Interventional Time Series Priors for Causal Foundation Models
arXiv:2603.11090v1 Announce Type: new Abstract: Prior-data fitted networks (PFNs) have emerged as powerful foundation models for tabular causal inference, yet their extension to time series remains limited by the absence of synthetic data generators that provide interventional targets. Existing time series benchmarks generate observational data with ground-truth causal graphs but lack the interventional data required for training causal foundation models. To address this, we propose \textbf{CausalTimePrior}, a principled framework for generating synthetic temporal structural causal models (TSCMs) with paired observational and interventional time series. Our prior supports configurable causal graph structures, nonlinear autoregressive mechanisms, regime-switching dynamics, and multiple intervention types (hard, soft, time-varying). We demonstrate that PFNs trained on CausalTimePrior can perform in-context causal effect estimation on held-out TSCMs, establishing a pathway toward foundat
arXiv:2603.11090v1 Announce Type: new Abstract: Prior-data fitted networks (PFNs) have emerged as powerful foundation models for tabular causal inference, yet their extension to time series remains limited by the absence of synthetic data generators that provide interventional targets. Existing time series benchmarks generate observational data with ground-truth causal graphs but lack the interventional data required for training causal foundation models. To address this, we propose \textbf{CausalTimePrior}, a principled framework for generating synthetic temporal structural causal models (TSCMs) with paired observational and interventional time series. Our prior supports configurable causal graph structures, nonlinear autoregressive mechanisms, regime-switching dynamics, and multiple intervention types (hard, soft, time-varying). We demonstrate that PFNs trained on CausalTimePrior can perform in-context causal effect estimation on held-out TSCMs, establishing a pathway toward foundation models for time series causal inference.
Executive Summary
The article proposes CausalTimePrior, a framework for generating synthetic temporal structural causal models with paired observational and interventional time series. This framework enables the training of prior-data fitted networks for causal inference in time series data. The authors demonstrate the effectiveness of CausalTimePrior in performing in-context causal effect estimation, paving the way for foundation models in time series causal inference.
Key Points
- ▸ Introduction of CausalTimePrior, a framework for generating synthetic temporal structural causal models
- ▸ Enables training of prior-data fitted networks for causal inference in time series data
- ▸ Demonstrated effectiveness in performing in-context causal effect estimation
Merits
Flexibility
CausalTimePrior supports configurable causal graph structures and multiple intervention types
Realism
The framework generates synthetic data that mimics real-world time series data with nonlinear autoregressive mechanisms and regime-switching dynamics
Demerits
Complexity
The framework may be challenging to implement and require significant computational resources
Limited Generalizability
The effectiveness of CausalTimePrior may be limited to specific domains or types of time series data
Expert Commentary
The introduction of CausalTimePrior marks a significant advancement in the field of causal inference, particularly in the context of time series data. The framework's flexibility and realism make it an attractive solution for developing more accurate and robust predictive models. However, the complexity of the framework and its potential limitations in terms of generalizability must be carefully considered. Further research is needed to fully explore the potential of CausalTimePrior and its applications in various domains.
Recommendations
- ✓ Further research should be conducted to evaluate the effectiveness of CausalTimePrior in various domains and types of time series data
- ✓ The development of more user-friendly and accessible implementations of CausalTimePrior would facilitate its adoption in practice
