Paper: arXiv 2408.09960

Authors: Daniel Cunha Oliveira, Yutong Lu, Xi Lin, Mihai Cucuringu, Andre Fujita

Abstract

We introduce a novel framework to financial time series forecasting that leverages causality-inspired models to balance the trade-off between invariance to distributional changes and minimization of prediction errors. To the best of our knowledge, this is the first study to conduct a comprehensive comparative analysis among state-of-the-art causal discovery algorithms, benchmarked against non-causal feature selection techniques, in the application of forecasting asset returns. Empirical evaluations demonstrate the efficacy of our approach in yielding stable and accurate predictions, outperforming baseline models, particularly in tumultuous market conditions.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)

Why this score: The paper employs advanced causal inference theory and invariant prediction frameworks, with substantial mathematical derivations. It provides a comprehensive empirical evaluation on real financial data (SPY and macroeconomic indicators, 2000-2022), comparing multiple causal discovery algorithms against baselines and assessing economic performance through a trading strategy.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Causal Discovery for<br>Asset Return Forecasting"] --> B["Data Inputs:<br>Equities & Financial Indicators"]
  B --> C["Methodology:<br>Causal Discovery vs.<br>Non-Causal Feature Selection"]
  C --> D["Computational Process:<br>Causality-Inspired<br>ML Forecasting Models"]
  D --> E{"Key Findings:<br>Superior Stability & Accuracy<br>Under Distributional Changes"}

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a novel causality‑inspired framework for financial time series forecasting that balances invariance to distributional changes with prediction error minimization.

Source sentence
We introduce a novel framework to financial time series forecasting that leverages causality‑inspired models to balance the trade‑off between invariance to distributional changes and minimization of prediction errors.

Assumes 4

  • Causal invariance: there exists a set of variables S* such that the conditional distribution of the outcome given S* is consistent across all environments. statistical
    Source sentence
    The key assumption is causal invariance: there exists a set of variables S∗such that the conditional distribution of the outcome given S∗is consistent in all environments e, h ∈E
  • No unmeasured confounder affects the treatment‑outcome relationship. model
    Source sentence
    The method also assumes no unmeasured confounder affects the treatment‑outcome relationship.
  • The study adheres to linear models for causal future selection. model
    Source sentence
    As we focus more on the performance of causal future selection, we adhere to linear models in this study
  • SPY price data are taken at the last business day of each month and FRED‑MD data at the first business day, implying alignment of monthly observations. data
    Source sentence
    The SPY prices data are available on the last business day of every month, whereas the FRED‑MD data is on the first business.

Data

  • SPY · Bloomberg (WRDS) market-prices commercial
    2000‑02‑02 to 2023‑04‑28 · SPDR S&P 500 ETF (SPY)
    Source sentence
    dataset contains monthly closing prices of SPY from February 2nd, 2000, to April 28th, 2023, and is collected from Bloomberg.
  • FRED‑MD · Federal Reserve Economic Data (FRED) macro public
    1960‑01‑01 to 2022‑12‑30 · U.S. macroeconomic indicators
    Source sentence
    dataset ... begins on January 1st, 1960, and concludes on December 30th, 2022.

Recognised sources: FRED Bloomberg WRDS

Costs

Unclear - sources disagree The platform classifier and the paper-text extraction give different answers, so no value is asserted.

Platform classifier: Not modeled

Paper text extraction: Not applicable

Finds 5

  • Modified VAR‑LiNGAM consistently achieves the lowest MAE and RMSE across both regular and crisis periods. performance
    Source sentence
    First, the Modified VAR–LiNGAM consistently outperforms (lowest prediction error) other methods in terms of MAE and RMSE, regardless of the period.
  • Causal methods (e.g., Modified VAR‑LiNGAM, Dynotears) have lower RMSE than the non‑causal benchmark SFS‑Linear during crises. predictive-accuracy
    Source sentence
    Over the GFC and COVID‑19 pandemic, the SFS‑linear method consistently shows higher RMSE than causal methods.
  • Dynotears yields lower RMSE than SFS‑Linear in crisis periods, highlighting the advantage of invariant prediction. predictive-accuracy
    Source sentence
    The bottom plot in Figure 1 highlights the difference in RMSE between SFS‑Linear and Dynotears. Values above 0 indicate that Dynotears yields a lower RMSE than SFS‑Linear.
  • During stable periods, causal methods perform similarly to the benchmark, indicating comparable efficiency when distributions do not shift. performance
    Source sentence
    During regular periods with stable economic conditions, the prediction accuracy of causal methods is similar to the benchmark.
  • Causal feature selection models exhibit smaller increases in MAE from regular to crisis periods compared to the non‑causal SFS‑Linear model. statistical
    Source sentence
    The last column of Table 1 depicts the MAE increase in percentage terms from regular periods to crisis periods

