Methodological Insights into Structural Causal Modelling and Uncertainty-Aware Forecasting for Economic Indicators

TL;DR

The paper addresses robust causal understanding and uncertainty-aware forecasting in macroeconomic time series by integrating LPCMCI-based causal discovery with GPDC, and zero-shot probabilistic forecasting via Chronos. It reveals a strong Growth→GDP link, weak inflation connectivity, and pronounced unemployment autoregression, then demonstrates accurate one- and two-step-ahead unemployment forecasts with 90% predictive intervals that support anomaly detection. This combination enables policy-relevant interpretation and robust monitoring of regime shifts without task-specific training. The work highlights the value of fusing causal structure learning with probabilistic language-model-based forecasting to improve forecasting robustness and detection of structural breaks in economic indicators.

Abstract

This paper presents a methodological approach to financial time series analysis by combining causal discovery and uncertainty-aware forecasting. As a case study, we focus on four key U.S. macroeconomic indicators -- GDP, economic growth, inflation, and unemployment -- and we apply the LPCMCI framework with Gaussian Process Distance Correlation (GPDC) to uncover dynamic causal relationships in quarterly data from 1970 to 2021. Our results reveal a robust unidirectional causal link from economic growth to GDP and highlight the limited connectivity of inflation, suggesting the influence of latent factors. Unemployment exhibits strong autoregressive dependence, motivating its use as a case study for probabilistic forecasting. Leveraging the Chronos framework, a large language model trained for time series, we perform zero-shot predictions on unemployment. This approach delivers accurate forecasts one and two quarters ahead, without requiring task-specific training. Crucially, the model's uncertainty-aware predictions yield 90\% confidence intervals, enabling effective anomaly detection through statistically principled deviation analysis. This study demonstrates the value of combining causal structure learning with probabilistic language models to inform economic policy and enhance forecasting robustness.

Figures (8)

• Figure 1: Visualisation of the normalised time series analyised.
• Figure 2: Identified contemporaneous causal relationships between GDP, inflation, economic growth, and unemployment, see huang2019causal.
• Figure 3: Time-lagged causal graph representation of multivariate time series, visualised across four variables from time steps $t-3$ to $t$. Directed edges indicate statistically significant causal links, with arrow colour denoting the strength and direction of the multivariate conditional independence (MCI) measure computed using the partial correlation test with analytic significance.
• Figure 4: Time-lagged causal graph representation of multivariate time series, visualised across four variables from time steps $t-3$ to $t$. Directed edges indicate statistically significant causal links, with arrow colour denoting the strength and direction of the multivariate conditional independence (MCI) measure computed using GPDC.
• Figure 5: Zero-shot forecast of unemployment using Chronos. Blue denotes the actual values, orange the predicted mean, and the shaded area indicates the 90% prediction interval.