# q-fin.TR Published articles for q-fin.TR. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Towards Robust Reinforcement Learning Market Making: Lessons from Loss Function Experimentation DevFeed: [Towards Robust Reinforcement Learning Market Making: Lessons from Loss Function Experimentation]() Original publisher: [Read original article]() Author: Siddharth Nand, Bora Guney Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Reinforcement learning](), [Machine learning]() Tags: [algorithmic-trading](), [cs-lg](), [evaluation](), [experimentation](), [financial-markets](), [inventory](), [management](), [market-making](), [q-fin-cp](), [q-fin-tr](), [reinforcement-learning](), [risk](), [risk-management](), [robust]() ### AI overview The abstract reports that reinforcement learning market-making strategies with penalties for inventory and cash management were more resilient and adaptable than profit-maximizing approaches in the paper's empirical evaluation. It presents balancing profit and risk management as important, but provides no details on the loss formulations, experimental setup, or measured outcomes. ### Source excerpt arXiv:2611.09012v1 Announce Type: new Abstract: Market makers play a vital role in ensuring efficient trade within financial markets. Recent literature has delved into training market-making systems using reinforcement learning. This paper delves into the impact of various loss function formulations on risk management behavior within reinforcement-based market making. Through empirical evaluation, we contrast profit-maximizing strategies with risk-averse approaches, finding that penalties for inventory and cash management foster more resilient and adaptable market-making strategies. Our results underscore the significance of striking a balance between profit maximization and risk management in algorithmic trading, providing actionable insights for refining strategies in practical settings. ## Does a Trading Policy Need to Know What It Trades? Microstructure Invariance and Zero-Shot Transfer of Execution Policies DevFeed: [Does a Trading Policy Need to Know What It Trades? Microstructure Invariance and Zero-Shot Transfer of Execution Policies]() Original publisher: [Read original article]() Author: Hieu Tran, Hong Yu Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [World models](), [Representation learning](), [few-shot]() Tags: [crypto](), [cs-ce](), [cs-lg](), [execution](), [information](), [invariance](), [nasdaq](), [policies](), [q-fin-tr](), [queue](), [raw](), [reinforcement-learning](), [trading](), [train](), [training](), [units](), [zero-shot]() ### AI overview In exact order-by-order replay, execution policies trained on 954 Nasdaq stocks transferred to unseen stocks with costs within 0.015 spreads per share of policies trained with those stocks; using each stock's history did not help. Transfer was similarly close to policies trained on a second exchange when episodes counted equally, and the gap was wider for crypto. In mean pre-fee cost, microstructure-invariant units outperformed raw units, z-scores and asset descriptors by less than one basis point. Much of that edge came from information and a shared cost scale that could also be supplied as plain inputs. None of the tested representations extrapolated to unseen spread regimes. The authors caution that a near-zero gap against a pooled comparator is weak evidence of a good representation, since that comparator detects only a small share of losses from a poor one. ### Source excerpt arXiv:2611.08701v1 Announce Type: new Abstract: Execution algorithms are calibrated asset by asset, because the queues, trade rates and bid-ask spreads they face differ by orders of magnitude across assets. For reinforcement learning this is zero-shot generalisation across contexts of very different scales, and microstructure invariance proposes a normalisation fixed by theory: measured in each asset's own units of time, size and price, markets look alike. In an exact order-by-order replay of public Nasdaq data, we train policies once, on 954 stocks in 2019, to choose between waiting in the queue and crossing the spread, and evaluate them unchanged on 1,964 unseen stocks in 2020-21, 2,629 stocks in 2025-26, crypto pairs and a second exchange. First, a policy that sees a stock only through its recent trade rate, trade size, spread and price increment (tick) costs within 0.015 spreads per share (a fifth of what learning saves) of a policy whose training data included the stock, and the stock's own history does not help it; it comes within a similar margin of policies trained on the second exchange when every episode counts equally, and within a wider one on crypto. Second, in mean cost before fees, invariant units beat raw units, z-scores and asset descriptors, but by under a basis point: most of their advantage is information and a common cost scale that a learner can also take as plain inputs, and none of the representations tested extrapolates to unseen spread regimes. Third, a near-zero transfer gap to a pooled comparator is weak evidence that a representation is right: we show that such a comparator detects