Quant Letter
The week in quantitative finance research: 90 papers from arXiv, SSRN and RePEc, screened and ranked, 10–16 Oct 2026.
This week pivots toward machine learning in discovery and execution. Foundation models learn market regimes from order flow; genetic programming and RL agents discover alpha factors across assets and frequencies. But a survey of 47 LLM trading systems finds reported margins vanish on retest, and a massive pre-registered sweep of 18,630 crypto strategies finds none viable. See Foundation Models Learn Market Regimes, EvoQuantLab, and LLM-Based Trading Systems Survey.
If you read five papers this week, read these.
Autoregressive transformers trained on limit-order-book message streams learn amortized Bayesian inference over market regimes, matching optimal prediction and enabling accurate parameter calibration.
A survey of 47 LLM trading systems across six markets finds that reported margins shrink or vanish when rerun, with best evidence placing language models upstream of execution.
A review of LLM applications finds that contextual interpretation improves some tasks but measurement error persists, and trading signals do not always generate profits.
Rejected factors earn 1.30 Sharpe in their home market versus 0.85 overall, yielding 3.8% six-factor alpha per year through venue reallocation constrained by signal-to-noise limits.
Topics drawing unusually many papers this week.
2.4% of this week's 755 new papers, 2.2x its share over the previous 3 weeks (1.1%).
5.8% of this week's 755 new papers, 1.4x its share over the previous 3 weeks (4.1%).
1.7% of this week's 755 new papers, 1.9x its share over the previous 3 weeks (0.9%).
What the quant community starred on GitHub this week, from a panel of 10,589 developers who follow financial machine learning (297 of the 860 we checked starred something). Counts only: we never publish who starred what.
Open source analytics agent for agentic analytics and chat-with-data.
★ 1,937TypeScript
MCP server with 139 tools for real-time global intelligence data.
★ 959Python
Production-grade Rust-native trading engine with event-driven architecture.
★ 30.2k+399 this weekRust
Terminal-style trading dashboard built on free public market data.
★ 1,764+342 this weekTypeScriptNew repo
Agent memory with episodic retrieval for financial decision-making.
★ 3+2 this weekPythonNew repo
GPU-accelerated quantitative finance library.
★ 24+15 this weekJupyter Notebook
Financial services tools and libraries.
★ 40.0k+863 this weekPython
Backtesting engine for thousands of trading ideas.
★ 9,394+78 this weekPython
Reverse engineer anything with agents, from app behavior down to native binaries.
★ 283k+229k this weekTypeScript
Agent skill for creating detailed, interactive isometric illustrations of systems.
★ 13.5k+13.5k this weekJavaScriptNew repo
Open-source, clean-room reimplementation of Microsoft Word in pure Rust.
★ 13.5k+11.3k this weekRustNew repo
Tool for automatic PS5 executables porting to Linux and Windows.
★ 39.3k+16.0k this weekC++New repo
Intentional crafting engine for artists, designers, and filmmakers.
★ 29.9k+17.7k this weekRust
Search your computer by file contents: photos, PDFs, documents, recordings, videos.
★ 4,220+4,219 this weekSwiftNew repo
Now published in Production and Operations Management · featured 1 May 2024, 5 days after release; 6 citations
Now published in SIAM J. Financial Math. · featured 23 Oct 2024, 12 days after release; 2 citations
Now published in Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society · featured 30 May 2025, 6 days after release; 1 citations
Filter by venue, topic or fanfare.
No papers match these filters.
Quantitative finance and ML-for-finance preprints
30 papers · 26 figures
Autoregressive transformers trained on limit-order-book message streams learn amortized Bayesian inference over market regimes, matching optimal prediction and enabling accurate parameter calibration.
Fanfare 4ML & AI MethodsFernando AceroPDF
A training-free framework combines genetic programming and LLM diagnosis loops to discover robust alpha factors across daily and hourly frequencies in equities and cryptocurrency.
Fanfare 4ML & AI MethodsHaolin Yang et al.PDF
A reinforcement-learning framework trains multi-turn tool-using LLM agents to discover factor expressions by conditioning on market descriptions and querying real-data feedback.
