Granger Causality vs. Forecasting I am trying to explain Granger Causality to my coworkers and have this question. Granger causality tells you that variable $X$ provides helpful information about future values of $Y$, but that is what can also be inferred from any forecasting model. In your forecasting model if $X$ is significant and included in the model then that means $X$ provides helpful information about future values of $Y$. So what is delivered extra in the Granger causality model?
 A: 
Granger causality tells you that variable $X$ provides helpful information about future values of $Y$ <...>

The devil is in the details. Granger causality considers the incremental benefits on forecasting $Y$ due to using the history of $X$ extra to using the history of $Y$ alone. That is, the benchmark forecast is based on past values of $Y$ alone while the challenger forecast uses the past values of $X$ in addition. Meanwhile, if past values of $Y$ are excluded when making the prediction (which may or may not be the case based on the excerpt cited above) and $X$ helps predict $Y$, we cannot say much about Granger causality.

Here is Definition 1 from p. 428 of Granger's original paper "Investigating causal relations by econometric models and cross-spectral methods" (1969):

Let $U_t$ be all the information in the universe accumulated since time $t-1$ and let $U_t-Y_t$ denote all this information apart from the specified series $Y_t$.
Definition 1: Causality. If $\sigma^2(X|U)<\sigma^2(X|\overline{U-Y})$, we say that $Y$ is causing $X$, denoted by $Y_t \rightarrow X_t$. We say that $Y_t$ is causing $X_t$ if we are better able to predict $X_t$ using all available information than if the information apart from $Y_t$ had been used.

A: Granger causality is a weak form of causal inference, particularly when compared with Pearl-type causality and Bayesian networks. Forecasts, on the other hand, make no assumptions about possible causal relationships and, given that, demonstrate association for the purposes of prediction.
