Granger Causal Networks and Indirect Feedback

2026-07-08 · Source: Towards Data Science · Field: Finance & Economics — Economic Analysis & Policy, Capital Markets & Investment Management · Depth: Advanced, long

Summary

The article introduces Granger Causal Network Graphs to address a key limitation in econometric vector autoregressive (VAR) models: distinguishing direct, indirect, and aggregate feedback among endogenous variables. Traditional VAR studies struggle to isolate these impacts due to computational complexity. The author demonstrates building these causality network graphs, G(e,d), using real-world financial data, including log returns of the NASDAQ-100 ETF (QQQ), SPY, and technical indicators like RSI, pctB, Volume, and Range. A pairwise Granger Causal Network Graph (99% CI, lag=9) revealed significant causal relationships, such as RSI's effect on QQQ log returns. Comparing this with a VAR-based Granger Causal Network, the analysis highlights how intermediate feedback can obscure direct effects. For example, pctB's direct impact on QQQ.diff is offset by stronger intermediate paths, and QQQ.Volume's influence is primarily indirect. The article suggests future work on conditional Granger causality networks to uncover deeper structural information.

Key takeaway

For quantitative analysts or econometricians modeling financial time series, understanding the nuanced interplay of direct and indirect causal effects is crucial. You should consider employing Granger Causal Network Graphs to visualize and differentiate these feedback mechanisms, especially when traditional VAR models fall short in isolating structural relationships. This approach can reveal hidden intermediate paths, like how QQQ.Volume indirectly impacts QQQ.diff, guiding more robust model specification and potentially improving out-of-sample forecasting by moving beyond isolated pairwise causality.

Key insights

Granger Causal Network Graphs visualize direct and indirect causal relationships between time series variables, overcoming VAR model limitations.

Principles

• Correlation does not imply causation.
• Causality networks reveal direct and indirect feedback.
• Structural causality requires controlling intermediate effects.

Method

Construct G(e,d) using pairwise Granger causality, comparing restricted and unrestricted bivariate models via F-test with a p-value cutoff (e.g., 1%) and AIC for lag selection.

In practice

• Visualize financial time series causality (e.g., QQQ, SPY, RSI).
• Identify intermediate feedback paths between variables.
• Compare bivariate vs. VAR networks for structural insights.

Topics

• Granger Causality
• Causal Networks
• Vector Autoregression
• Econometrics
• Financial Time Series
• Structural Causality

Best for: AI Scientist, Data Scientist, Research Scientist, Domain Expert

Related on AIssential

• CausalPOI: Spatio-Temporal Graph-Based Causal Modeling for Cold-Start POI Check-in Forecasting — CausalPOI forecasts cold-start POI check-ins by modeling causal effects through spatio-temporal functional interaction graphs.
• Time Series Gaussian Chain Graph Models — A novel time series Gaussian chain graph model captures complex blockwise dependencies and causal relations using frequency-domain decomposition.
• RAVEN: A Regime-Aware Variable-context Expert Network for Financial Time Series Forecasting — RAVEN adaptively determines optimal temporal context for financial time series forecasting using a Mixture-of-Experts framework to handle non-stationarity.
• Causal Semantic Alignment for LLM-based Time Series Forecasting — CVAformer disentangles time series variables into invariant and dynamic components to improve LLM-based forecasting accuracy by mitigating confounding effects.
• Moment Matters: Mean and Variance Causal Graph Discovery from Heteroscedastic Observational Data — Separating mean and variance causal influences improves interpretability and intervention design in heteroscedastic systems.
• Neural Network Parameter-optimization of Gaussian Pre-marginalized Directed Acyclic Graphs — A new graphical structure and neural network duality enable robust parameter optimization for latent variable causal models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards Data Science.