Fanfare 3ML & AI MethodsPeiqi Jiang et al.PDF
An reinforcement-learning framework trains language models to construct executable symbolic factors from financial literature, aligned via market feedback, outperforming baselines on Chinese and U.S. equity indices.
Fanfare 3ML & AI MethodsJie Huang et al.PDF
A multiscale phase-type Hawkes model measures how fast prices absorb macroeconomic news versus self-exciting echoes, showing direct absorption in about one second and extended responses up to 137 seconds.
Fanfare 3ML & AI MethodsSayak DuttaPDF
StochLOB replaces Gaussian noise with jump-Lévy and stable-noise corruptions to capture bursty limit-order-book flow, improving tail-window stress diagnostics on held-out LOB data.
Fanfare 3ML & AI MethodsArshia Abolghasemi et al.PDF
RICE-Alpha uses event graphs and historical reliability to improve LLM-based stock forecasting, achieving stronger results on Nasdaq-100 and Hang Seng panels than competing LLM agents.
Fanfare 3ML & AI MethodsTong Liu et al.PDF
XForecast introduces a benchmark combining stock prices with curated financial news to predict four-week-ahead returns, using weighted metrics that emphasize news-driven opportunities.
Fanfare 3ML & AI MethodsJaehoon Lee et al.PDF
A generalized Fisher transform and permutation-equivariant flow-matching network generate valid correlation matrices with 100% validity and superior feature distances versus GAN and diffusion baselines.
Fanfare 3ML & AI MethodsGautier MartiPDF
A Monte Carlo Tree Search framework selects predictive features by out-of-sample validation rather than training error, improving stability and interpretability under multicollinearity.
Fanfare 3ML & AI MethodsCameron Mouangue and Tristan CazenavePDF
HE-CV-UCB adaptively eliminates misspecified surrogate models and achieves no-regret Conditional-Value-at-Risk optimization when kernel hyperparameters are unknown.
Fanfare 3ML & AI MethodsValentin Margraf et al.PDF
The study proposes a neural network architecture for market making that models self-exciting order arrivals through learnable exponential features and trains via reinforcement learning to optimize quoting policy.
Fanfare 3Trading, Microstructure & ExecutionFernando Acero et al.PDF
A causal prior-data fitted network trained on synthetic market priors estimates true market impact by isolating price movement from trading flow versus information, addressing bias in classical models.
Fanfare 3Trading, Microstructure & ExecutionDennis Thumm and Ying ChenPDF
Six defects from data, code, and configuration caused live intraday trading to diverge from backtest replay, none visible in summary statistics, each requiring specific controls.
Fanfare 2Trading, Microstructure & ExecutionAlex Vidovich and Zivana ZerjalPDF
A dual-timescale multi-agent framework generates trading strategies conditioned on current risk and maintains ongoing position governance between strategy revisions on cryptocurrency, preserving returns after transaction costs.
Fanfare 2Trading, Microstructure & ExecutionJunzhou Su et al.PDF
The research derives break-even screening rates for liquidity providers defending against stale quotes, showing that one-second repricing requires far less screening than Gaussian models predict.
Fanfare 2Trading, Microstructure & ExecutionAleksei Adadurov et al.PDF
The study shows that automated market maker pools matched to transaction-cost-aware benchmarks have zero loss-versus-rebalancing, clarifying the investment interpretation of pool design.
Fanfare 2Trading, Microstructure & ExecutionMasaaki FukasawaPDF
The research develops a statically and dynamically arbitrage-free model for European option prices with smooth interpolation, fitting market data within bid-ask spreads for over 95% of relevant options.
Fanfare 3Derivatives & VolatilityMagnus Wiese et al.PDF
The paper shows that combining dynamic trading in the underlying with static positions in three vanilla options reduces variance-swap hedging error by more than 95% across different market roughness regimes.
Fanfare 3Derivatives & VolatilityPaolo Di Tella et al.PDF
An MLP-initialized reinforcement-learning framework optimizes exit timing for short-straddle options, achieving 37.54% cumulative return and a Sharpe ratio of 1.705 out-of-sample over 2024-2026.
