From an old albeit still relevant (and funny) American Journal of Agricultural Economics article (link opens a .pdf) by Wally Thurman and Mark Fisher:
The notion of Granger causality is simple: If lagged values of X help predict current values of both X and Y, then X is said to Granger cause Y. We implement this notion by regressing eggs on lagged eggs and lagged chickens; if the coefficients on lagged chickens are significant as a group, then chickens cause eggs. A symmetric regression tests the reverse causality. We perform the Granger causality tests using one to four lags. The number of lags in each equation is the same for eggs and chickens.
To conclude that one of the two “came first,” we must find unidirectional causality from one to the other. The test results indicate a clear rejection of the hypothesis that eggs do not Granger cause chickens. They provide no such rejection of the hypothesis that chickens do not Granger cause eggs. Therefore, we conclude that the egg came first.
What I like about Thurman and Fisher’s paper is that it neatly illustrates the point that for all the media talk of causality surrounding Christopher Sims’ work on Granger causation when he and Thomas Sargent won the Nobel prize for economics last year, Granger causation is not the same as actual causation.
(HT: Jeremy Petranka.)