{"paper_id":"2503.02680","title":"VWAP Execution with Signature-Enhanced Transformers: A Multi-Asset Learning Approach","authors":["Remi Genet"],"paper_date":"2025-03-04","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2503.02680v1","page_url":"https://thequant.space/flowcharts/vwap-execution-with-signature-enhanced-transformers-a-multi/","scores":{"empirical_rigor":8,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a globally‑fitted transformer model with path‑signature features (GFT‑Sig) for VWAP execution that can be trained across multiple assets and achieve performance comparable to or better than asset‑specific models.","quote":"The proposed architecture combines a transformer‑based design inspired by arXiv:2406.02486 with path signatures for capturing geometric features of price‑volume trajectories, as in arXiv:2406.17890. The empirical analysis, conducted on hourly cryptocurrency trading data from 80 trading pairs, shows that the globally‑fi","summary":"In this paper I propose a novel approach to Volume Weighted Average Price (VWAP) execution that addresses two key practical challenges: the need for asset-specific model training and the capture of complex temporal dependencies. Building upon my recent work in dynamic VWAP execution…","strategy":"GFT-Sig"},"assumptions":{"status":"available","items":[{"assumption":"Volume conservation constraint and non‑negativity of normalized order allocation (˜qt ≥ 0).","type":"execution","stated":"explicit","quote":"Under the volume conservation constraint and non‑negativity constraints ˜qt ≥0, this decomposition separates the overall slippage ST into a price deviation component and a volume allocation error component."},{"assumption":"Future prices and volumes are unknown at execution time, requiring accurate predictions.","type":"market","stated":"implied","quote":"This optimization problem is particularly challenging because future prices and volumes are unknown at execution time, which necessitates accurate predictions of market dynamics while managing execution risk."},{"assumption":"Rolling two‑week median normalization (shifted forward by 12‑hour horizon) removes volume trends and prevents look‑ahead bias.","type":"data","stated":"explicit","quote":"First, a rolling two‑week median normalization, shifted forward by the 12‑hour prediction horizon, removes volume trends and prevents look‑ahead bias."},{"assumption":"Path‑signature features summarise recent price‑volume trajectories in a compact form and can be used as additional contextual features.","type":"model","stated":"implied","quote":"By conditioning on signature‑derived features that summarize recent price‑volume trajectories in a compact form, the model is better equipped to capture nuanced temporal dependencies relevant to VWAP execution."}]},"datasets":{"status":"available","items":[{"name":"Hourly trading data from 80 cryptocurrency pairs","provider":"Binance","kind":"market-prices","access":"commercial","period":"up to July 1, 2024","universe":"80 cryptocurrency pairs listed on Binance","quote":"The experiments utilize a comprehensive dataset of hourly trading data from 80 cryptocurrency pairs listed on Binance. The data spans from each asset’s inception up to July 1, 2024, providing a rich and diverse sample for evaluation.","known_id":"binance"}],"known":[{"id":"binance","name":"Binance","provider":"Binance","kind":"market-prices","access":"public","quote":"3.1 Dataset and Preprocessing The experiments utilize a comprehensive dataset of hourly trading data from 80 cryp-tocurrency pairs listed on Binance.","mentions":4,"used_in":3}],"sample":{"start":null,"end":null,"market":"cryptocurrency trading pairs","asset_class":"crypto","frequency":"intraday"},"data_resolution":"intraday-bars"},"transaction_costs":{"status":"available","modeled":"partial","headline_net_of_costs":null,"detail":"Reductions in absolute (basis points) and quadratic (millionths) deviations from market VWAP are reported relative to a TWAP benchmark.","quote":"Table 2 summarizes the real‑time performance of our VWAP execution strategy relative to a TWAP benchmark for order durations of 30, 120, 480, and 1440 minutes. Overall, the proposed method consistently reduces both the absolute and quadratic deviations from the market VWAP.","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"partial","evidence_card_quoted":true},"card_detail":"Reductions in absolute (basis points) and quadratic (millionths) deviations from market VWAP are reported relative to a TWAP benchmark.","card_quote":"Table 2 summarizes the real‑time performance of our VWAP execution strategy relative to a TWAP benchmark for order durations of 30, 120, 480, and 1440 minutes. Overall, the proposed method consistently reduces both the absolute and quadratic deviations from the market VWAP.","meaning":"costs appear only as a robustness check, a flat fee without spread or impact, or outside the headline numbers"},"results":{"status":"available","evaluation_type":"execution-cost","findings":[{"finding":"The globally‑fitted GFT‑Sig model achieves superior absolute and quadratic VWAP loss compared to asset‑specific models.","kind":"performance","quote":"The empirical analysis, conducted on hourly cryptocurrency trading data from 80 trading pairs, shows that the globally‑fitted model with signature features (GFT‑Sig) achieves superior performance in both absolute and quadratic VWAP loss metrics compared to asset‑specific approaches."},{"finding":"Improvements increase with order duration, with quadratic deviation reductions up to 70‑74% and absolute deviation reductions of 20‑47%.","kind":"performance","quote":"Notably, the reduction in quadratic deviation is even more pronounced than that in absolute deviation."},{"finding":"The model generalizes to out‑of‑sample assets, maintaining performance across different market conditions.","kind":"performance","quote":"Notably, these improvements persist for out‑of‑sample assets, demonstrating the model's ability to generalize across different market conditions."}],"reported_performance":{"strategy":"GFT-Sig","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"available","supports":[],"contradicts":[{"paper":{"paper_id":"2611.15646","title":"Idiosyncratic Market Effects in Global Time Series Momentum","paper_date":"2026-11-11","authors":["Andrew Butler","Maciej Zawadzki","Adam Butler"],"empirical_rigor":7,"page_url":"https://thequant.space/flowcharts/idiosyncratic-market-effects-in-global-time-series-momentum/"},"question":"Do globally‑trained models outperform asset‑specific models for VWAP execution and time‑series momentum?","this_paper_finding":"The globally‑fitted GFT‑Sig model achieves superior absolute and quadratic VWAP loss compared to asset‑specific models.","other_paper_finding":"Fitting nonlinear models to individual markets yields a marginal out‑of‑sample performance improvement over global models.","other_paper_quote":"fitting nonlinear models to individual markets is a robust and often overlooked modelling approach that can result in a marginal improvement in out‑of‑sample performance.","note":"A shows global GFT‑Sig beats asset‑specific on absolute and quadratic VWAP loss; B finds only marginal out‑of‑sample gain for market‑specific nonlinear models, contradicting A.","confidence":0.467}],"qualifies":[],"compared_with":30,"citations":{"cited_by_in_corpus":1,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":1},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":true,"code_url":"https://github.com/remigenet/DynamicVWAPTransformer","note":null},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T17:43:27Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}