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 32 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 3 papers in the archive (3 by independent authors)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (114)

  1. Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance (2005)
  2. Root mean square error (RMSE) or mean absolute error (MAE)?—Arguments against avoiding RMSE in the literature (2014)
  3. The Sharpe Ratio (1994)
  4. The Average Drawdown of the Brownian Motion (2004)
  5. Maximum drawdown and the Kelly criterion (2000)
  6. Performance measurement in a downside risk framework (1994)
  7. Discovering Causal Signals in Images (2016) — link
  8. CASTLE: Regularization via Auxiliary Causal Graph Discovery (2020)
and 106 more
  1. Causal Regularization (2019) — link
  2. New Introduction to Multiple Time Series Analysis (2005) — link
  3. FRED-MD: A Monthly Database for Macroeconomic Research (2016)
  4. Inferring causation from time series in Earth system sciences (2019) — link
  5. Spurious Correlation: A Causal Interpretation (1954) — link
  6. Causal inference by using invariant prediction: identification and confidence intervals (2016)
  7. Invariant Causal Prediction for Sequential Data (2019) — link
  8. Elements of Causal Inference: Foundations and Learning Algorithms (2017)
  9. Invariance, Causality and Robustness (2018) — link
  10. Causal inference with multiple time series: principles and problems (2013) — link
  11. Causal network reconstruction from time series: From theoretical assumptions to practical estimation. (2018) — link
  12. Causal discovery and inference: concepts and recent methodological advances (2016) — link
  13. Introduction to the foundations of causal discovery (2017) — link
  14. Time-series forecasting with deep learning: a survey (2021) — link
  15. Empirical Asset Pricing via Machine Learning (2020)
  16. Analysis of financial time series (2005) — link
  17. Causal Machine Learning: A Survey and Open Problems (2022) — link
  18. Causal Structure Learning (2017) — link
  19. Causation, Prediction, and Search (2000)
  20. Probabilistic Graphical Models (2009)
  21. Causality (2009)
  22. Structural Equations with Latent Variables (1989)
  23. Forecasting stock returns (2013)
  24. The distribution of the flora in the alpine zone. 1 (1912)
  25. A linear non-Gaussian acyclic model for causal discovery. (2006)
  26. A comprehensive look at the empirical performance of equity premium prediction (2008)
  27. Dividend yields are equity risk premiums (1984)
  28. The dividend-price ratio and expectations of future dividends and discount factors (1988)
  29. Efficient tests of stock return predictability (2006)
  30. Trade co-occurrence, trade flow decomposition, and conditional order imbalance in equity markets (2022) — link
  31. Order imbalance and individual stock returns: Theory and evidence (2004)
  32. Predicting excess stock returns out of sample: Can anything beat the historical average? (2008)
  33. Getting the most out of macroeconomic information for predicting excess stock returns (2016)
  34. Macroeconomic factors and the correlation of stock and bond returns (2002)
  35. Macroeconomic factors do influence aggregate stock returns (2002)
  36. Order imbalance, liquidity, and market returns (2002)
  37. Foundations of technical analysis: Computational algorithms, statistical inference, and empirical implementation (2000)
  38. Stock market prices do not follow random walks: Evidence from a simple specification test (1988)
  39. Naive trading rules in financial markets and wiener-kolmogorov prediction theory: A study of" technical analysis" (1991)
  40. Optimal structure identification with greedy search (2002)
  41. Review of causal discovery methods based on graphical models (2019)
  42. An algorithm for fast recovery of sparse causal graphs (1991)
  43. Causation, prediction, and search (2000)
  44. Learning high-dimensional directed acyclic graphs with latent and selection variables (2012)
  45. Deep neural networks, gradient-boosted trees, random forests: Statistical arbitrage on the S&P 500 (2017)
  46. Literature review: Machine learning techniques applied to financial market prediction (2019)
  47. Machine learning techniques and data for stock market forecasting: A literature review (2022)
  48. Macroeconomics and reality (1980)
  49. Hypothesis testing in time series analysis (1951)
  50. Stock price prediction using the ARIMA model (2014)
  51. Vector autoregressions (2001)
