Paper: arXiv 2506.23619

Authors: Guillaume Coqueret, Martial Laguerre

Abstract

This paper investigates the impact of posterior drift on out-of-sample forecasting accuracy in overparametrized machine learning models. We document the loss in performance when the loadings of the data generating process change between the training and testing samples. This matters crucially in settings in which regime changes are likely to occur, for instance, in financial markets. Applied to equity premium forecasting, our results underline the sensitivity of a market timing strategy to sub-periods and to the bandwidth parameters that control the complexity of the model. For the average investor, we find that focusing on holding periods of 15 years can generate very heterogeneous returns, especially for small bandwidths. Large bandwidths yield much more consistent outcomes, but are far less appealing from a risk-adjusted return standpoint. All in all, our findings tend to recommend cautiousness when resorting to large linear models for stock market predictions.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • 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

Why this score: The paper features high math complexity with extensive theoretical derivations (e.g., Proposition 1, risk decompositions, large-dimensional asymptotics) and rigorous empirical analysis including Monte Carlo simulations, sub-period backtests, and sensitivity analysis on bandwidth parameters for equity premium forecasting.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Assess impact of posterior drift<br>on overparametrized model forecasting"] --> B["Methodology<br>Estimate linear models with varying bandwidths<br>Analyze out-of-sample forecasting accuracy"]
  B --> C["Data Input<br>Equity premium historical data<br>Train vs. Test samples with regime shifts"]
  C --> D["Computational Process<br>Simulate varying holding periods<br>(e.g., 15 years) across sub-periods"]
  D --> E["Key Finding 1<br>Small bandwidths show<br>highly heterogeneous returns"]
  D --> F["Key Finding 2<br>Large bandwidths offer consistency<br>but poor risk-adjusted returns"]
  E & F --> G["Conclusion<br>Recommend cautiousness using large<br>linear models for stock market prediction"]

Cited References (62)

  1. A penalized two-pass regression to predict stock returns with time-varying risk premia (2023)
  2. The term structure of equity risk premia (2021)
  3. Benign overfitting in linear regression (2020)
  4. Comment on `\textit T he Virtue of Complexity in Return Prediction ’ (2023)
  5. Optimal Inference for Spot Regressions (2024)
  6. Simplified: A Closer Look at the Virtue of Complexity in Return Prediction (2025)
  7. The Limited Virtue of Complexity in a Noisy World (2025)
  8. On the Appeal of Complexity (2023)
and 54 more
  1. Time-varying parameters as ridge regressions (2023)
  2. Double trouble in double descent: B ias and variance (s) in the lazy regime (2020)
  3. Predictive regressions with time-varying coefficients (2012)
  4. Unifying Grokking and Double Descent (2023)
  5. High-Dimensional Learning in Finance — → 2506.03780
  6. Pockets of predictability (2023)
  7. Time-varying risk premium in large cross-sectional equity data sets (2016)
  8. A survey on concept drift adaptation (2014)
  9. Consistent change-point detection with kernels (2018)
  10. On stable factor structures in the pricing of risk: do time-varying betas help or hurt? (1998)
  11. A comprehensive 2022 look at the empirical performance of equity premium prediction (2024)
  12. Simplicity versus Complexity: The Role of Historical Average in Kelly, Malamud, and Zhou’s (2024) RFF Model (2025)
  13. Surprises in High-Dimensional Ridgeless Least Squares Interpolation (2020)
  14. Surprises in high-dimensional ridgeless least squares interpolation (2022)
  15. The virtue of complexity in return prediction (2024)
  16. The statistics of Sharpe ratios (2002)
  17. Learning under concept drift: A review (2018)
  18. A linear adjustment-based approach to posterior drift in transfer learning (2024)
  19. Seemingly Virtuous Complexity in Return Prediction (2025)
  20. The S harpe R atio: S tatistics and Applications (2021)
  21. Grokking: G eneralization beyond overfitting on small algorithmic datasets (2022)
  22. Transfer Learning Meets Functional Linear Regression: No Negative Transfer Under Posterior Drift (2025)
  23. Can Machines Learn Weak Signals? (2025)
  24. Explaining grokking through circuit efficiency (2023)
  25. A comprehensive look at the empirical performance of equity premium prediction (2008)
  26. A short proof of the Marchenko–Pastur theorem (2016)
  27. Random Forests for Change Point Detection (2023) — link
  28. The Conditional CAPM and the Cross-Section of Expected Returns (1996)
  29. Change Point Estimation in High Dimensional Markov Random-Field Models (2016) — link
  30. Multiscale Change Point Inference (2014) — link
  31. High-dimensional change-point detection under sparse alternatives (2019)
  32. A survey of methods for time series change point detection (2017)
  33. A kernel multiple change-point algorithm via model selection (2019)
  34. Homogeneity and change-point detection tests for multivariate data using rank statistics (2015)
  35. Detecting “disorder” in multidimensional random processes (1981)
  36. An Efficient Two Step Algorithm for High Dimensional Change Point Regression Models Without Grid Search. (2019)
  37. High-dimensional asymptotics of prediction: Ridge regression and classification (2018)
  38. Spectral convergence for a general class of random matrices (2011) — link
  39. The nonstationarity-complexity tradeoff in return prediction (2025)
  40. Eigenvectors of some large sample covariance matrix ensembles (2011)
  41. Topics in random matrix theory (2023)
  42. Random matrix methods for machine learning (2022)
  43. Anisotropic local laws for random matrices (2017)
  44. Strong convergence of the empirical distribution of eigenvalues of large dimensional random matrices (1995)
  45. Conformal Prediction Under Generalized Covariate Shift with Posterior Drift (2025)
  46. On the limit of the largest eigenvalue of the large dimensional sample covariance matrix (1988)
  47. Limit of the smallest eigenvalue of a large dimensional sample covariance matrix (2008)
  48. A non-asymptotic theory of Kernel Ridge Regression: deterministic equivalents, test error, and GCV estimator (2024) — link
  49. Dimension free ridge regression (2022) — link
  50. Forecasting stock market returns: The sum of the parts is more than the whole (2011)
  51. Dimension-free deterministic equivalents for random feature regression (2024) — link
  52. Six lectures on linearized neural networks (2023) — link
  53. Asymptotics of ridge (less) regression under general source condition (2021)
  54. On the optimal weighted $l_2$ regularization in overparameterized linear regression (2020)

1 of 62 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.