{"paper_id":"2401.05414","title":"On the Three Demons in Causality in Finance: Time Resolution, Nonstationarity, and Latent Factors","authors":["Xinshuai Dong","Haoyue Dai","Yewen Fan","Songyao Jin","Sathyamoorthy Rajendran","Kun Zhang"],"paper_date":"2023-12-28","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2401.05414v2","page_url":"https://thequant.space/flowcharts/on-the-three-demons-in-causality-in-finance-time-resolution/","scores":{"empirical_rigor":6.5,"math_complexity":7,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a rank‑based latent causal discovery method that leverages rank constraints of the cross‑covariance matrix to uncover underlying causal structure, including latent variables, and provides a systematic solution that works when time‑resolution mismatch, non‑stationarity, and latent ","quote":"Rank Constraint and Motivation. Rank-based Latent Causal Discovery Algorithm.","summary":"Financial data is generally time series in essence and thus suffers from three fundamental issues: the mismatch in time resolution, the time-varying property of the distribution - nonstationarity, and causal factors that are important but unknown/unobserved. In this paper, we follow a causal…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The underlying data generation follows a linear instantaneous causal model so that log returns can be treated as linearly aggregated i.i.d. observations.","type":"model","stated":"explicit","quote":"To deal with time resolution, mismatch, we assume the underlying procedure follows Eq. 1 and we use the log return of each stock, formulated as returnt = log(pricet −pricet−1).","support":"full"},{"assumption":"Noise terms in the structural causal model are non‑Gaussian to enable the use of the Generalized Independent Noise (GIN) condition for direction identification.","type":"statistical","stated":"explicit","quote":"Here we aim at making use of Generalized Indepen-dent Noise (GIN) condition (Xie et al., 2020) to decide the directions, by assuming the noise terms ϵVi in Definition 1 are non-gaussian.","support":"full"},{"assumption":"Within each segment identified by change‑point detection the data are stationary; otherwise rank‑deficiency constraints would be invalid.","type":"data","stated":"explicit","quote":"5.2 Baysian Change Point Detection To utilize our rank-based latent causal discovery method we need to assume that our data is stationary; otherwise the nonstationary would induce rank deficiency constraints that cannot correctly reflect the underlying causal structure.","support":"full"},{"assumption":"Latent variables exist and act as common children of observed variables (e.g., X1‑X4 share a latent L1).","type":"model","stated":"explicit","quote":"9 Dong, Dai, Fan, Jin, Rajendran, and Zhang Example 2 Suppose we observe four variables X1, X2, X3, X4 as common children of a latent variable L1.","support":"full"},{"assumption":"Observed variables are jointly Gaussian when estimating causal coefficients, allowing a factor‑analysis‑like likelihood formulation.","type":"statistical","stated":"explicit","quote":"Similar to the typical factor analysis setting (Gorsuch, 2014), we assume that ϵV are all gaussian and thus X are jointly gaussian, and we find the causal coefficients A by maximizing the log-likelihood of observational data: arg max A,Ψ −n|X| 2 log 2π −n 2 log |ΣX| −n 2 tr((ΣX)−1 ˆΣX), (6) s.t., diagonal entries of ΣL are all 1, (7) where n is the number of i.i.d. observations of X, tr(˙) returns the trace of a","support":"full"},{"assumption":"The latest data are more informative for handling concept drift.","type":"market","stated":"implied","quote":"To handle concept drift, prior works usually rely on the assumption that the latest data contains more useful information.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"SP100 stock price data (log daily and hourly returns)","provider":null,"kind":"market-prices","access":"unknown","period":"2019-2023","universe":"S&P 100 constituents","quote":"4.4 The Whole Procedure and Potential Usage We will apply the proposed rank-based latent causal discovery method to SP100 stock price data, which suffers from the three demos - time resolution mismatch, non-stationarity, and the existence of latent variables. The CD-NOD analysis was conducted on these ten selected stocks, utilizing their log daily return data spanning from 2019 to 2023.","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"The rank‑based latent causal discovery algorithm successfully recovers the Markov equivalence class of the underlying causal graph despite the presence of latent variables.","kind":"theoretical","quote":"4.3 Determining All Directions and Estimating Causal Coefficients Our rank-based latent causal discovery algorithm is able to output the Markove equivalence class of Omin(Os(G)).","support":"full"},{"finding":"Empirical validation on SP100 data shows that the method identifies four stocks (PFE, BA, AAPL, NFLX) whose causal generating mechanisms change over time.","kind":"statistical","quote":"The discovered causal graph indicates four stocks whose causal generating mechanisms are subject to changes over time: Pfizer (PFE), Boeing (BA), Apple (AAPL), and Netflix (NFLX).","support":"full"},{"finding":"Applying the method to different time windows (2017, 2019, 2021) reveals how change points such as Covid‑19 affect the underlying causal graph.","kind":"performance","quote":"Our first experiment is to show how the change point, e.g., Covid-19, influences the underlying graph.","support":"partial"},{"finding":"The approach can handle all three demons—time‑resolution mismatch, non‑stationarity, and latent factors—simultaneously, as demonstrated in the empirical study.","kind":"theoretical","quote":"Based on this understanding we propose rigorous methods to address each issue as well as a systematic solution that works even when the three issues are present simultaneously. • Our empirical study validates the effectiveness of our proposed methods in real-life scenarios in finite sample cases.","support":"full"},{"finding":"Estimated causal coefficients exhibit meaningful relationships, e.g., high coefficient (0.84) between energy companies XOM and CVX, and shared latent structure among finance stocks BAC, USB, JPM.","kind":"statistical","quote":"This also implies that BAC, USB, JPM share something in common that is not shared by other financial stocks. As for the energy companies XOM and CVX, we observe that they are highly related with a high coefficient 0.84.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":21,"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":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"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":2,"extracted_at":"2026-11-14T19:59:40Z"},"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/"}