Q1. You can use either values (P or F). The null is there is no Granger causality (X does not cause Y, please see below). An F test can be used to determine whether the null is valid. If the F statistics is greater than the table value, the null is rejected (X does cause Y).
Yt = a + b1*Yt-1 + b2*Yt-2 + ... + c1Xt-1 + c2Xt-2 + ...
Null: All the cj = 0, j = 1 …k
Q2. Follow 第 2 楼, AIC or SBC can be used to determine the number of lags. But more importantly, you need to use LM test to make sure there is no autocorrelation exists in the residuals.
Q3. I can not answer it since no graph is presented.
Send me an email if you have any further questions.