The Impossible Trinity of Time-Series Validation: A Conservation Law among Training Sufficiency, Test Coverage, and Temporal Causality

Validating a model on a time series asks for three things at once: each training run should use most of the sample (sufficiency), the test sets should together cover most of the sample (coverage), and training data should come before test data (causality). We prove that the three cannot be had together and price each one. Let $α$ be the smallest training fraction over folds, $β$ the fraction of the sample covered by tests, $Λ$ the fraction of the sample used as training data from the future of a test point, and $δ$ the distance from a test point to the nearest training point in its future. Every scheme on a sample of length $T$ satisfies $α+β\le 1+Λ$ and $α+\min\{β,δ/T\} \le 1$, and under $β$-mixing the leakage bias at a test point is at most $2Mβ_{\mathrm{mix}}(δ)$. In words: going beyond the causal frontier $α+β=1$ requires training on the future; that future data must sit within $(1-α)T$ of a test point; and its harm depends on its distance, not its amount. Hence expanding walk-forward is exactly the Pareto frontier of causal validation, $k$-fold cross-validation buys the most future data, and purged $k$-fold with an embargo pays in distance instead, which is cheap when the process forgets quickly but cannot repair the part of causality demanded by non-stationarity. On pure noise, shuffled 5-fold reports an information coefficient of $+0.32$, while contiguous 5-fold, using the same amount of future data, reports $+0.004$.

1. 15 May 2024

   TKAN: Temporal Kolmogorov-Arnold Networks

   The article presents Temporal Kolomogorov-Arnold Networks (TKANs), a new neural network design that merges the benefits of Recurrent Neural Networks and Long Short-Term Memory for improved multistep time series forecasting.

   SSRN

   189cites
2. 21 Feb 2024

   Robust agents learn causal world models

   The research suggests that intelligent agents must learn an approximate causal model to generalize to new domains, impacting fields like transfer learning and causal inference.

   Machine learningIn International Conference on Learning Representations

   102cites
3. 2 Feb 2026

   Directional-shift Dirichlet ARMA models for compositional time series with structural break intervention

   The article introduces a new Bayesian model that analyzes compositional time series data, effectively handling structural breaks and enhancing forecasting accuracy during these changes.

   arXivIn International Journal of Forecasting

   4cites
4. 27 Oct 2025

   FinCARE: Financial Causal Analysis with Reasoning and Evidence

   KG+LLM for Financial Causal Discovery: Combines SEC knowledge graphs, LLM reasoning, and causal discovery to build more accurate finance‑grounded causal models.

   arXiv

   1cites
5. 13 Sep 2025

   FinZero: Launching Multi-modal Financial Time Series Forecast with Large Reasoning Model

   Multi-modal Forecasting: FinZero, a pre-trained model fine-tuned by the Uncertainty-adjusted Group Relative Policy Optimization method, is introduced in the article, enhancing the accuracy, adaptability, and scalability of financial time series forecasting.

   arXiv

   5cites
6. 29 Aug 2025

   FinCast: A Foundation Model for Financial Time-Series Forecasting

   Time-Series Forecasting Model: FinCast, a new model for financial time-series forecasting, outperforms existing methods by effectively capturing diverse patterns without needing domain-specific adjustments.

   arXivIn Proceedings of the 34th ACM International Conference on Information and Knowledge Management

   19cites