Granger Causality

Overview


Granger causality is a definition of causality proposed by Clive Granger in the study of economic data. Granger proposed that a time series {% X_t %} can be said to be causal to another time series {% Y_t %} if information about current and past values of {% X_t %} is useful in predicting value of {% Y_t %}.

Within an information theory perspective, one could assert that the mutual information of causally related variables is greater than zero.

Granger Test


The test that Granger proposed made some assumptions.

  • Both time series are covariance stationary
  • The optimal forecast is linear
The Granger test runs a regression of {% Y_t %} against lagged values of {% X_t %} and tests for signficance.
{% Y_t = \alpha + \beta_1 X_{t-1} + \beta_2 X_{t-2} + ... + \beta_n X_{t-n} %}
The test Granger proposed was the standard F-Test. That is, the F test tests the hypothesis that not all the {% \beta %}'s given above are zero.