Papers for

financial technology developers

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Agent4Crypto improves cryptocurrency trading with adaptive multi-agent system

Agent4Crypto: A Dual-Reflective and Market-Feedback-Aligned LLM-Based Multi-Agent System for Cryptocurrency Trading

Abstract: Cryptocurrency markets provide a natural testbed for autonomous AI agents because they are digital-native, high-frequency, and machine-speed environments, yet existing methods often struggle with non-stationary market regimes, opaque decisions, and fixed aggregation rules. In this paper, we propose Agent4Crypto, a dual-reflective and market-feedback-aligned multi-agent system. The framework coordinates specialized agents for data collection, analysis, risk assessment, reflection, and trading, synthesizing heterogeneous signals from prices, indicators, news, and charts in an interpretable decision pipeline. Our central design is a trader-centric decision mechanism: three style-specific trader agents, Aggressive, Neutral, and Conservative, generate candidate actions and natural-language rationales from the same market context, while their influence is learned rather than manually fixed. A PPO-based gating policy dynamically allocates trader weights through a dual-reflective feedback process. One reflection stream evaluates post-trade decision quality from realized portfolio outcomes and distills reusable lessons into a Financial ReasoningBank for retrieval in subsequent trader prompts, while the other estimates trader-level counterfactual contribution, indicating which style has recently aligned better with market movements. The PPO state combines Counterfactual Contribution Memory with market features, previous trader weights, and current trader actions, so trader influence adapts across bull, sideways, and bear regimes. Experiments on BTC, ETH, and XRP show that Agent4Crypto improves total return and Sharpe ratio over traditional strategies, machine learning and reinforcement learning baselines, and existing LLM-based trading systems, demonstrating the value of interpretable trader competition and market-feedback-driven adaptation in volatile cryptocurrency markets.

Mon 9 NovMultiagent SystemsComputation and Language
The gist
Cryptocurrency markets are fast and complicated, making it hard for automated systems to trade effectively. The authors developed Agent4Crypto, a system that uses multiple specialized agents with different trading styles to analyze various market data and trade accordingly. It learns how much influence each style should have based on past trading success and market conditions. The system also reflects on its own past decisions to improve future trades. Tests on several cryptocurrencies show that Agent4Crypto can make better trading decisions compared to previous methods.
Open → 2611.12269v1

Policy learning creates fast counterfactual explanations for time series

Scalable Counterfactual Explanations for Time Series via Policy Learning

Abstract: Generating counterfactual (CF) explanations for time series is challenging due to complex temporal dependencies and the computational inefficiency of traditional instance-specific optimization. We propose a policy learning framework for counterfactual explanation in time series, trained as a one-step contextual-bandit policy-learning problem. By training an actor-critic agent in the latent space of a pre-trained autoencoder, our method learns a transferable policy that captures global structural regularities. Unlike existing methods that rely on iterative optimization for each new input, the learned policy enables direct generation of valid and plausible counterfactuals through a single forward pass, avoiding per-instance optimization at inference time. Experiments on 13 UCR benchmark datasets show that our framework achieves high validity and strong plausibility across multiple anomaly detection metrics. Overall, by decoupling counterfactual generation from per-instance optimization, the proposed method provides an efficient and scalable alternative to existing approaches for time series counterfactual explanations.

Mon 9 NovMachine LearningArtificial Intelligence
The gist
Understanding why time series predictions change can be tricky because time sequences have complex patterns. This paper presents a new method that learns to produce explanations quickly by training a policy that works for many examples, instead of figuring things out from scratch each time. The method uses an autoencoder to simplify the data and a trained agent to generate explanations in one step. This approach was tested on multiple standard datasets and found to produce valid and plausible explanations efficiently.
Open → 2611.11704v1

Trading system using language models beats S&P 500 in 2026

Inside the Trading Loop: Markets, Models, and Moves with LLM Agents

Abstract: We built a hybrid multi-agent trading system that combines large language model (LLM) analysts with momentum-based portfolio construction and rule-based execution. Company-news and macroeconomic agents return structured reports to a coordinator that sets stock weights and cash reserves. We evaluate the portfolio engine using its broker-recorded paper account, which operated with deterministic fallbacks, and compare LLM-enabled configurations through retrospective replays. From March 11 to August 28, 2026, the account returned 34.5%, compared with 14.4% for the S&P 500 fund SPY over the same dates. This raw return advantage came with a larger maximum drawdown. Reconstructed policies also finish above matched SPY and QQQ references when assessments are used in the same session or deferred to the next. Design comparisons examine cash authority, selective invocation, and response contracts. A constant-cash control gives a return estimate close to stance-based sizing, and shorter report contracts improve observed parsing but can change recommendations.

Sun 8 NovArtificial Intelligence
The gist
This paper studies a trading system that uses large language models (LLMs) to analyze company news and economic data, combined with other methods to decide which stocks to buy or sell. The system’s portfolio delivered a higher return than a popular stock market fund between March and August 2026, though it also had bigger losses at times. The authors explored how different design choices affect performance, like when to ask the LLM for advice and how to handle cash reserves. Their approach showcases how AI can assist but not fully replace traditional trading rules.
Open → 2611.11657v1

Tool-using AI agents get safer decision checks against prompt attacks

SPILLWAY: Separating Target Admission from Sanitization to Defend Tool-Using Agents against Prompt Injection

Abstract: Tool-using agents act on email, files, and payments, so an instruction hidden in a web page or a shared document can make an agent send data or money to an attacker. Many runtime defenses put a language model on the decision that lets an action through, so the decision depends on a model that may read the attacker's text and adds model calls to every step. Provenance defenses keep the model off the final check but need a model to plan or route each task first. Where an action goes can instead be read from the tools' typed returns: an email's sender or a payment's account arrives in a typed field, while injected instructions arrive as free text. SPILLWAY turns this observation into a deterministic floor with no per-task plan. The check admits a call when each target value that appears in tool output but not in the user's task was returned by a read-only tool in a typed field or is covered by a fixed delegation rule. A deterministic sanitizer removes instruction-like regions its fixed rules match before the model reads them and retains the removed spans as evidence, and a local judge may restore a refused call, never one whose target lies in the removed text. Across three backbones, and seven task suites on one of them, no standard templated injection redirects an action to an attacker's target under SPILLWAY, at the highest observed mean utility and composite scores (NRP, ERS) of the defenses we run under that attack. A forged chat template also produces a raw zero, and mean in-scope attack success under a released reinforcement-learning attacker is .007. SPILLWAY reaches this floor with no defense-triggered backbone re-run and almost no added model call. As personal AI agents act on users' email, files, and payments, SPILLWAY gives them an auditable refusal floor on where their actions go that keeps them useful and holds across backbones.

