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

TL;DR

DoFlow is a flow-based generative model that provides accurate observational, interventional, and counterfactual predictions for multivariate time series, while also enabling anomaly detection and causal reasoning.

Contribution

It introduces DoFlow, a novel flow-based model over causal DAGs that unifies causal inference with generative time series forecasting, including counterfactuals.

Findings

Achieves accurate observational forecasting on synthetic and real data.

Enables causal forecasting for interventional and counterfactual queries.

Effectively detects anomalies using explicit likelihoods.

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 the broader goal of unifying causal reasoning and generative modeling for complex dynamical systems.