Comment by photon137

Even if you're looking for causation, detrending is usually necessary to obtain consistent estimators (in statistical terms). For example, Granger causality works with two stationary timeseries:

https://en.wikipedia.org/wiki/Granger_causality