Wed 4 NovCryptography and SecurityArtificial Intelligence
The gist
People use AI agents to help manage emails, files, and payments, but bad actors can trick these agents by hiding sneaky instructions inside normal-looking text. The authors created SPILLWAY, a method that separates deciding where an action should go from checking for harmful instructions. SPILLWAY uses clear rules to catch suspicious instructions before the AI reads them and only allows actions when the target information is safe and comes from trusted sources. This approach reduces the chance that attackers can redirect payments or emails and keeps the AI agents helpful without slowing them down.
Open → 2611.07222v1

Entropy regularization improves investment and spending decisions under market changes

Entropy-Regularized Reinforcement Learning for the Regime-Switching Consumption-Investment Problem

Abstract: This paper establishes an entropy-regularized reinforcement learning (ERRL) framework for the continuous-time consumption-investment problem under a regime-switching market model with general utility functions. By incorporating randomized exploratory policies and Shannon entropy regularization into the stochastic optimal control formulation, we derive closed-form representations for the optimal exploratory consumption and investment strategies. Specifically, the optimal investment strategy follows a multivariate Gaussian distribution, whereas the optimal consumption strategy takes the form of a Gibbs measure---which specializes to a Gamma distribution under logarithmic utility and a non-standard exponential distribution under power utility. Crucially, our theoretical analysis reveals that the classical deterministic optimal decisions correspond to the mode---rather than the mean---of these exploratory distributions. Furthermore, we establish the convergence of the exploratory value function to its classical unregularized counterpart as the temperature parameter vanishes. For algorithmic design, we develop a continuous-time q-learning framework grounded in a novel continuous-time q-function, prove the policy improvement theorem, and establish martingale characterizations for both on-policy and off-policy learning. We find that parameterizing the actor network based on theoretical distribution families can help accelerate convergence and enhance numerical stability. Furthermore, numerical experiments across regime-switching environments provide empirical support for our theoretical findings, demonstrating stable loss convergence as well as the monotonicity and concavity of the learned value functions.

Sat 31 OctMachine Learning
The gist
Planning how to spend and invest money during changing market conditions is hard. This paper introduces a new way to make better, more flexible choices by adding a kind of randomness (called entropy regularization) to the decision process. The researchers found exact formulas for how to mix and match investment and spending strategies when markets switch between different states. Their approach also smoothly matches traditional investment methods as the randomness goes away. They tested their method in simulated markets and saw it learns steadily and reliably.
Open → 2611.01046v1

AStockArena audits real-time financial trading decisions beyond returns

AStockArena: Beyond Returns, Can We Entrust Real Capital to the Decisions of Financial Agents?

Abstract: Large language models (LLMs) are increasingly used in dynamic financial trading, yet real-time trading benchmarks mainly evaluate agents by cumulative returns and risk-adjusted performance. Such outcome metrics cannot show whether trades are faithfully reported, supported by available evidence, and consistent with account states. We introduce AStockArena, a real-time decision-auditing benchmark for continuous trading in China's A-share market. Under shared market conditions and execution constraints, we evaluate 13 open- and closed-source models over 3,198 sequential decisions. AStockArena reconstructsthe actions actually posted to accounts by aligning reports, trade requests, execution responses, and account states, and audits them with eight checks spanning execution traceability (O), evidence sufficiency (RK), and financial meta-awareness (B), combining record verification with consensus expert annotation. Of 1,070 trading points, 72.99% exhibit at least one audit gap; model-level rates range from52.2% to 91.2%, with evidence gaps most prevalent. Historical trade-filtering replay further shows thatportfolio outcomes are sensitive to these flagged trades. Decision-process auditing thus offers a verifiable complement to return-based evaluation of financial agents.

Sat 31 OctComputation and LanguageArtificial Intelligence
The gist
Financial trading AIs are often judged only by how much money they make or how risky their trades are. The authors point out that these measures miss whether the trades were actually done as reported and if there is enough evidence supporting them. They built AStockArena, a tool that checks over 3,000 trading decisions from AI models on China’s stock market to see if trades match records and rules. They found many issues with the reported trades, showing that just looking at profits isn’t enough to trust these AI traders. This auditing helps verify if AI financial agents make reliable decisions.
Open → 2611.00515v1

Financial assistants often answer wrong questions with right numbers

Right Numbers, Wrong Questions: Evaluating LLMs on Underspecified Financial Queries

Abstract: Financial assistants answer questions that sound factual but depend on what the user leaves unsaid: what a data figure measures, which product variant is meant, and the investor's situation and the rules that apply. An assistant that ignores these can still answer with a real number from its data, one that answers a different question. We call this frame collapse. We study it in Indian mutual funds, where product variants and tax rules are formally defined, so the correct answer can be computed for every reading. One study tests each challenge. Given an ambiguously labelled figure, models used it, often under the user's label. Given every alternative labelled, they listed them all, but the layout decided which came first. Missing a fact about the investor, weaker agents assumed a default or an outdated tax rule; requiring the fact did not guarantee its use. Because such answers match their data, checks against the data pass them. An AI judge caught mislabelled figures only when told what each field means, and still passed answers that silently chose another reading. We show how these failures arise, not how often, and propose controls for financial assistants.

