Granger causality
Test whether one history improves another forecast
Granger causality asks whether past values of one variable improve prediction of another conditional on the specified information set. It is usually tested through restrictions on lag coefficients in a time-series model. Direction, lag length and conditioning variables define the claim. Cointegrated systems require an appropriate error-correction treatment, including potential long-run predictive channels.
Choose this for directional predictive questions in regularly observed time series when relevant histories can be aligned. Use a lag structure consistent with timing and a model appropriate for stationarity or cointegration.
Strengths
- Gives a precise test of incremental predictive information
- Can distinguish directional and feedback patterns
Limitations
- Omitted common causes can create predictive associations
- Temporal aggregation and lag selection can change apparent direction
Know the boundary
Granger causality is predictive precedence within an information set. It does not by itself establish an intervention effect or rule out confounding.