Curated items for this scope are still being collected. Below are the latest related updates; items meeting the curation criteria will be marked separately.
The paper proposes a branching signature model that generalizes the classical geometric rough path based on the branching rough path framework, proving its universal approximation theorem, which can approximate high-level signatures through iterative combinations of low-level signature mappings. The model can explicitly extend the original path to a high-dimensional space via the mapping Ψ, implementing it as the classical geometric signature of the extended path, and its effectiveness is verified in two applications: rough volatility calibration and optimal execution.
This arXiv paper defines the recovery of tick-level queue history from aggregated L2 limit order book snapshots as a conditional partial identification problem. It constructs a path-preserving compiler that only varies among three cancellation allocation rules—front-of-queue, quantity-weighted random, and back-of-queue—while fixing the aggregated path, reconciled market removals, latent order partitioning, and new order arrivals.
The study addresses market-making challenges in Polymarket's thin order books, using the JaxMARL-HFT framework to train cooperative multi-agent market-making models on 417 real markets. The market-making agents are split into quoters and risk managers, trained using recurrent independent proximal policy optimization (IPPO) combined with RNN policies.
Testing showed that the improved single-agent market-making model quoted more frequently and had smaller out-of-sample losses, but none of the models achieved positive average portfolio value in out-of-sample evaluation, with only single-seed training completed.
This arXiv paper tests the degree to which 4 AI trading agents adhere to their own strategies and risk rules, based on 2627 closed trades from February 12 to August 27, 2026. 18% of the trades were live orders on Hyperliquid, while the rest were simulated at real-time prices, with a uniform round-trip fee of 0.09%.
This study introduces LOBGrad, a conditional DDPM framework for probabilistic order book forecasting. The framework combines an autoregressive RNN encoder with a diffusion-based generator, and establishes an excess-risk oracle inequality for the single-step DDPM noise prediction objective under exponential mixing and regularity assumptions.
This study introduces topological data analysis (TDA) and persistent homology (PH) into forecasting models in the field of prediction markets, exploring whether the underlying structure of market belief states has predictive value. The research findings show that single-parameter PH can improve the predictive power and calibration accuracy of linear forecasting models, and that two-parameter multi-persistent homology can produce similar gains for nonlinear models. The paper is 41 pages long with 19 figures, and belongs to the field of Trading and Market Microstructure (q-fin.TR).
To address the need for transaction cost estimation in scenarios without quote data, researchers developed an empirical characteristic function method with first-order bias correction for spread estimation, and established the asymptotic distribution of the corrected estimator under appropriate conditions. Numerical experiments show that this method has a bias-variance trade-off compared to the Roll model, and offers computational efficiency advantages over the Hasbrouck model.
This arXiv paper reproduces Calvano et al.'s 2020 simulated Bertrand duopoly Q-learning pricing experiment and challenges the conventional punishment-mechanism explanation for why independently trained tabular Q-learning agents converge to supracompetitive prices.
The research team built behavioral models for three types of trading agents—rational arbitrageurs, manipulators, and retail momentum followers—based on reinforcement learning, and conducted a pilot study on hourly cryptocurrency data (AAVE, ETC, SUI). The results show that the behavioral characteristics of different agents have differentiated impacts on market pressure, liquidity, volatility, and risk-adjusted returns. The research conclusions support formulating differentiated regulatory rules for different behavioral types, rather than uniformly restricting volatility or turnover rates.
Researchers propose OrderFlowDiT, a conditional diffusion Transformer trained with an unbalanced optimal transport (UOT) auxiliary loss, to predict the conditional distribution of limit order submissions over the next five minutes. Across 1,950 validation conditions covering 30 U.S. stocks, adding UOT reduces the full-field transport cost by 45.8% and the transport energy score by 26.1% compared with training using only mean squared error, with improvements in both metrics for every stock.
This study analyzes trade execution under concave power-law impact and Hurst-scaled price fluctuations. It determines the optimal execution horizon by maximizing the Sharpe ratio of realized trading P&L and identifies conditions for an interior optimum. The derived execution schedule can reduce envelope costs by 10-17% compared to a TWAP schedule over the same duration. The results are published on arXiv.
This paper proposes a multivariate state-dependent Hawkes process (SDHP) to model cross-product liquidity dynamics between quarter-hourly and hourly contracts in the Dutch EPEX SPOT continuous intraday market. The model is calibrated using Level 3 order book data from the first month of each quarter in 2024, dividing trading sessions into normal and stress states. The study finds that approximately 95% of orders are driven by their own order flow, with cross-excitation primarily flowing from quarter-hourly contracts to hourly contracts, and this dominance significantly weakens in stress states.