Wed 28 OctArtificial IntelligenceComputation and Language
The gist
Financial assistants sometimes give a number that looks correct but actually answers a different question than the user intended. The authors show this happens when questions leave out important details, like which financial product or tax rule applies. They studied Indian mutual funds where details are well defined, finding that models often pick default or outdated assumptions without realizing it. This means even answers that match data can be wrong for the user’s situation. The authors offer ways to detect and control these mistakes.
Open → 2610.38631v1

Financial advice AI shows bias by gender caste and location in India

The Indian Financial Automated Advice Bias (IFAAB) Dataset : Auditing Socioeconomic and Demographic Disparities in LLM-Generated Financial Advice in India

Abstract: Large language models are increasingly used for financial guidance, yet little is known about whether their advice changes across social identities in Global South contexts. We introduce the Indian Financial Automated Advice Bias (IFAAB) benchmark, a counterfactual audit dataset of 31,734 GPT-4o-generated financial-advice responses for synthetic Indian personas varying by gender, caste, education, employment, residence, age, and district. Holding the underlying request fixed, we evaluate disparities in LLM-implied wealth, income, savings, projected gains, textual similarity, advice topics, and qualitative bias judgments. In the gender-only audit, male-labelled profiles receive 60.1\% higher initial total resources and 46.8\% higher post-advice total resources than comparable female-labelled profiles; in the caste-augmented audit, the post-advice total-resource gap remains 49.4\%. Adding district information reduces the aggregate post-advice total-resource gender gap to 7.9\%, revealing strong geographic conditioning. Standalone caste comparisons also show systematic stratification: relative to General-category profiles, OBC, SC, and ST profiles receive significantly lower post-advice total resources, respectively. Textual audits show low semantic similarity between paired gendered responses, with an average TF--IDF cosine similarity score of 0.44. Qualitative analysis shows that women are more often steered toward cautious saving, community support, and basic financial literacy, while men receive more growth-oriented advice involving investment, entrepreneurship, and asset accumulation. IFAAB provides a reusable benchmark for evaluating fairness, suitability, and consumer-protection risks in LLM-mediated financial advice, and highlights the need for locally grounded, intersectional evaluation of AI systems in high-stakes economic domains.

Wed 28 OctComputers and SocietyArtificial Intelligence
The gist
Large language models that give financial advice in India sometimes treat people differently based on gender, caste, and where they live. The authors created a big dataset to test how AI advice changes for fake Indian profiles with different social traits. They found that men and people from higher castes often get suggestions that lead to more money, while women and lower caste individuals get more cautious advice. The study shows that AI can reflect social biases and that testing for fairness in local contexts is important.
Open → 2610.38239v1

Concordat enforces safe rules for ai agents handling sensitive tasks

Concordat: Typed Sessions for LLM Agents

Abstract: Agents built on large language models are increasingly allowed to take consequential actions: cancelling bookings, issuing refunds, changing accounts. The rules that govern those actions, such as "obtain the customer's explicit confirmation before changing anything", are usually stated in the prompt and followed only as reliably as the model chooses to follow them. We present Concordat, a Python library that treats an agent as an object whose interface is a finite-state protocol, in the spirit of typestate and session types, and enforces that protocol on the model at run time. On every turn the model is offered only the tools that are legal in the current protocol state, calls outside that set are rejected before they have side effects, decisions reserved for humans can only be requested, and token and tool-call budgets are charged and split linearly across handoffs. Protocols are checked when they are defined, and a subclass of an agent is rejected if it could bypass its parent's protocol, a behavioural form of Liskov substitution. We evaluate a Concordat agent against the standard agent of tau-bench on its 50 airline tasks, adding simulated customers who pressure the agent to skip confirmation. The standard agent made changes without the required "yes" in 13 of 50 tasks with gpt-4.1 and in 29 of 128 task-runs with Claude Haiku 5.5; the Concordat agent made none, with no significant difference in task success. tau-bench's own reward did not detect most of these violations. Concordat is open source.

Tue 27 OctArtificial IntelligenceMultiagent Systems
The gist
Sometimes AI agents do things like cancel bookings or change accounts, which need careful rules to keep people safe. Usually, these rules are just instructions the AI might ignore. The authors created Concordat, a tool that makes sure AI agents follow strict rules step-by-step and can only do allowed actions. It stops the AI from making changes without proper permission, catching mistakes before they happen. Tests showed it prevented rule-breaking better than older methods without making the AI less successful.
Open → 2610.37461v1

Ai financial advice shifts with changing self reported risk levels

Same Person, Different Advice: Auditing LLM Financial Advice with Repeated Risk-Tolerance Responses

Abstract: AI financial advisers set a client's stock allocation largely from one stated-risk question, yet people do not answer it the same way twice: in a national US panel, 41% of households chose a different answer two years later, and in a laboratory retest a third of people moved two or more points on a 0-10 scale after six months. Seven large language models from five providers gave one-turn allocation advice for 800 real household profiles in which only the risk answer changed. The recommended stock share moved by 10 to 16 percentage points on average, and by 24 to 35 when the answer changed, against 0.6 to 4.2 points for the same prompt asked twice; in the laboratory retest the rating alone moved advice by 11 to 18 points. Without any note, the models respond to the answer about as strongly as when told that it is very reliable (less so for the two Claude models). If that instructed response is how each model maps attitude to advice, discounting the answer for its measured reliability, as classical test theory prescribes, would cut the swings by a factor of 2.4 to 3.3. The swing is largest for clients under 40. A note that the answer is noisy steadies some models, but shifts their weight to the client's age and pulls cautious clients towards a generic middle allocation: clients unwilling to take any risk are advised 22% stocks instead of 4%. Three extra money questions reduce the swing for some models, with no statistically detectable change relative to how much advice differs across people. In a scripted test, one clarifying question removes nearly all of the swing from an answer the client then corrects (33.5 points to 1.9 on average) while keeping the response to a change the client confirms. We argue that same-person retest audits should be a standard check for AI advisers, and that what to do with an unreliable answer is a question of client autonomy as well as of statistics.

