{"paper_id":"2408.15846","title":"Trading with Time Series Causal Discovery: An Empirical Study","authors":["Ruijie Tang"],"paper_date":"2024-08-28","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2408.15846v2","page_url":"https://thequant.space/flowcharts/trading-with-time-series-causal-discovery-an-empirical-stud/","scores":{"empirical_rigor":6,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a workflow that uses causal discovery algorithms to build a trading strategy based on identified causal structures in stock price data.","quote":"An investment strategy was developed based on the causal structures identified by these algorithms.","summary":"This study investigates the application of causal discovery algorithms in equity markets, with a focus on their potential to build investment strategies. An investment strategy was developed based on the causal structures identified by these algorithms. The performance of the strategy is evaluated…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The market is liquid and short selling is permitted without borrowing costs.","type":"market","stated":"explicit","quote":"For simplicity, we assume a liquid market where short selling is permitted without incurring borrowing costs."},{"assumption":"A fixed daily transaction cost of 0.1% is applied to trades.","type":"execution","stated":"explicit","quote":"we impose a fixed daily transaction cost of 0.1% [7]."},{"assumption":"The trading strategy is dollar‑neutral with equal weighting on long and short sides.","type":"execution","stated":"explicit","quote":"our trading strategy employs a dollar‑neutral portfolio, ensuring that the dollar amounts of long and short positions are balanced"},{"assumption":"Linear regression is used for one‑day‑ahead predictions.","type":"model","stated":"explicit","quote":"using the one‑day‑ahead predictions obtained through regression"},{"assumption":"VarLiNGAM assumes linear relationships, non‑Gaussian independent errors, and no latent confounders.","type":"model","stated":"explicit","quote":"The LiNGAM algorithm assumes that each variable is a linear function of its causes plus an error term ei, which follows a non‑Gaussian distribution with non‑zero variance. Additionally, the error terms ei are independent across variables, and the system is assumed to contain no latent confounders."},{"assumption":"Time lag attributes are removed from tsFCI and VarLiNGAM causal graphs for analysis.","type":"data","stated":"explicit","quote":"the causal graphs generated by tsFCI and VarLiNGAM are compressed by removing the time lag attributes for subsequent analysis."}]},"datasets":{"status":"available","items":[{"name":"SP500 component stocks","provider":"[15]","kind":"market-prices","access":"public","period":"2009.09.01 - 2019.12.31","universe":"SP500 constituents","quote":"The data, sourced from [15], covers a period of 10 years (2009.09.01 - 2019.12.31) and includes prices for SP500 and CSI300 component stocks."},{"name":"CSI300 component stocks","provider":"[15]","kind":"market-prices","access":"public","period":"2009.09.01 - 2019.12.31","universe":"CSI300 constituents","quote":"The data, sourced from [15], covers a period of 10 years (2009.09.01 - 2019.12.31) and includes prices for SP500 and CSI300 component stocks."},{"name":"Nancy Pelosi portfolio","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":"past five years","universe":"12 stocks in Pelosi's current portfolio","quote":"We collected stock prices from her current portfolio, comprising 12 stocks, over the past five years from Yahoo Finance.","known_id":"yahoo-finance"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"unclear","headline_net_of_costs":null,"detail":null,"quote":null,"source":"conflict","rule":"R4b","candidates":{"front_matter":"partial","evidence_card":"yes","evidence_card_quoted":true},"conflict_kind":"graded","card_detail":"fixed daily transaction cost of 0.1% per trade","card_quote":"we impose a fixed daily transaction cost of 0.1% [7].","meaning":"the platform classifier and the paper-text extraction disagree, so no value is asserted"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"VarLiNGAM is the most effective algorithm, handling large datasets well and generating the highest returns across all datasets.","kind":"performance","quote":"Among the three algorithms, VarLiNGAM is the most effective, handling large datasets well and generating the highest returns on all datasets."},{"finding":"Incorporating a causal discovery phase generally improves portfolio returns compared to a self‑cause‑only baseline.","kind":"performance","quote":"Causal discovery proves more effective than the self‑cause‑only setting. The portfolio supported by causal discovery significantly outperforms the self‑cause‑only portfolio."},{"finding":"Causal discovery is feasible for large asset classes (SP500, CSI300) but ineffective for the small Pelosi portfolio.","kind":"negative","quote":"causal discovery proves ineffective when applied to smaller asset classes like Pelosi, where the true driving forces of its component stocks fall outside the analysis scope."},{"finding":"VarLiNGAM is the only algorithm that can process 400–500‑stock datasets; tsFCI and TiMINo are limited to much smaller datasets.","kind":"performance","quote":"VarLiNGAM is the only feasible option for analyzing the stock market with 400–500 stocks, while tsFCI and TiMINo are constrained to processing much smaller datasets."},{"finding":"Using a smaller time lag and selecting 1%‑6% of stocks as winners yields the best returns for the trading strategy.","kind":"performance","quote":"For our trading strategy, setting the number of winner stocks to 1%-6% of the size of the original asset class yields the best returns."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":35,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"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/kaaaylaaa/causality-based-trading","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":"2026-11-14T18:03:18Z"},"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/"}