only a small share of what a poor representation loses, and confirm it in synthetic markets and on pendulums. ## Behavioral Prior Elicitation via LLM for Heterogeneous-Agent Bitcoin Flash-Crash Simulation DevFeed: [Behavioral Prior Elicitation via LLM for Heterogeneous-Agent Bitcoin Flash-Crash Simulation]() Original publisher: [Read original article]() Author: Tran Phuong Linh Pham, Thi Minh Phuong Ngo, Phuong Thao Nguyen, Quang Minh Nguyen, Thai Hoa Ngo, Trong Nghia Nguyen Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Large Language Model](), [Simulation]() Tags: [bitcoin](), [counterfactual-analysis](), [crash](), [crash-simulation](), [cs-ai](), [flash](), [large](), [large-language-model](), [leverage](), [limit-order-book](), [llm](), [q-fin-tr](), [simulation](), [stat-ap](), [synthetic](), [tails]() ### AI overview The study uses an offline LLM, conditioned on empirical microstructure anchors, to elicit behavioral priors for four agent types in a heterogeneous BTCUSDT limit order book simulator. Its four-stage pipeline builds a 100 ms event panel from 66 detected flash crashes, validates the priors, then uses synthetic simulations for causal discovery and counterfactual analysis. Across 500 runs, LLM-elicited priors yielded a 0.176 crash rate versus 0.008 with uniform priors and produced heavier return tails. In the frozen counterfactual, lowering leverage amplification by one within-sample standard deviation reduced predicted crash rate to 0.136; lowering directional pressure had only a marginal effect. The findings concern a controlled synthetic environment. ### Source excerpt arXiv:2611.03660v1 Announce Type: cross Abstract: Bitcoin perpetual futures trade continuously with shallow liquidity and high price impact, making them prone to millisecond-scale flash crashes. Agent-based limit order book simulators are a natural tool for studying such events, but their behavioral parameters are usually hand-crafted or require full order book histories that are rarely available at scale. We propose a four-stage pipeline that constructs a reproducible 100 ms BTCUSDT flash-crash event panel from 66 detected events, elicits behavioral priors for four agent archetypes using an offline large language model conditioned on empirical microstructure anchors, instantiates a heterogeneous limit order book simulator with the validated priors, and applies causal discovery plus counterfactual analysis to the resulting synthetic panel. Across 500 simulation runs, LLM-elicited priors produce a crash rate of 0.176 versus 0.008 under uninformative uniform priors and generate heav- ier return tails. In the frozen counterfactual analysis, reducing leverage amplification by one within-sample standard deviation lowers the predicted crash rate from 0.176 to 0.136, whereas reducing directional pressure yields only a marginal change. These find- ings suggest that empirically anchored LLM elicitation can improve simulator fidelity and help identify plausible crash-amplification mechanisms in a controlled synthetic environment ## FinLLM-KG: Knowledge Graph-Enhanced LLMs for Financial Sentiment Analysis in Algorithmic Trading DevFeed: [FinLLM-KG: Knowledge Graph-Enhanced LLMs for Financial Sentiment Analysis in Algorithmic Trading]() Original publisher: [Read original article]() Author: Giorgos Iacovides, Oliver Hannibal, Wuyang Zhou, Danilo Mandic Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [база знаний](), [Large Language Model](), [Retrieval Augmented Generation (RAG)](), [Inference](), [Finance]() Tags: [algorithmic-trading](), [cs-cl](), [cs-lg](), [explainability](), [financial](), [knowledge-graph](), [llms](), [q-fin-st](), [q-fin-tr]() ### AI overview FinLLM-KG adds temporally grounded company knowledge from SEC 10-K filings to causal LLM sentiment predictions at inference time, without changing model weights. It supplies context as either a natural-language company profile or a small set of relevance-ranked factual triples, letting predictions draw on firm-level information as well as the article. The framework is described as model-agnostic, with almost zero computational cost, and links sentiment scores to sentences and retrieved facts to make their evidence inspectable for human review. Simulations reportedly show consistently better portfolio performance than the state-of-the-art baseline; the excerpt provides no quantitative results or evaluation details. ### Source excerpt arXiv:2610.36083v1 Announce Type: cross Abstract: Recent advances in Generative AI have substantially improved financial sentiment analysis through post-trained financial large language models (LLMs). However, existing approaches remain confined to predicting the sentiment of each article in isolation and operate as black boxes, which limits contextual understanding and hinders deployment in high-stakes financial decision-making. To this end, we introduce FinLLM-KG, a model-agnostic framework that combines causal LLMs with a financial knowledge graph, in order to inject temporally grounded, company-level knowledge from SEC 10-K filings at inference time, without updating model weights. Within FinLLM-KG, context