Fri 16 OctComputation and LanguageComputers and Society
The gist
People often give different answers about how much financial risk they want to take when asked the same question at different times. The authors tested multiple large language models that give stock advice based on one risk answer and saw that the recommended stock amounts changed a lot when the risk answer changed. They found that models treat risk answers as very reliable even though people’s self-reports can be inconsistent. Adding extra questions or clarifying responses helps stabilize the advice. The authors suggest checking AI advisers by repeating risk questions to better handle unreliable answers.
Open → 2610.23060v1

Strategic federated learning improves defense against covert attacks

StratFL: Countering Covert Model Poisoning Attacks with Strategic Federated Learning

Abstract: Advanced persistent threats (APTs) are a major concern in federated learning (FL) deployments in critical sectors such as healthcare, and finance. Traditional defense mechanisms fail to contain APTs due to their intermittent attacking behavior. This paper addresses a specific type of APT called covert model poisoning (CMP) where the attacker intermittently injects poisoned model updates in a covert manner. The interaction between the FL server and a CMP attacker is modeled as an imperfect repeated game where both players cannot observe each other's actions. A novel adversarial search algorithm called StratFL is designed alongside novel belief update algorithms at both the server and adversary. Experiments are designed using MNIST and CIFAR-10 datasets to evaluate and compare StratFL's performance against state-of-the-art methods. Empirical convergence of belief update algorithms is demonstrated across various security vignettes. At the same time, StratFL is found to exhibit an improved accuracy of 10% compared to state-of-the-art methods.

Thu 15 OctCryptography and SecurityArtificial IntelligenceComputer Science and Game Theory
The gist
Federated learning allows many users to train AI models together without sharing their data, but attackers can secretly corrupt the shared model updates, making it less accurate. This paper studies a sneaky kind of attack where bad updates happen only sometimes, which traditional defenses miss. The authors model the interaction between the defender and attacker as a game where neither side fully sees the other's moves, and create a new method called StratFL that strategically detects and counters these hidden attacks. Tests show StratFL achieves about 10% better accuracy than other leading methods on common image datasets.
Open → 2610.22626v1

Market making improves with reinforcement learning on self-exciting order flows

Reinforcement Learning with Hawkes-Structured Networks for Market Making

Abstract: Market making under realistic order-flow dynamics is challenging due to the clustered, self-exciting nature of order arrivals, which are naturally modeled by non-Markovian Hawkes processes. We propose a model-free reinforcement learning (RL) framework for market making with general Hawkes order-flow dynamics. We first establish non-asymptotic performance bounds that quantify the loss incurred by approximating the infinite-dimensional Hawkes state using a finite collection of exponentially weighted features. Guided by this analysis, we introduce the Hawkes-Structured Network (HSN), a novel neural architecture that encodes historical order flow through a learnable mixture-of-exponentials filter. HSN is trained end-to-end using model-free RL, jointly optimizing the feature representation and quoting policy directly from reward feedback. Experiments on single- and multi-asset market-making problems with mixture-of-exponentials, power-law, and regime-switching kernels demonstrate that HSN outperforms general-purpose neural architectures, achieving higher trading rewards while implicitly recovering the underlying order-flow intensities.

Mon 12 OctMachine Learning
The gist
Making markets—buying and selling financial assets—can be tricky because orders to trade tend to happen in bursts and are influenced by past activity. This paper shows how to use reinforcement learning, a kind of AI that learns by trial and error, combined with a special way to remember past order patterns, to better handle these bursts. The researchers created a neural network that directly learns both how to recognize order flow patterns and how to trade from rewards. Their method works better than typical neural networks in various trading scenarios.
Open → 2610.16524v1

Credit alignment method improves cross-tokenizer model learning scores

CAST: Credit Alignment for Cross-Tokenizer On-Policy Distillation

Abstract: Cross-tokenizer on-policy distillation (OPD) requires aligning teacher and student sequences before their scores can be compared. Decoded-text alignment provides such correspondence at the chunk level, but leaves a distinct problem unresolved: credit alignment. A chunk-level signal specifies how much a span should change, but not how that signal should be distributed across student tokens, while tokens without teacher correspondence remain unconstrained. We propose CAST (Chunk-aware credit Assignment with Selective Token Anchoring), a shift-or-stay framework that addresses both issues. For aligned chunks, CAST redistributes each chunk's advantage across student tokens using predictive uncertainty and local teacher-signal smoothness, emphasizing learning opportunities supported by locally consistent teacher feedback while preserving the chunk's total advantage. The resulting weights exactly solve a KL-regularized credit-allocation problem. For unaligned positions, CAST selectively anchors the student to its initial policy through a KL penalty, limiting drift where teacher supervision is unavailable. Because credit assignment is separated from chunk construction, CAST also applies to punctuation-defined or fixed-length chunks under shared tokenization. Across six mathematics, science, and code benchmarks, CAST improves the average over the strongest evaluated baseline by $2.45$ and $1.32$ points for cross-tokenizer transfer from Qwen3-8B to Mistral-7B and Llama-3.1-8B, respectively, and by $2.10$ points for shared-tokenizer transfer to Qwen3-1.7B.

Fri 9 OctMachine LearningComputation and Language
The gist
When computers learn from one model to improve another using different ways of breaking up sentences into tokens, it’s hard to tell exactly how well the learning is going token by token. The authors propose a new method called CAST that carefully shares credit for model improvements across token groups and keeps parts of the student model stable where the teacher gives no guidance. This helps the student model learn more accurately from the teacher, even when they use different tokenizers. Their tests show that CAST improves performance across various math, science, and coding tasks.
Open → 2610.13257v1

Market maker algorithms learn better bid ask pricing with less error

Optimal Regret for Online Market Making with Limit Order Book

Abstract: We study online learning in market making, where, at each round, a market maker posts bid and ask prices before observing the market price and the private valuation of an incoming trader. In this setting, Maran et al. 2026 introduce a feedback model motivated by limit order books, in which the trader's valuation is revealed only if no transaction occurs. Assuming that trader valuations are drawn i.i.d. from an unknown distribution while market prices are chosen adversarially, they establish an expected regret bound of $\widetilde{\mathcal{O}}(T^{2/3})$. In this work, we improve upon this guarantee by establishing a high-probability regret bound of $\widetilde{\mathcal{O}}(\sqrt{T})$. As a warm-up, we first consider the full-feedback setting. We introduce a discretization of the bid-ask space based on two coupled grids and combine it with Hedge to achieve the desired regret rate. Building on these ideas, we then address the substantially weaker feedback induced by a limit order book and develop an algorithm that achieves the same guarantee. Finally, we investigate the limits of learnability in fully adversarial environments, where the valuations may vary arbitrarily as well. Perhaps surprisingly, we show that when both market prices and trader valuations are chosen adversarially, sublinear regret is impossible even under full feedback, thereby motivating our stochastic assumption on the valuations.

