On Wed, 15 Sep 2010, Claudio wrote: > This is a stupid question: how to perform a Granger-Sims > causality in R?
I presume you mean, in gretl. > The GS causality test is like the Granger causality, but > includes not only lags, but also leads of the variable... As Sims (1972) puts it, "If and only if causality runs one way from current and past values of some list of exogenous variables to a given endogenous variable, then in a regression of the endogenous variable on past, current and future values of the exogenous variables, the future values of the exogenous variables should have zero coefficients." You can't specify leads (negative lags) with gretl's VAR command, but VARs are just estimated via OLS so you can easily do this in gretl. Here, for example, is a Sims test for one-way causality of US car sales (QNC) by US consumers' income (INCOME): open data9-7 list Xlags = INCOME(-1 to -4) list Xleads = INCOME(+1 to +4) ols QNC 0 Xlags INCOME Xleads omit Xleads --quiet "omit" gives F(4,46) = 0.263 with p-value 0.90, so the hypothesis of one-way causality (all the leads of INCOME have zero coefficients) is not rejected. Allin Cottrell