is supplied either as a natural-language company profile or as a small set of relevance-ranked factual triples. This enables sentiment predictions to account for broader firm-level information rather than the article alone, at almost zero computational cost and for any causal LLM. In addition, attributing each article-level sentiment score to individual sentences and to the retrieved knowledge-graph facts makes the evidence behind the prediction inspectable, thereby enhancing explainability and supporting human-in-the-loop review. Simulations demonstrate that FinLLM-KG consistently improves portfolio performance over the state-of-the-art baseline. ## How do standard time series forecasting benchmarks translate to finance? A broad study of risk adjusted performance DevFeed: [How do standard time series forecasting benchmarks translate to finance? A broad study of risk adjusted performance]() Original publisher: [Read original article]() Author: Adir Saly-Kaufmann, Kieran Wood, Jan-Peter Calliess, Stefan Zohren Published: 2026-10-26T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Finance](), [Time Series]() Tags: [architectures](), [asset](), [attention](), [benchmarks](), [costs](), [cs-lg](), [dataset](), [deep-learning](), [distributions](), [finance](), [financial-forecasting](), [forecasting](), [futures](), [hybrid](), [lstm](), [models](), [network](), [performance](), [q-fin-tr](), [risk](), [time-series](), [time-series-forecasting]() ### AI overview On a 15-year daily futures dataset, a comparison of 16 forecasting architectures using volatility-targeted Sharpe ratio optimization found that standard benchmark success did not reliably carry over to finance. Several generic Transformers and State Space Models underperformed a simple AR(1) baseline. A Variable Selection Network combined with an LSTM achieved the highest Sharpe ratio, while xLSTM was most robust to transaction costs. The abstract attributes the challenge to returns' low signal-to-noise ratio, non-stationarity, and heavy tails. ### Source excerpt arXiv:2603.01820v2 Announce Type: replace-cross Abstract: Forecasting financial asset returns is challenging due to low signal-to-noise ratios, non-stationarity and heavy-tailed distributions. We benchmark 16 deep learning architectures for time series forecasting, including Transformers, State Space Models (SSMs), xLSTMs and recurrent hybrids, on a 15-year daily futures dataset using an end-to-end volatility-targeted Sharpe ratio optimisation task. Our results challenge prevailing assumptions in the literature. Several widely used 'state-of-the-art' architectures, including generic Transformers and SSMs, underperformed a simple AR(1) benchmark in a realistic financial forecasting task. The strongest results came from comparatively simple hybrid approaches using recurrent structures: a Variable Selection Network combined with an LSTM achieved the highest Sharpe ratio, while xLSTM showed the greatest robustness to transaction costs. These findings suggest that success on standard time-series benchmarks does not reliably transfer to finance, and that architectures tailored to the unique properties of asset returns may remain more effective than generic attention-based approaches for financial forecasting. ## Understanding the Inner Workings of a State-Space Foundation Model for Limit Order Books DevFeed: [Understanding the Inner Workings of a State-Space Foundation Model for Limit Order Books]() Original publisher: [Read original article]() Author: Valentin Mohl, Alexandre Bismuth, Aramis Marti-Shahandeh, George Nigmatulin, Kang Li, Stefan Zohren, Jakob Foerster, Mihai Cucuringu, Anisoara Calinescu Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Mamba](), [AI Foundation Models](), [model architecture](), [World models]() Tags: [cs-lg](), [foundation-model](), [mamba](), [market-microstructure](), [order-flow-imbalance](), [q-fin-tr](), [stocks](), [trading](), [volume](), [world-model]() ### AI overview An 8.2M-parameter Mamba-3 model trained on tokenized limit order book messages from eight large-cap US equities learns a decodable representation of market dynamics. Order-flow imbalance is recovered at about R² = 0.59, above controls and baselines; the representation also extends to unseen stocks and captures traded volume and volatility. Steering the model's internal state along a feature axis shifts its next-message beliefs monotonically in the direction implied by that feature. These findings suggest an internal market representation, while the abstract says practical use in real-world trading remains uncertain. ### Source excerpt arXiv:2610.28753v1 Announce Type: cross Abstract: Autoregressive foundation models offer state-of-the-art capabilities to simulate the market by generating the message flow of limit order books (LOBs). Yet, without understanding their inner workings, research directions are hard to narrow and practical use in real-world trading environments remains uncertain. This study therefore asks what these models actually learn about market microstructure. We find that, similarly to games where sequence models build an internal representation of