Wed 7 OctComputer Science and Game TheoryMachine Learning
The gist
This work studies how a market maker can learn to set good bid and ask prices over time when trading with many buyers and sellers. The authors improve previous methods by designing algorithms that make fewer mistakes, even when the market price changes unpredictably. They also show that if the buyers’ valuation changes completely randomly in a worst-case way, it’s impossible for any algorithm to do well. This helps explain when learning to set prices can and cannot succeed.
Open → 2610.09691v1

Language models rely on identity when financial details are missing

Thin Evidence, Thick Priors: How Language Models Substitute Identity for Missing Financial Facts

Abstract: People increasingly ask large language models what to do with their money, yet seldom describe their finances in full. This paper asks what a model does with the gap. Holding finances fixed and changing only who the investor is said to be, we grade the financial evidence in the prompt from eight facts to none and measure how far the recommended equity allocation moves. Across 96,600 prompts to Llama-3.1-8B-Instruct, built from 100 financial profiles, 138 personas and seven disclosure conditions, the average gap between two personas with identical finances rises from 4.78 percentage points at full disclosure to 10.34 points with no financial facts. A two-way cluster bootstrap counting duplicated prompts once places the ratio at 2.16 (95% interval 1.69 to 2.79), and the rise is already 1.69-fold with a single fact left. Identity explains 5% of within-profile variation in advice at full disclosure and 96% with no disclosure. Household size is the only attribute whose influence grows reliably as evidence is withdrawn. Once standard errors are clustered on the persona, the unit to which identity was assigned, most attribute-specific interactions reported in the conference version lose significance, and gender instead appears as a small standing gap that full disclosure does not close. Stating risk appetite alone brings the swing into the range seen with two to seven generic facts. With no facts, the model's one-line rationale cites incomes, debts and savings it was never told, and these invented finances turn adverse more often for larger households. Inside the network, gender is linearly decodable at every layer, and ablating the gender direction at five layers leaves the aggregate identity swing unchanged. Advisory systems built on such models should be audited at the disclosure levels users actually reach, and judged across the whole identity space rather than one attribute at a time.

Tue 6 OctArtificial Intelligence
The gist
When people ask language models for financial advice but don’t share many details about their money, the models fill in gaps by guessing based on who the person is. The authors tested this by keeping financial facts the same but changing who the investor was described as, then seeing how the advice changed. They found the model’s recommendations shifted more when less financial information was given, with identity traits like household size influencing advice more as fewer facts were shared. The model even invented financial details it wasn’t told about, showing these tools depend heavily on assumptions about identity when facts are missing.
Open → 2610.07798v1

Cost effective numerical question answering over complex tables with llms

Cost-Effective Numerical QA over Semi-Structured Table: Structuring, Resolution, Planning

Abstract: Semi-structured tables encode rich semantic information through diverse layout elements, such as hierarchical row and column headers. Answering numerical questions over such data is challenging because it requires a joint effort of accurate structural understanding of tables, question uncertainty resolution, and complex query intents interpretation. Existing methods are rarely effective in handling the above multifaceted challenges in a holistic manner; moreover, they rely on LLMs without considering financial implications. We propose SemiBonsai, a cost-effective, LLM-powered framework featuring: (1) a Multiway Layered Table Structurer that converts a table into a multiway layered tree organized by structural elements; (2) a Context-Aware Uncertainty Resolver that grounds underspecified phrases to context-coherent structural elements; (3) a Plan-Guided Reasoner that decomposes complex query intents into a query plan for reasoning; (4) a Budget-Aware Bandit Router that learns per-instance LLM utilities and optimizes LLM selection under fixed monetary constraints. Experiments on three benchmarks show that SemiBonsai improves both QA effectiveness and cost-effectiveness. Notably, the Table Structurer improves answer accuracy by 46% over alternative tree construction methods, and Router achieves a 5% average relative gain in answer accuracy compared with existing routing methods across budgets.

Tue 6 OctDatabases
The gist
Answering number questions from tables is hard when the tables have complicated layouts. The authors created a system called SemiBonsai that breaks down tables into clear parts, figures out unclear question words using context, and plans how to answer complex questions step-by-step. Their system also smartly chooses which AI to use for each question to save money. Tests show it answers questions more accurately and cheaply than older methods.
Open → 2610.07749v1

Skill packages improve financial AI agents for portfolio and risk tasks

Knowledge or Calculator? Decomposing the Skill Premium in Verifiable Financial Agent Workflows

Abstract: Financial AI agents must do more than retrieve facts: investment workflows require correct quantitative execution, reliable use of procedural resources, and auditable structured outputs. We introduce FinSkillBench, an evaluation suite of 2,603 point in time episodes across 12 subtasks in portfolio construction, risk management, and fundamental analysis, with hidden regenerable ground truth and task specific deterministic verifiers. Executing 17,820 episodes across 9 models and 3 resource conditions, the paired analysis across 8 models shows that curated skill packages raise mean scores by +16.2 points (0.366 to 0.528), whereas skills generated within a single episode add only +0.5 points while consuming more tokens and turns. We then decompose the curated premium by granting human authored procedural documents and executable domain tools separately: documents alone add +5.6 points, tools alone add +19.5 points, and their combination is subadditive. The premium is strongly workflow dependent: executable tools dominate numerically intensive workflows, documentation matters more when procedural or output schema guidance is the bottleneck, and interpretive tasks benefit from both. The effects are sign stable across 10 scoring variants and cluster bootstrap analyses, and an independently implemented second harness reproduces the directional pattern while showing that effect magnitudes depend on how tools and data are exposed. Overall, a measured "skill premium" is a property of the full model, resource, and harness system rather than of the underlying model alone.

