{"paper_id":"2509.25055","title":"AlphaSAGE: Structure-Aware Alpha Mining via GFlowNets for Robust Exploration","authors":["Binqi Chen","Hongjun Ding","Ning Shen","Jinsheng Huang","Taian Guo","Luchen Liu","Ming Zhang"],"paper_date":"2025-09-29","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2509.25055v2","page_url":"https://thequant.space/flowcharts/alphasage-structure-aware-alpha-mining-via-gflownets-for-ro/","scores":{"empirical_rigor":6,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes AlphaSAGE, a structure‑aware, diversity‑seeking framework that combines a relational graph convolutional encoder, a GFlowNet generator, and a dense multi‑faceted reward to mine formulaic alphas.","quote":"we introduce AlphaSAGE (Structure-Aware Alpha Mining via Generative Flow Networks for Robust Exploration), a novel framework is built upon three cornerstone innovations: (1) a structure-aware encoder based on Relational Graph Convolutional Network (RGCN); (2) a new framework with Generative Flow Networks (GFlowNets); a","summary":"The automated mining of predictive signals, or alphas, is a central challenge in quantitative finance. While Reinforcement Learning (RL) has emerged as a promising paradigm for generating formulaic alphas, existing frameworks are fundamentally hampered by a triad of interconnected issues. First…","strategy":"AlphaSAGE"},"assumptions":{"status":"none_found","items":[]},"datasets":{"status":"available","items":[{"name":"CSI300","provider":null,"kind":"market-prices","access":"unknown","period":"2010-01-01 to 2024-12-31 (train/val/test","universe":"Chinese market constituents","quote":"market ... The split of the training set/validation set/test set for the Chinese market is defined as follows: 2010-01-01 to 2020-12-31 / 2021-01-01 to 2021-12-31 / 2022-01-01 to 2024-12-31."},{"name":"CSI500","provider":null,"kind":"market-prices","access":"unknown","period":"2010-01-01 to 2024-12-31 (train/val/test","universe":"Chinese market constituents","quote":"market ... The split of the training set/validation set/test set for the Chinese market is defined as follows: 2010-01-01 to 2020-12-31 / 2021-01-01 to 2021-12-31 / 2022-01-01 to 2024-12-31."},{"name":"S&P500","provider":null,"kind":"market-prices","access":"unknown","period":"2010-01-01 to 2020-12-31 (train/val/test","universe":"U.S. market constituents","quote":"For the US market: 2010-01-01 to 2016-12-31 / 2017-01-01 to 2017-12-31 / 2018-01-01 to 2020-12-31."}],"known":[],"sample":{"start":"2018-01","end":"2020-12","market":"S&P500","asset_class":"equities","frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":false,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"unclear","evidence_card_quoted":false},"meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"AlphaSAGE achieves the highest correlation metrics (IC 0.079, ICIR 0.496, RIC 0.094, RICIR 0.583) and the best portfolio outcomes (annualized return 7.62%, max drawdown -17.3%, Sharpe 1.71) across CSI300, CSI500, and S&P500.","kind":"other","quote":"AlphaSAGE ranks first on all correlation metrics, with notably higher ICIR/RICIR, and these gains translate into the best portfolio outcomes (highest annualized return, lowest drawdown, highest Sharpe)."},{"finding":"Ablation studies show that replacing the sequence encoder with a GNN provides the largest single lift across correlation and risk metrics, highlighting the value of structure‑aware representations.","kind":"other","quote":"Replacing the sequence encoder with a GNN provides the largest single lift across correlation and risk metrics, underscoring the value of structure-aware representations."},{"finding":"Adding the novelty reward improves signal quality and tradability by reducing redundancy among factors, while the entropy regularizer yields higher IC/RIC, AR, and Sharpe with controlled MDD.","kind":"other","quote":"the novelty reward (NOV) raises both signal quality and trad‑ability by reducing redundancy among factors. Finally, the entropy regularizer (ENT) yields the best overall results—higher IC/RIC, AR, and Sharpe with controlled MDD."}],"reported_performance":{"strategy":"AlphaSAGE","sharpe":6.32,"sharpe_benchmark":null,"annual_return_pct":19.47,"total_return_pct":null,"max_drawdown_pct":4.2},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"available","supports":[],"contradicts":[],"qualifies":[],"compared_with":22,"citations":{"cited_by_in_corpus":6,"by_independent_authors":6,"how_cited":{"used_as_baseline":1,"uses_method_or_data":2,"extends":0,"discusses_findings":0,"mentions":3,"unclassified":0},"independent_papers":[{"paper":{"paper_id":"2611.19937","title":"AllocFlow: End-to-End Portfolio Trading Action Generation from Raw Market Data","paper_date":"2026-11-15","authors":["Annan Wang","Yuchen Xie","Yu Cen","Leon Liangyu Chen","Qianyu Julie Zhu","Xin Li"],"empirical_rigor":8,"page_url":"https://thequant.space/flowcharts/allocflow-end-to-end-portfolio-trading-action-generation-from-raw-market-data/"},"how":"used_as_baseline","quote":null},{"paper":{"paper_id":"2610.06947","title":"FactorBench: A Portfolio-Aware Benchmark for Automated Factor Mining","paper_date":"2026-10-03","authors":["Zhuohan Wang","Carmine Ventre"],"empirical_rigor":9,"page_url":"https://thequant.space/flowcharts/factorbench-a-portfolio-aware-benchmark-for-automated-factor-mining/"},"how":"uses_method_or_data","quote":null},{"paper":{"paper_id":"2610.04959","title":"AlphaPADI: Formulaic Alpha Discovery via Pool-Aware Hierarchical Discrete Diffusion","paper_date":"2026-10-04","authors":["Yanzheng Jin","Pengyang Shao","Yunshan Ma","Haowen Pan","Naixin Zhai","Chen-Hui Song"],"empirical_rigor":6,"page_url":"https://thequant.space/flowcharts/alphapadi-formulaic-alpha-discovery-via-pool-aware-hierarchical-discrete/"},"how":"uses_method_or_data","quote":null}],"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/BerkinChen/AlphaSAGE","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":"2026-11-14T17:26:05Z"},"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/"}