[2511.02137] DoFlow: Flow-based Generative Models for Interventional and Counterfactual Forecasting on Time Series

Statistics > Machine Learning arXiv:2511.02137 (stat) [Submitted on 4 Nov 2025 (v1), last revised 1 Mar 2026 (this version, v2)] Title:DoFlow: Flow-based Generative Models for Interventional and Counterfactual Forecasting on Time Series Authors:Dongze Wu, Feng Qiu, Yao Xie View a PDF of the paper titled DoFlow: Flow-based Generative Models for Interventional and Counterfactual Forecasting on Time Series, by Dongze Wu and 2 other authors View PDF HTML (experimental) Abstract:Time-series forecasting increasingly demands not only accurate observational predictions but also causal forecasting under interventional and counterfactual queries in multivariate systems. We present DoFlow, a flow-based generative model defined over a causal Directed Acyclic Graph (DAG) that delivers coherent observational and interventional predictions, as well as counterfactuals through the natural encoding-decoding mechanism of continuous normalizing flows (CNFs). We also provide a supporting counterfactual recovery theory under certain assumptions. Beyond forecasting, DoFlow provides explicit likelihoods of future trajectories, enabling principled anomaly detection. Experiments on synthetic datasets with various causal DAG structures and real-world hydropower and cancer-treatment time series show that DoFlow achieves accurate system-wide observational forecasting, enables causal forecasting over interventional and counterfactual queries, and effectively detects anomalies. This work contributes to ...