iris: The financial significance is tied to the positions that creditors fund and recover, as borrower level debt and collateral determine conditional recovery under liquidation.
Ruth: That brings us to the exposure reporting section, how does that translate into something practical for a trader?
Iris: The worst three of twelve common markets under all three headline rules show a ratio of means at thirty percent that lies between one point six and two point five.
Warren: And the worst three markets block numbers three, four, and eight under all rules, carrying one thousand three hundred forty two USDC against five hundred sixty six for the worst three balanced markets with no feedback.
Iris: The ratio of those worst-three means is between one point six and two point five, and the common arm’s exposure sits in the lever-up-first markets where no feedback scenario write-offs average one thousand two hundred eighty four USDC, which is eight point seven percent of their pre shock debt.
Ruth: What about the pool price trigger they mentioned? How does that change things?
Iris: The pool price trigger at a ten percent shock has four lever-up-first common markets with positive exposure, at most five hundred thirteen USDC, which is three point one percent of that market’s debt.
Warren: It sounds like the spot valued replay reproduces every recorded execution and price under the reference valued cap.
Iris: That's right, the observed gap narrows from zero point eight two eight to zero point six eight eight capacity units with reference valuation, and common-market borrowing averages one thousand four hundred fifteen K zero and repayment is three hundred thirty five K zero under the spot valued cap.
Ruth: So, even with this simulated environment, they still found a difference in how debt is handled based on action order.
Iris: The pre-specified test rejects complete-order invariance with a p value of two over eight thousand, which is the smallest attainable two sided value.
Warren: It seems only the observed plan and its lever up or delever mirror produce a T value this large, as shown in appendix figure A two.
Iris: The pre specified estimate is T equals one point one two units of initial unused reference borrowing capacity.
More episodes
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- The day in conversation for 2026-11-18
- Daily Summary for 2026-11-18
- 2601.05274-On the use of case estimate and transactional payment data in neural networks for individual loss reserving
- 2009.00868-Loss-Given-Default Modeling by Post-Last Passage Time Process
- 2606.30070-Financial Resilience Evaluation: From Conditional Expectations to Dynamic Convex Risk Measures
- 2212.12687-Measuring price impact and information content of trades in a time-varying setting
- Daily Summary for 2026-11-17
- 2401.08323-Dynamic portfolio selection under generalized disappointment aversion