March 10th, 2010, 10:55 pm
It is not quite as simple as that.In essence Granger causality involves building a (linear) model to predict A based on past values of B, then building a model to predict B based on past values of A, and finally comparing these two models.The model building requires that a human being (econometrician) make some decisions (eg. how may lags are you going to use in each model, are you going to use levels or differences, raw data or logs) and, importantly, review the fitted equations to make sure the models are reasonably satisfactory. It would be difficult to completely automate this process IMHO.I other words it is not like a Student t test where you just plug some number into a formula, there is a small but nonzero element of judgement/skill involved. You must have experience in fitting equations to data.