FUND-RELATED GRAPH REPRESENTATION FOR MARGINAL EFFECTIVENESS IN MULTI-FACTORS QUANTITATIVE STRATEGY

Biao Li, Yang Du, Zhichen Lu, Gang Kou, Yunxi Yan, Yong Shi

Primary Area: general machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Quantitative strategy, Representation Learning, Marginal effectiveness, Capital Flow

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: With increasing research attention in the quantitative trading community on multi-factors machine learning strategies, how to obtain higher-dimensional and effective features from finance market has become an important research topic in both academia and industry area. In general, the effectiveness of new data, new factors, and new information depends not only on the strength of their individual effects but also on the marginal increment they bring relative to existing factors. In this paper, our research focuses on how to construct new factors from the relational graph data. We construct six capital flow similarity graphs from the frequency of joint occurrences of the inflows or outflows of the net fund between stocks within the same period. Moreover, three composite multi-graphes from the six basics are built to exploit the capital flow similarities. Experiments demonstrate the marginal improvement contributing to these proposed graphs. Learned by the multi-factor XGBoost model, the new dataset integrating with representations of the fund-related graphs exceeds the baseline multi-factor model in the Information Coefficient(IC), TOP group excess returns, long-short returns, and index-enhanced portfolio returns in A-share market.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2179