the board, an 8.2M-parameter Mamba-3 state-space model trained autoregressively on the tokenized message stream of eight large-cap US equities learns an internal representation of the market and its dynamics. In particular, order-flow imbalance, a central driver of short-horizon price moves, is decodable at $R^2 \approx 0.59$, far above controls and baselines. This representation extends to unseen stocks, suggesting that these models learn general representations rather than ticker-specific patterns. The model also generalizes across metrics beyond order-flow imbalance: results hold across traded volume and volatility. Additionally, the representation is causally steerable: pushing the state along a feature's axis shifts the model's next-message beliefs as that feature implies, monotonically in dose, consistent with the emergence of a world model. ## LOB-Impact Benchmark: Emergent Market Impact in Generative Limit Order Book Models DevFeed: [LOB-Impact Benchmark: Emergent Market Impact in Generative Limit Order Book Models]() Original publisher: [Read original article]() Author: George Nigmatulin, Valentin Mohl, Sascha Frey, Kang Li, Aramis Marti-Shahandeh, Mihai Cucuringu, Anisoara Calinescu, Jakob Foerster, Stefan Zohren Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Benchmark](), [World models]() Tags: [autoregressive](), [benchmark](), [cs-lg](), [family](), [generation](), [generative](), [generative-models](), [impact](), [limit-order-book](), [market-impact](), [models](), [nasdaq](), [neural](), [next](), [next-token-prediction](), [open-source](), [orders](), [persistence](), [q-fin-tr](), [simulator](), [square-root-law](), [stock]() ### AI overview The benchmark tests whether autoregressive limit order book generators reproduce market impact when execution-style metaorders are injected. Across five models and five NASDAQ stocks, it assesses free-generation order-flow persistence and price diffusivity, impact-law exponent and amplitude during injection, and post-execution relaxation; visible injections are compared with blind simulator-only controls to distinguish learned response from book mechanics. Measured impact exponents range from 0.63 to 1.21, above the traditional 0.5 square-root-law value. The authors report no observable impact decay during relaxation in the tested models, tentatively linking this limitation to short context windows, and release the benchmark as open source. ### Source excerpt arXiv:2610.27100v1 Announce Type: cross Abstract: Realistic market impact is essential for generative limit order book (LOB) models to serve as world models for trading agents. If a simulator misprices this mechanism, agents learn policies that are optimal in the simulator but wrong in the real market. Existing metrics assess generated streams purely through distributional comparison with historical data, leaving counterfactual response to an agent's own orders largely unexplored. Here, we show that market impact emerges naturally in neural LOB generators trained solely for next-token prediction. We inject execution-style metaorders into the generated message flow of five autoregressive models across five NASDAQ stocks and measure the response using estimators from the empirical impact literature, validated on baselines whose physics is known in advance. Our benchmark evaluates any message-level generator in three phases: free generation (reading order-flow persistence against price diffusivity); metaorder injection (yielding the impact law's exponent and amplitude); and a post-execution window (measuring relaxation). In addition, comparing each visible injection to a blind, simulator-only control isolates the model's learned response from standard book mechanics. For every model on every stock, we measure how the impact responds as a function of cumulative metaorder participation. Interestingly, we observe emergent impact in generative models of market data, with exponents in the range of 0.63 to 1.21, slightly above the traditional 0.5 of the square-root law of impact. In the family of models tested, we are so far unable to observe impact decay during relaxation. We tentatively attribute this to the still-short context windows of the models. We release the benchmark as open source to validate future generative models for market data. ## The Control Layer Behind Rule-Following Agents Trading Across Market Conditions DevFeed: [The Control Layer Behind Rule-Following Agents Trading Across Market Conditions]() Original publisher: [Read original article]() Author: Saad Naja, Nasser Odtallah, Jake Van-Praag, Sarah May Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [google-adk](), [test](), [reproducibility](), [audit]() Tags: [agent-evaluation](), [agents](), [ai](), [ai-trading](), [ai-trading-agents](), [cs-ai](), [cs-lg](), [cs-ma](), [evaluation](), [q-fin-rm](), [q-fin-tr](), [risk](), [rules](), [strategies](), [strategy](), [test](), [tests](), [toolkit](), [trade](), [trading]() ### AI overview Across 2,627 closed positions from February to August 2026, four AI trading agents showed high adherence to declared strategies and risk rules, even though one was