Statistics, experimentation, causal reasoning · insight-0030
Read "X Granger-causes Y" as "X helps forecast Y", never as "changing X changes Y"
- Evidence type
- documented fact
- Difficulty
- medium
- Roles
- Data scientist, Data analyst
- Industries
- aviation, finance, energy
Summary: situation, action, reason, boundary
Situation
A report claims competitor fares Granger-cause your bookings and proposes reacting to every competitor change.
Action
Use Granger tests for forecasting feature selection, and state them as predictive. For causal claims, name the intervention, the confounders and a design (experiment, quasi-experiment) that addresses them.
Reason
The Granger definition asks whether past X improves prediction of Y given past Y; it says nothing about what happens under intervention.
Boundary
Macro indicators, marketing spend and sales, sensor series and competitor monitoring.
Symptom or risk
Predictive precedence is mistaken for a causal lever; a common driver such as a holiday announcement can move both series with different lags.
Exceptions
Under strong, explicit assumptions (no unobserved common causes, correct lag structure), Granger results can support a causal argument, but the assumptions carry the weight.
Trade-offs
Predictive features are still valuable; just do not price decisions on them.
Pitfalls
Testing many lags and series pairs and reporting only the significant direction.
Lessons and concepts
Claims and their sources
Granger causality is defined by improvement in predicting one series from the past of another, which is predictive precedence rather than an intervention effect.