Fanfare 3Derivatives & VolatilityJianning Wang et al.PDF
A continuous-time volatility model incorporating realized measures forecasts one-day-ahead integrated variance, showing strong out-of-sample performance relative to GARCH and related benchmarks across 17 equity indices.
Fanfare 2Derivatives & VolatilityHongyu Yao et al.PDF
A first-order attribution framework maps feature exposures to portfolio-performance sensitivities, enabling post-training intervention that improves net Sharpe ratios without retraining.
Fanfare 3Portfolio & AllocationYunqi Liang et al.PDF
A data-driven approach combines vine-copula simulation with reinforcement learning to optimize dynamic portfolio selection across equities and gold with CVaR penalties.
Fanfare 2Portfolio & AllocationGabriël M. J. Gelissen and Fengyi YuanPDF
A gradient-boosted classifier predicts pension sponsor distress one to two quarters ahead from SEC filings, reaching 0.887 AUC-ROC, with application to dynamic self-investment caps in sponsor securities.
Fanfare 3Risk, Credit & BankingRavil AkhtyamovPDF
Sequential backtests for Expected Shortfall using only observed losses and ES forecasts control false rejections, with time-series extension under AR-GARCH structure proving convergent under grid refinement.
Fanfare 2Risk, Credit & BankingRuodu Wang and Jingcheng YuPDF
Using Ginnie Mae buyout options, the study finds that nonbank issuers repurchase fewer delinquent loans when funding markets freeze, delaying about $19 billion in mortgage resolution compared to depositories.
Fanfare 2Risk, Credit & BankingArka P. BandyopadhyayPDF
A realized tail-risk model using high-frequency intraday data forecasts value-at-risk and expected shortfall with competitive performance across U.S. stock returns and multiple tail levels.
Fanfare 2Econometrics & ForecastingKeisuke Sasaki and Yasuhiro OmoriPDF
The study characterizes how latent common signals generate spurious cross-excitation in Hawkes models and uses third-order cumulants to identify true excitation mechanisms, with applications to cryptocurrency data.
Fanfare 2Econometrics & ForecastingJiwon Jung and Kiseop LeePDF
GPT-4o evaluates research correctness with predictive value for publication but cannot judge originality or significance, suggesting technical peer review can delegate to models while contribution assessment requires people.
Fanfare 4LLMs & TextHai Lu et al.PDF
Passive-voice fraction in earnings calls predicts temporary mispricings for good news and longer-run underperformance for bad news, independent of standard sentiment measures.
Fanfare 3Asset Pricing & FactorsLynn Tao et al.PDF
New working papers in finance, economics and ML
30 papers
A survey of 47 LLM trading systems across six markets finds that reported margins shrink or vanish when rerun, with best evidence placing language models upstream of execution.
Fanfare 4LLMs & TextYebo Feng et al.
Language models analyzing analyst tone in M&A calls reveal that fear and anxiety about deals predict completion delays, failure, and long-run underperformance even when markets initially react favorably.
Fanfare 3LLMs & TextRuiyuan Chen et al.
A news reader trained only on pre-1995 text produces stable topics but earns 1.12 monthly Sharpe via ensemble with other readers, significant before multiple-testing adjustment.
Fanfare 2LLMs & TextShivesh Prakash et al.
Rejected factors earn 1.30 Sharpe in their home market versus 0.85 overall, yielding 3.8% six-factor alpha per year through venue reallocation constrained by signal-to-noise limits.
Fanfare 4Asset Pricing & FactorsHuannian Jin
The research reveals that pooling returns across market states creates an infinite-factor structure unconditionally, meaning infinitely many priced components and no finite principal-component basis for expected returns.
Fanfare 4Asset Pricing & FactorsMikhail Chernov et al.
Using partial least squares on 18 tail-risk proxies, the study constructs an index that strongly predicts the equity premium both in- and out-of-sample.
Fanfare 3Asset Pricing & FactorsYakun Liu and Lei Zhang
Stock prices and consumption share a long-run trend whose deviations mean-revert and predict market returns over one to two years, identifying consumption as a macroeconomic anchor for asset values.
Fanfare 3Asset Pricing & FactorsCarlo A. Favero et al.