Fri 2 OctArtificial Intelligence
The gist
FinSkillBench is a new test suite that checks how well financial AI agents perform tasks involving investment and risk management. The authors found that using pre-made skill sets helps the agents do better than creating skills on the fly. They separated the effects of human-written instructions and automated tools, showing tools help more with number-heavy tasks while instructions help where procedures matter. This means the whole system of model, tools, and instructions—rather than just the AI model—affects performance.
Open → 2610.03564v1

Compound AI systems improve reliability with failure patterns and fixes

Compound AI System Reliability: A Failure Taxonomy and Resilience Pattern Catalog from 150 Production Incidents

Abstract: Deploying compound AI systems reliably and safely requires understanding failure modes that emerge at component boundaries, not within individual models. Cascading errors propagate across component boundaries, silent quality degradation evades standard monitoring, and coordination failures yield incorrect collective behavior from individually correct parts. We analyze 150 production incident reports from open-source compound AI projects and anonymized enterprise deployments to construct a taxonomy of 23 failure modes organized into five categories: retrieval failures, generation failures, tool failures, orchestration failures, and integration failures. For each category, we propose resilience patterns with measured effectiveness from controlled fault injection experiments. Circuit breakers reduce cascade propagation by 89%, output quality gates catch 73% of silent degradation before user impact, and component isolation reduces blast radius by 64%. Systems implementing three or more resilience patterns from our catalog reduce mean-time-to-recovery (MTTR) by 71% compared to unstructured monitoring baselines. We release the incident taxonomy and pattern catalog as a practitioner resource.

Thu 1 OctSoftware EngineeringArtificial IntelligenceDistributed, Parallel, and Cluster Computing
The gist
Big AI systems made of many parts sometimes fail in unexpected ways when these parts interact. The authors studied 150 real failures from different AI projects to figure out common problems and grouped them into five types. They also tested ways to make these systems stronger, like using circuit breakers to stop errors from spreading or quality checks to catch hidden problems early. Systems using several of these fixes recovered much faster after problems than those without them. They share their findings to help others build safer AI systems.
Open → 2610.02503v1

Multimodal models improve financial forecasting from earnings calls

MM-FinEval: A Multi-Task Multimodal Benchmark for Real-World Financial Forecasting

Abstract: Financial forecasting from earnings conference calls requires models to reason over complex corporate disclosures, market expectations, and subtle communication signals. However, existing financial benchmarks are often limited to unimodal inputs or single-task settings, making it difficult to evaluate whether multimodal large language models (LLMs) can support real-world financial analysis. In this paper, we introduce MM-FinEval, a novel benchmark designed to evaluate multimodal LLMs across multiple financial tasks. MM-FinEval spans a diverse timeline from 2019 to 2022. The entire proposed dataset contains 2,045 S\&P 500 conference earning calls as inputs and 12 financial task labels as outputs. Each input contains three modalities: a word-to-word text transcript of the earning call, the corresponding presentation slides used during the call, and the entire audio recording. To establish a rigorous evaluation framework, we analyze 19 baseline models across three distinct model categories: Image-Text, Audio-Text, and Any-to-Any configurations. We observe that small-size Any-to-Any models processing all three modalities achieve strong performance, even when compared against larger proprietary models restricted to two-modality inputs. This indicates that our tri-modal dataset design introduces useful, non-redundant information. These results validate that text, audio, and visual data serve as important, complementary signals that mimic the decision-making process of expert human analysts.

Tue 29 SeptMachine LearningArtificial Intelligence
The gist
Financial forecasting often relies on understanding complex information from company earnings calls, which include spoken words, presentation slides, and audio cues. The authors created a new benchmark called MM-FinEval that tests how well AI models can analyze all three types of information together for multiple financial prediction tasks. They found that smaller models using all three data types perform better than some larger models using fewer types. This shows that combining text, audio, and visuals adds helpful information that mimics human analysts.
Open → 2609.38523v1

Finance agents get automatic expert-guided evaluation rubrics

FinAutoRubric: Expert-Guided Automatic Rubric Generation for Evaluating Financial Research Agents

Abstract: Evaluating finance research agents requires rubrics that reflect expert standards and fix the values correct as of an information cutoff. Expert-reviewed finance benchmarks rely on fixed, per-item rubrics, which are costly to extend and cannot encode each institution's own standard. In FinAutoRubric, experts specify reusable evaluation guidance, while agents and code carry out query-specific rubric generation, review, and validation. This expert guidance governs every agent, as prompts and as rules that code enforces, and a Task Bank of reusable criteria carries it across tasks. In long-horizon loops that follow the expert guidance, a writer agent researches every expected value and a reviewer agent verifies it, and failures escalate to a human. On three expert-authored finance benchmarks, its rubrics track expert scoring as closely as the strongest evaluated generator while stating the expert rubric's expected value for more criteria, their scores agree with human grading, and in-house analysts prefer them in a blind review. The released 100-query FinAutoRubric Benchmark, built from in-house analysts' key questions across 78 tasks and eight asset classes, shows that rubrics from an earlier model generation still leave headroom for a later one.

Mon 28 SeptArtificial Intelligence
The gist
Evaluating financial research agents needs careful and expert-based rules to judge their answers accurately. The authors created FinAutoRubric, a system where experts give general guidelines that computer agents use to build detailed, specific rules for checking answers. These rules help write and review financial questions automatically, with human checks for problems. The system was tested on real finance questions and matched expert scores well, even earning preference from analysts in blind tests.
Open → 2609.35744v1

Ceo arena tests ceo agents competing in simulated markets

CEO Arena: Evaluating Long-Horizon Multi-Agent Decision-Making in Competitive Markets

Abstract: Long-horizon competition tests agents' ability to coordinate business decisions under uncertainty and adapt to changing rival strategies. We introduce CEO Arena, a benchmark that uses matched replacement evaluation to assess operating returns alongside an agent's effects on rivals and the market. Each CEO agent is compared with a reference policy in the same company under the same economic seed, holding other agents' identities and assignments fixed while all agents adapt. In a shared eight-company market spanning 500 simulated days, CEOs make sequential decisions on pricing, procurement, marketing, research and development, and service using private company information and noisy market signals, under resource constraints and delayed feedback. We evaluate eight LLM-based CEO agents in 27 main runs and 26 robustness runs. In the main evaluation, most agents have negative mean returns, and private gains can accompany market losses. Robustness analyses suggest that aggregate patterns extend beyond the original rule-based baseline; four of the 56 directed pairs show relatively stable effects. Memory, action, and accounting traces suggest demand capture and rivals' pricing and spending responses as possible explanations. CEO Arena provides a controlled testbed for studying long-horizon agent competition, strategic interaction, and market externalities.