unprofitable. Most positions were simulated against live prices with a 0.09% round-trip fee; 18% used real Hyperliquid orders. Agents kept 97.5% of gross losses within declared stops, and 23 of 24 runs respected the five-position limit. Profitability tracked win frequency relative to each strategy's break-even rate, while the same market data produced different results under different strategies. The study offers rerunnable trade-log tests and regime tests requiring published Bitcoin prices and regime labels. These are trade-level findings, not account-level returns or proof of a lasting edge; the authors present discipline as a necessary but insufficient test before allocating capital. ### Source excerpt arXiv:2610.26324v1 Announce Type: cross Abstract: Trading discipline often fails people. Our study tested whether four AI trading agents followed their strategies and risk rules. Offering traders the evidence and an agent evaluation toolkit to separate luck from following the rules, before giving agents meaningful autonomy or capital. Our study analysed 2,627 closed positions from 12 February to 27 August 2026 across the same asset universe and execution conditions. Most positions were simulated against live prices with a 0.09% round-trip fee; 18% involved real Hyperliquid orders. Each position declared a strategy, stop and target, with a limit of five open positions per agent. The fleet finished net positive after fees. Rule adherence was high across all four agents, including the unprofitable one. Profitability tracked how often agents won relative to their break-even win rates. Institutional-style controls ran by default and contained losses through a 29% market drawdown and recovery, with documented exceptions. The discipline held for all four agents. Agents kept 97.5% of gross loss within declared stops, and typical losses closed at declared levels. Declared risk remained stable after losses and broadly stable across calm and volatile months. Twenty-three of 24 runs respected the position limit; logs identify the exception's size and duration. The same live data produced different results across assigned strategies. The trade-log tests can be rerun from the published trade snapshot; regime tests also require the published Bitcoin prices and regime labels. Section 3.5 specifies the stop-breach tolerance. These trade-level results are not account-level returns or proof of a lasting edge. Our study provides an evaluation toolkit to help traders distinguish rule-following from luck before committing capital. For anyone considering allocating capital to agents to trade, discipline should be the entry test, not the only test. ## Probabilistic Limit Order Book Forecasting with Conditional Diffusion Models DevFeed: [Probabilistic Limit Order Book Forecasting with Conditional Diffusion Models]() Original publisher: [Read original article]() Author: Junyu Chen, Franklin Liu, Moshu Xu, Lijian Yang Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [VAE](), [Predictive AI](), [Time Series]() Tags: [conditional](), [cs-lg](), [diffusion](), [diffusion-models](), [econ-em](), [forecasting](), [limit-order-book](), [q-fin-tr]() ### AI overview LOBGrad combines an autoregressive RNN encoder with a conditional DDPM generator to forecast limit order books probabilistically. The authors establish an excess-risk oracle inequality for its one-step noise-prediction objective under exponential mixing and regularity assumptions; with contractive dynamics, the conditioning-window dependence can be reduced to dependence on latent-state dimension. On BTC/USDT, GOOGL, and META, it improves on an encoder-free diffusion ablation across reported queue metrics and is competitive with other baselines. The abstract also reports reproduction of key LOB stylized facts and reasonably calibrated queue-based predictive intervals, while a stylized downstream task illustrates potential forecast value. ### Source excerpt arXiv:2610.25495v1 Announce Type: new Abstract: Modeling limit order book (LOB) dynamics is central to market forecasting and execution-strategy evaluation. While diffusion models achieve strong empirical performance, existing LOB snapshot models typically lack a learned recurrent encoder and formal statistical guarantees. We introduce LOBGrad, a conditional DDPM framework for probabilistic LOB forecasting that combines an autoregressive RNN encoder with a diffusion-based generator. In particular, we establish an excess-risk oracle inequality for the one-step DDPM noise-prediction objective under exponential mixing and regularity assumptions. We further show that, under contractive dynamics, the dependence on the conditioning-window length can be reduced to a dependence on the latent-state dimension. Experiments on BTC/USDT, GOOGL, and META show that LOBGrad improves on its encoder-free diffusion ablation across all reported queue metrics and obtains competitive results against other baselines. It also reproduces the key stylized facts of LOB microstructure, and yields reasonably calibrated queue-based predictive intervals. A stylized downstream task illustrates the potential value of the generated forecasts.