The study finds no equity size premium in Brazil, but an unexplained 6.8 to 8.8 percentage point credit spread for small firms that rises sharply with corporate delinquency.
Fanfare 3Asset Pricing & FactorsRuben Xavier De Freitas
The CAPE ratio significantly predicts ten-year real stock returns across 1881–2023 even after rigorous statistical tests, with out-of-sample predictive power highly dependent on market regime.
Fanfare 3Asset Pricing & FactorsSanay Kotian
A frozen Gaussian VaR model, pre-registered and anchored in Bitcoin, passed coverage tests but showed clustering in 2020, 2022 and 2023, while exponentially weighted variance improved.
Fanfare 4Derivatives & VolatilityRishith Sharma
Single-stock leveraged ETFs incur excess financing costs averaging 6.2% annually through total-return swaps, fully passed to investors and substantially larger than disclosed expense ratios.
Fanfare 3Derivatives & VolatilityJinming Xue et al.
A blockchain-verified sweep of 18,630 cryptocurrency and FX trading variants found no strategies exceeding a Deflated Sharpe of 0.95, with median detectable Sharpe of 3.49 and collective underperformance versus random selection.
Fanfare 4Trading, Microstructure & ExecutionMobin Moharami Gargari
Trading-cost aware portfolio construction that reweights holdings by liquidity and trades only when forecasts exceed the bid-ask spread lifts net Sharpe from negative to positive on machine learning predictions.
Fanfare 3Trading, Microstructure & ExecutionDaniele Bianchi and Teng Jiao
Integer programming identifies arbitrage across linked prediction-market contracts where prices violate logical constraints, recovering 1.9 million relations and earning 2.2% guaranteed profit after spreads and fees.
Fanfare 3Trading, Microstructure & ExecutionDaojing Zhai and Hongjian Shi
Month-end index rebalancing that increases duration generates predictable demand and 1.19 basis points of monthly return concentrated on the rebalancing day in government bond futures.
Fanfare 3Trading, Microstructure & ExecutionAmar Soebhag et al.
Order-level latency variation at China's exchange shows that HFT raises short-horizon price noise and adverse-selection costs but slightly improves pricing efficiency.
Fanfare 3Trading, Microstructure & ExecutionLi Guo et al.
The study shows that dealer balance-sheet constraints, collateral convenience yields, and client outside options jointly determine repo haircuts, with leverage ratios reshaping market functioning and monetary policy transmission.
Fanfare 3Trading, Microstructure & ExecutionBenedikt Ballensiefen et al.
Exchange-rate hedging costs explain about 16 percent of remittance fees, especially in thin currency markets, and induce operators to adjust variable fees while raising exit risk.
Fanfare 3Trading, Microstructure & ExecutionLeonardo Flori and Nicola Limodio
A new calculus framework for rough volatility below the semimartingale regime enables bias-corrected estimation of leverage effects and roughness with central limit theorems, rejecting semimartingale assumptions for recent S&P data.
Fanfare 4Econometrics & ForecastingHaijian GUO et al.
Random forests and gradient boosting forecast financial crises across 147 years and 17 countries with 77 percent accuracy, using real-sector predictors as the dominant source.
Fanfare 3Econometrics & ForecastingEmile du Plessis and Ulrich Fritsche
A survey reviews machine learning methods for predicting stock returns in time series and cross-section, covering LASSO regularization, deep learning, genetic programming, and LLM applications.
Fanfare 3ML & AI MethodsDavid Rapach and Guofu Zhou
Transformer-based models capture pricing signals via self-attention that boost out-of-sample Sharpe ratios across one-, three-, six-, and twelve-month stock return forecasts, outperforming standard machine learning approaches.
Fanfare 3ML & AI MethodsTobias Cramer et al.
Encoding commodity futures curves as time-maturity surfaces and applying convolutional networks extracts carry information that predicts returns beyond standard factors and improves the mean-variance frontier.
Fanfare 3ML & AI MethodsYuecheng Jia et al.
Comparing predictive models across US and Indian equity markets shows that additional capacity helps when signal is strong but not when it is weak, with simplest models matching complex ones in high-noise regimes.