Mon 28 SeptMultiagent Systems
The gist
Running a business involves making many decisions over time while dealing with uncertain markets and competitors trying to outsmart each other. The authors created CEO Arena, a simulation where AI agents act as CEOs managing companies, deciding prices, marketing, and development with limited info. They tested multiple AI CEOs to see how well they perform alone and how their actions affect competitors and the market overall. The results show that even successful private gains can harm the market, revealing complex interactions among competing decision-makers.
Open → 2609.34821v1

AlphaOpsBench evaluates trading strategy execution in prediction markets

AlphaOpsBench: Benchmarking End-to-End Alpha Strategy Operationalization in Prediction Markets

Abstract: Large language models increasingly generate quantitative trading strategies, yet existing benchmarks assume standardized assets, numerical features, or directly compilable strategy representations---assumptions that prediction-market strategies violate, since a coarse idea may leave the traded outcome, causal information source, signal definition, threshold, sizing, order policy, exit, and settlement behavior unspecified. We introduce \textsc{AlphaOpsBench}, which evaluates end-to-end operationalization from source-grounded economic hypotheses to auditable executable programs over 581 source-preserving strategy records and a lifecycle-scale Polymarket dataset with 1.28 million binary markets, 183.6 million cleaned executions, settlement evidence, and limit-order-book history, comparing Direct generation against a Staged design-then-code protocol. In a corrected independent-generation study over 36 controlled tasks and 24 preregistered real strategies, strict end-to-end validity remains rare: Direct and Staged obtain 35/180 and 20/180 canonical passes on the controlled cohort and no confirmed pass on the real cohort, and repeated generations vary substantially in model-owned economic choices. By contrast, 775,725 of 783,655 scheduled historical replays complete, showing that replayability is a far weaker property than source-faithful operationalization. Financial outcomes depend on the declared execution model and available historical evidence, and fee and liquidity experiments show that execution costs alter subsequent trading paths rather than acting only as ex-post deductions. \textsc{AlphaOpsBench} thus separates strategy fidelity, behavioral validity, historical executability, and financial performance in an evidence-aware benchmark for LLM-based quantitative research in prediction markets.

Fri 25 SeptComputational Engineering, Finance, and Science
The gist
Creating trading strategies for prediction markets is tricky because many details about how trades happen are often missing from basic ideas. The authors created AlphaOpsBench, a new way to test how well large language models can turn rough strategy ideas into real, working programs over a large dataset of real market activity. They found it is very hard for models to fully and correctly follow through from idea to execution, with most attempts failing strict tests. Their benchmark helps to separate if a strategy is true to the original idea, behaves as expected, can replay historical trades, and produces real financial results.
Open → 2609.31390v1

Large language model agents cause fragile financial systems often failing

Financial Fragility in Societies of LLM Agents: Coordination Failures and Stabilizing Mechanisms

Abstract: Individually protective decisions can produce avoidable collective failures. As large language model (LLM) agents take on greater roles in financial decision-making, financial AI safety must therefore be considered not only at the level of individual agents, but also at the level of the systems they jointly create. We study this problem with FRAIL, a controlled experimental framework that places LLM agents in three dynamic financial environments---bank runs, debt rollover, and reward crowdfunding---where agents' decisions reshape the financial conditions faced by others. Across seven leading LLMs, we find widespread collective fragility even when no agent is instructed to destabilize the system: 77\% of baseline bank-run episodes and 83\% of debt-rollover episodes end in failure. We then compare three interaction mechanisms based on compensated commitments, centralized commitment agreements, and participant-led coalitions. All three improve aggregate outcomes, but no single mechanism performs best across all financial structures. Across mechanisms, successful stabilization shares a common temporal pattern: broad commitment forms early, before defensive behavior becomes self-reinforcing. Our findings show that individually capable agents do not automatically form safe financial systems, highlighting system-level evaluation and interaction design as central problems for financial AI safety. Code is available at https://anonymous.4open.science/r/FinFrail-CF26.

Fri 25 SeptArtificial Intelligence
The gist
Sometimes, when smart automated agents make financial decisions alone, their combined actions can cause big problems like financial crashes. The authors study how groups of AI agents behave in tricky financial situations, showing that they often end up causing failures even when not trying to. They test different ways these agents can work together to prevent these failures but find no single solution works best for all cases. Their work shows that building safe financial AI means looking beyond individual agents to how they act as a system.
Open → 2609.30940v1

Insurance reserve platform speeds up life policy calculations by 100 times

Insurance Reserve Intelligence Platform

Abstract: Insurance reserve estimation is a fundamental actuarial task supporting premium pricing, solvency assessment, financial reporting, capital planning, and risk management. Classical reserve methods based on Thiele's differential equation provide a rigorous and interpretable foundation for life insurance valuation, but repeated reserve calculations become computationally expensive in sensitivity analysis, optimization, and large-scale scenario evaluation. This paper presents an Insurance Reserve Intelligence Platform for term-life reserve modelling that combines a classical Thiele-equation solver with a Physics-Informed Neural Network (PINN) enhanced by Knowledge-Informed Neural Network (KINN) losses. The framework includes synthetic policy generation, risk-adjusted premium calculation, classical reserve trajectory generation, reserve-ratio dataset construction, configurable neural training, validation diagnostics, sensitivity and elasticity analysis, prototype optimization workflows, and interest-rate scenario testing. A key refinement is the use of premium ratio and the explicit separation of pricing-time and scenario-time interest-rate semantics. The final model uses seven features: elapsed time, issue age, pricing interest rate, scenario interest rate, premium ratio, sum assured, and mortality intensity. It predicts a standardized reserve ratio instead of raw reserve values, improving numerical stability across policies with different sums assured. The model achieved an R2 of 0.9887, MAE of 785.48, and RMSE of 1212.76 on the test set. On 200 policies, PINN/KINN inference was approximately 119.53 times faster than the classical solver. Results show strong predictive accuracy, physics consistency, and boundary performance, while highlighting remaining limitations in monotonicity and out-of-distribution generalization.

