Granger causality is both "reliable" and "controversial." It is reliable in the sense that it tests for a relationship over time between two variables, that is, do changes in A precede in a regular way changes in B. Now, Granger tests do a good job of ferreting out such relationships, but, as numerous critics have pointed out, such temporal precedence even on a regular basis is far from being the same thing as "causality" in some true philosophical sense, although I am not going to attempt answer what that is here. Certainly one problem is still the standard old causation/causality problem with it even being possible that what follows ("truly") causes what precedes, a standard example being the cock crowing just before the sun rises (presumably the latter causes the former, to the extent there is any causation involved).
Two other problems are that there may be no such thing as causation (I'll leave that one alone) and that even if there is such an apparently regular temporal relationship observed rigorously by a Granger test, it may not remain in the future. Enough for now. Barkley Rosser -----Original Message----- From: Doug Henwood <dhenwood at panix.com> To: lbo-talk at lists.panix.com <lbo-talk at lists.panix.com> Date: Thursday, March 18, 1999 2:32 PM Subject: RE: Junk Econometrics [was School vouchers]
>Max Sawicky wrote:
>
>>Another common transgression is using correlation to justify a
>>particular direction of causation, notwithstanding the
>>availability of (admittedly imperfect) tests for causation.
>
>I keep running into so-called Granger tests. Are these reliable, or
>controversial?
>
>Doug
>