Fanfare 3ML & AI MethodsYash Vats
Machine learning forecasts of equity, bond, bill and housing returns from 1870-2020 across 16 countries improve Sharpe ratios substantially when trained on pooled cross-country data rather than country-specific samples.
Fanfare 3Portfolio & AllocationRay Friedenberger and Felix Reichenbach
An interpretable machine-learning framework linking regime identification to cross-asset portfolio construction earns 0.60 net Sharpe, not significantly different from a 60/40 benchmark.
Fanfare 2Portfolio & AllocationChalam Raja Ankam
The paper develops a Malliavin-informed policy-gradient estimator using stochastic control variates to reduce variance in continuous-time reinforcement learning for dynamic portfolio selection.
Fanfare 2Portfolio & AllocationAbdou Kaba
Cryptocurrencies sensitive to positive market jumps earn lower returns, a negative premium unexplained by coskewness and strongest when retail sentiment is high.
Fanfare 3Crypto & DeFiSuzanne S. Lee and Minho Wang
Elections in major democracies raise cryptocurrency correlation, reducing portfolio diversification, with the effect driven by shifting expectations about future regulatory environments.
Fanfare 3Crypto & DeFiYuan Tian and Yao Zheng
The study shows deposit inflows from quantitative easing at primary dealers predict subsequent lending and employment growth in exposed counties, revealing a concentrated channel for monetary transmission.
Fanfare 3Macro-Finance & RatesTuo Yang
Working papers curated by RePEc's NEP field reports
30 papers · 24 figures
A review of LLM applications finds that contextual interpretation improves some tasks but measurement error persists, and trading signals do not always generate profits.
Fanfare 4LLMs & TextZongwu Cai et al.PDF
FinBERT-based sentiment measures agree more closely with human raters than dictionary methods and better distinguish positive, neutral, and negative financial news articles.
Fanfare 3LLMs & TextStefano Borgioli et al.PDF
Large bitcoin trades reshape participant composition and trigger herding, raising volatility, while ethereum whales produce minimal market impact, revealing persistent information asymmetries.
Fanfare 3Crypto & DeFiKeith HazenPDF
The paper uses monetary history to identify five institutional foundations for scalable private money and assesses how contemporary stablecoins and the GENIUS Act address them.
Fanfare 3Crypto & DeFiMichael D. Bordo and Carolyn A. WilkinsPDF
The study proposes a multi-stage method combining neural networks and copulas to detect regime changes and finds cryptocurrency dependence with traditional assets increases sharply during crises.
Fanfare 2Crypto & DeFiVirginie Terraza and Aslı Boru İpekPDF
Nonbank financial institutions contract loan syndication during banking crises but expand bond underwriting, helping firms that maintain NBFI relationships weather downturns better.
Fanfare 3Risk, Credit & BankingBruno Albuquerque and Melih FiratPDF
ECB tightening passes nearly fully to mortgage rates but weakly to consumer credit, with younger borrowers facing stronger mortgage pass-through than older ones.
Fanfare 2Risk, Credit & BankingOlivier De Jonghe et al.PDF
A stochastic balance-sheet model quantifies how solvency and liquidity risks interact, finding that amplification effects from behavioral responses explain roughly one-third of default risk.
Fanfare 2Risk, Credit & BankingGiovanni Covi and Tihana ŠkrinjarićPDF
The analysis finds high public debt and growing non-bank footprint increase the risk of market dysfunction in government debt markets while making very favorable conditions more likely.
Fanfare 2Risk, Credit & BankingMathias Drehmann and Sonya ZhuPDF
Digital lending by non-banks reduces individual bank risk-taking and competition, with moderate systemic effects and stronger pressure on smaller banks.
Fanfare 2Risk, Credit & BankingEnrique Bátiz-Zuk et al.PDF
A mixture-of-experts approach combining ensemble models and logistic regression predicts developing-economy crises while decomposing risk into interpretable fiscal and financial components.
Fanfare 3ML & AI MethodsFrancisco Blasques et al.PDF
A calibrated model finds AI complements high-skill workers and equipment, with AI price declines expansionary but increased usage contractionary due to complementarity constraints.