  52. Graph similarity learning for change-point detection in dynamic networks (2024) — link
  53. Cross-impact of order flow imbalance in equity markets (2023) — link
  54. Causal discovery and inference: concepts and recent methodological advances (2016) — link
  55. A direct method of nonparametric measurement selection (1971)
  56. Trade co-occurrence, trade flow decomposition and conditional order imbalance in equity markets (2024)
  57. Empirical properties of asset returns: stylized facts and statistical issues (2001)
  58. Technical analysis and stock return predictability: An aligned approach (2018)
  59. Stock return predictability: Is it there? (2007)
  60. Predicting stock returns using industry-relative firm characteristics (2000)
  61. Book-to-market, dividend yield, and expected market returns: A time-series analysis (1997)
  62. Asset returns and inflation (1977)
  63. Forecasting the equity risk premium: the role of technical indicators (2014)
  64. Causal Factor Investing: Can Factor Investing Become Scientific? (2023)
  65. Correlation-based feature selection for machine learning (1999)
  66. Forecasting crude oil prices with a large set of predictors: Can LASSO select powerful predictors? (2019)
  67. Sparse signals in the cross-section of returns (2019)
  68. Combining multiple feature selection methods for stock prediction: Union, intersection, and multi-intersection approaches (2010)
  69. The adaptive selection of financial and economic variables for use with artificial neural networks (2004)
  70. A causal feature selection algorithm for stock prediction modeling (2014)
  71. LASSO: A feature selection technique in predictive modeling for machine learning (2016)
  72. Independent component analysis: algorithms and applications (2000) — link
  73. Nonlinear causal discovery with additive noise models (2008)
  74. Causality discovery with additive disturbances: An information-theoretical perspective (2009)
  75. Extensions of ICA for causality discovery in the Hong Kong stock market (2006)
  76. On the identifiability of the post-nonlinear causal model (2012) — link
  77. Survey and evaluation of causal discovery methods for time series (2022)
  78. The probability approach in econometrics (1944)
  79. Autonomy (1989)
  80. The logic of causal inference: Econometrics and the conditional analysis of causation (1990)
  81. Models, reasoning and inference (2000)
  82. Identifying the consequences of dynamic treatment strategies: A decision-theoretic overview (2010)
  83. Detecting and quantifying causal associations in large nonlinear time series datasets (2019)
  84. Dynotears: Structure learning from time-series data (2020)
  85. Estimation of a structural vector autoregression model using non-gaussianity. (2010)
  86. The M3-Competition: results, conclusions and implications (2000)
  87. The M4 Competition: Results, findings, conclusion and way forward (2018)
  88. A hybrid method of exponential smoothing and recurrent neural networks for time series forecasting (2020)
  89. An Analysis of the Adaptation Speed of Causal Models (2020) — link
  90. Recognition in Terra Incognita (2018)
  91. ``Why should I trust you?’’ Explaining the predictions of any classifier (2016)
  92. Testing for causality: A personal viewpoint (1980)
  93. Investigating causal relations by econometric models and cross-spectral methods (1969)
  94. On causal discovery from time series data using FCI (2010)
  95. Causal structure learning from multivariate time series in settings with unmeasured confounding (2018)
  96. A tutorial on spectral clustering (2007)
  97. Normalized cuts and image segmentation (2000)
  98. On spectral clustering: Analysis and an algorithm (2002)
  99. SPONGE: A generalized eigenproblem for clustering signed networks (2019)
  100. Co-trading networks for modeling dynamic interdependency structures and estimating high-dimensional covariances in US equity markets — → 2302.09382
  101. Detecting Lead-Lag Relationships in Stock Returns and Portfolio Strategies (2023)
  102. Correlation Matrix Clustering for Statistical Arbitrage Portfolios (2023)
  103. Robust Detection of Lead-Lag Relationships in Lagged Multi-Factor Models — → 2305.06704
  104. Dynamic Time Warping for Lead-Lag Relationships in Lagged Multi-Factor Models — → 2309.08800
  105. k-means++: The advantages of careful seeding (2006)
  106. Order-independent constraint-based causal structure learning. (2014)

3 of 114 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.