What if Tomorrow is the World Cup Final? Counterfactual Time Series Forecasting with Textual Conditions

Time series forecasting has become increasingly critical in real-world scenarios, where future sequences are influenced not only by historical patterns but also by forthcoming events. In this context, forecasting must dynamically adapt to complex and stochastic future conditions, which introduces fundamental challenges in both forecasting and evaluation. Traditional methods typically rely on historical data or factual future conditions, while overlooking counterfactual scenarios. Furthermore, many existing approaches are restricted to simple structured conditions, limiting their ability to generalize to the real-world complexities. To address these gaps, we introduce the task of counterfactual time series forecasting with textual conditions, enabling more flexible and condition-aware forecasting. We propose a comprehensive evaluation framework that encompasses both factual and counterfactual settings, even in the absence of ground truth time series. Additionally, we present a novel text-attribution mechanism that distinguishes mutable from immutable factors, thereby improving forecast accuracy under sophisticated and stochastic textual conditions.

Time series forecasting is widely used to predict future trends in areas such as energy, transportation, healthcare, and finance. However, real-world futures are often shaped by upcoming events that may change what happens next. For example, a traffic forecast may depend on whether a concert is canceled, or an energy forecast may change because of unexpected weather conditions. Existing forecasting methods usually focus only on what actually happened in the past and struggle to handle “what-if” situations or complex real-world descriptions. In this work, we introduce a new forecasting setting that allows models to predict future time series under different possible future events described in natural language. We also propose a new way to evaluate these forecasts, including situations where the true future outcome is unavailable. In addition, we design a method that helps the model identify which factors in a text description are likely to change and which are fixed, leading to more accurate and flexible predictions under uncertain future conditions.