Fanfare 3ML & AI MethodsAndré Luduvice and Roberto PinheiroPDF
Rising geopolitical risk fragments safe-asset demand: countries shift toward gold when sanction risk heightens, raising gold prices and Treasury yields as US convenience yields fall.
Fanfare 3Asset Pricing & FactorsKai Arvai et al.PDF
Analysis of IRS tax returns shows permanent income risk rose nearly 70% among the top 5% since the 1970s, depressing risk-free rates and widening wealth inequality.
Fanfare 3Asset Pricing & FactorsJ. Carter Braxton et al.PDF
A structural VAR shows climate policy uncertainty depresses equities and raises financial stress through credit channels, creating supply-driven downturns distinct from conventional uncertainty shocks.
Fanfare 3Asset Pricing & FactorsSerena Ionta Massimo GuidolinPDF
Stock and bond investors extract different value signals from earnings calls, with bondholders emphasizing operations and risk while equityholders focus on growth and innovation.
Fanfare 2Asset Pricing & FactorsJawad Addoum et al.PDF
The study compares monetary tightening in the 1980s and 2020s, finding that fiscal space—not exchange-rate flexibility alone—explains why emerging markets have avoided crisis this time.
Fanfare 3Macro-Finance & RatesDrishan Banerjee et al.PDF
A framework maps Middle East geopolitical risk to oil prices and inflation through scenario analysis updated in real time using market prices and expert assessments.
Fanfare 3Macro-Finance & RatesIrma Alonso-Alvarez et al.PDF
A model shows that as China's financial markets deepen, network effects allow the renminbi to compete with the dollar, triggering a discrete shift from cooperation to fragmentation.
Fanfare 3Macro-Finance & RatesZhengyang JiangPDF
Bank of England quantitative easing lowered corporate bond yields but did not increase real investment; instead firms used proceeds for buybacks and reduced bank borrowing.
Fanfare 2Macro-Finance & RatesMahmoud Fatouh et al.PDF
The research finds that reserve adequacy shapes how global financial shocks transmit to sovereign spreads and external debt in emerging markets, with over-accumulation dampening spread responses.
Fanfare 2Macro-Finance & RatesMateo Agustín FernándezPDF
A model shows asymmetric information about monetary policy can make exchange-rate risks uninsurable, generating foreign-exchange premia and opening a case for policy coordination.
Fanfare 2Macro-Finance & RatesMr. Suman S Basu and Pablo WinantPDF
Korea's carry-trade strategy strengthens public net worth near-term but cannot offset projected deficits or restore long-term fiscal sustainability without reform.
Fanfare 2Macro-Finance & RatesYiLi Chien et al.PDF
Switzerland and Japan face the highest risk of hitting the zero bound under current inflation targets; higher average inflation substantially reduces this constraint.
Fanfare 2Macro-Finance & RatesAntónio Afonso and Leonardo ZorziPDF
The research shows that firms in industries with larger network weights forecast inflation more accurately, and production networks endogenously increase nominal rigidity by about 30 percent.
Fanfare 3Econometrics & ForecastingRafael Gonçalves and Ina HajdiniPDF
Incorporating unstructured text from Federal Reserve transcripts into a DSGE model sharpens monetary policy inference and implies a flatter, less backward-looking Phillips curve.
Fanfare 3Econometrics & ForecastingSara Casella et al.PDF
Heterogeneous substitutability between corporate bonds implies security-level multipliers are much smaller than previous estimates, though portfolio multipliers remain substantial.
Fanfare 2Trading, Microstructure & ExecutionManav Chaudhary et al.PDF
Italian trade data reveals firms use invoicing currency bilaterally and by country to hedge country-specific risk, not just currency risk, reducing profit variance beyond net exposure.
Fanfare 2Trading, Microstructure & ExecutionEmanuele Brancati et al.PDF
Decomposing geopolitical risk into oil and non-oil components predicts US industry stock volatility better than aggregate measures across 49 sectors.
Fanfare 2Derivatives & VolatilityOnur Polat et al.PDF
Opening private markets to retail investors modestly improves their welfare while reducing firms' cost of capital and shrinking private equity alpha.
Fanfare 2Portfolio & AllocationLubos Pastor et al.PDF