Fri 25 SeptArtificial Intelligence
The gist
Estimating money that insurance companies need to keep aside for future claims is important but can take a long time, especially when testing lots of what-if scenarios. The authors built a platform that combines traditional insurance math with new AI techniques to make these calculations faster and still accurate. Their model predicts a ratio instead of raw values to keep things stable across different policies. They showed it can produce results over 100 times faster than usual methods while keeping good accuracy.
Open → 2609.30765v1

Central bank digital currency design integrates credit and policy tools

Sovereign Grassroots Currencies: A CBDC Architecture for Credit and Monetary Policy (Full Version)

Abstract: A Central Bank Digital Currency (CBDC) is central-bank money in digital form, held by the public. Leading designs have two limitations: conversion from bank deposits into CBDC can accelerate deposit flight, requiring safeguards; the CBDC stays outside credit creation and monetary-policy operations. Here we present a CBDC architecture that overcomes these limitations, based on grassroots currencies. It has three components: (1) Money: sovereign grassroots coins, which are digital debts of one unit of fiat currency issued by the central bank, constituting a direct CBDC; (2) Credit and Liquidity: non-sovereign grassroots coins, which are digital debts of one unit of the same fiat currency, redeemable at par, that can be issued by any person, natural or legal - adding credit; and (3) Interest: grassroots bonds, sovereign and non-sovereign - adding maturity, and with it interest, standard banking instruments, and the central bank's instruments of monetary policy. The central bank can therefore lend, absorb liquidity, set its rates and buy and sell securities in the coins and bonds the public holds, choosing the counterparties and terms of its credit operations, and without converting bank deposits into newly issued central bank money on demand. We prove that one unit of the fiat currency is the only arbitrage-free price of a grassroots coin whose issuer meets presentations, and argue that the central bank's lending rate and the rate on its own bonds bound what its counterparties pay and accept on comparable terms; the central bank can choose to deal with any counterparty, not just banks.

Wed 23 SeptDistributed, Parallel, and Cluster ComputingMultiagent Systems
The gist
Converting bank deposits into digital money can cause risks, and current digital currencies avoid helping with loans or controlling the economy. The authors propose a new digital currency system that includes digital coins and bonds issued by both the central bank and other people, allowing credit creation and interest rates. This lets the central bank lend money, control liquidity, and adjust interest rates without causing deposit withdrawals from banks. Their system keeps digital coins equal in value to regular money while enabling more flexible financial operations.
Open → 2609.27727v1

Benchmark tests how ai agents authorize payments securely

APort Vault: Benchmarking AI Agent Payment Authorization with the Open Agent Passport

Abstract: APort Vault is a benchmark for payment authorization in tool-using AI agents. It replays 4,371 attacks written by humans against a live payment agent during a public capture-the-flag event, across 14 models from 8 labs, five policy configurations and two replay tracks, with and without a deterministic pre-action check implementing the Open Agent Passport (OAP) specification. 225,964 evaluations completed. We report five distinct events per evaluation, because collapsing them is how an agent benchmark produces a number that does not survive review. Requests are common and their rate differs far more across configurations than across models, though each attack exists at exactly one configuration so policy and attack cohort vary together: 10.9% of model-alone evaluations at Level 1, 3.0% at Level 2, 0.1% at Level 3, 79.4% at Level 4. On the 1,293 Level 4 prompts, each evaluated on every model, request rates run from 71.2% to 84.3%, and 809 prompts (62.6%) elicited a request from all fourteen models, each ending in a successful payment to the level's allowlisted recipient. The authorization boundary is where the conditions diverge. At Levels 2 to 4, transfers to recipients the passport did not permit number 140 of 76,842 with the model alone and 0 of 69,297 behind the layer, and 105 against 0 on 68,970 matched model, prompt and track triples. The zero spans 790 source sessions, giving a per-session upper bound of 0.38%. It was not obtained by refusing payments: 25,370 payments executed behind the layer, while the policy denied 187 of the 25,640 transfer calls it evaluated, 148 of them for a forbidden recipient. We release the 225,964 evaluations, the level passports, the scoring code and the analysis script at huggingface.co/datasets/aporthq/vault-benchmark-v1 .

Fri 18 SeptCryptography and Security
The gist
Payment systems powered by AI need to be safe from attackers trying to trick them into making unauthorized payments. The authors created a large set of tests called APort Vault, which uses thousands of real attack attempts to evaluate how well AI agents follow payment rules. They also tested a security layer called the Open Agent Passport, which stopped all unauthorized payment transfers in these tests. The study shows the effectiveness of this extra checking step in keeping AI-driven payments secure.
Open → 2609.22076v1

How people judge AI financial advice depends on style and source

Trustworthy FinAInce: Unpacking How AI-Mediated Financial Advice is Judged

Abstract: As generative AI is increasingly used as a source of personal financial guidance, understanding how people appraise such advice is important for supporting appropriate reliance. We conducted a randomized vignette experiment with 285 U.S. adults across eight financial decisions, independently varying three advice styles---AI, expert, and online community---and displayed source labels while holding the underlying recommendation consistent. Advice style most strongly shaped message and safety appraisals, Expert labels selectively increased perceived source knowledge, and decision context primarily shaped risk and safety appraisals. These appraisals were associated with downstream judgments, with models explaining 69.2% of overall quality, 75.9% of trust, and 82.9% of intended reliance. Expert-style advice also remained most preferred when shown without source labels. Our findings have implications for understanding financial advice evaluation, distinguishing the roles of advice style and source labels, and designing financial AI that supports grounded evaluation rather than simply maximizing trust.

Thu 17 SeptHuman-Computer InteractionArtificial IntelligenceComputation and Language
The gist
Financial advice can come from AI, experts, or online communities, but people trust and rely on each source differently. The authors studied how labeling the advice source and the style of advice affect people's opinions on trustworthiness and safety. They found expert advice is generally preferred and seen as more knowledgeable, even if the advice content is the same. The decision context also changes how risky or safe the advice seems. These insights can help design financial AI tools that encourage careful evaluation rather than blind trust.
Open → 2609.20989v1