Jim Devine wrote: > But the valid approach to econometrics is hypothesis testing -- not > this "let's see what fits" approach.
This is the kind of reaction I'm trying to warn young people against -- a reaction based on a superficial understanding of a technique. Computationally, VAR is a procedure to estimate the parameters of a time-series model. It's up to you to decide which parameters you estimate. So, of course you can do hypothesis testing with VAR. It depends on which parameters you decide to estimate. If you have enough degrees of freedom (data), then you can estimate the model "coefficients" and higher moments as well -- as high as you wish. Typically, economists estimate var-covar matrices as a matter of routine (which gives you the weights to test your coefficients). Hypothesis tests for each coefficient (and for the whole model) are default procedure. If you wish, call it "empiricism" (if you don't care about conceptual precision) -- but not "crude." But here's the way for people to decide on their own whether it is crude or not -- spend some time getting your head around the principles, then take a bunch of data and use it yourself! How can it harm you? By the way, I disagree with Krugman that these techniques let the data speak (although he may have put things that way to abbreviate). Obviously, there are always assumptions built in. But they are rather mild -- and the point is precisely relaxing them to the extent the data allow. The remarkable thing, IMO, is how much true information you can get from data thus digested. _______________________________________________ pen-l mailing list [email protected] https://lists.csuchico.edu/mailman/listinfo/pen-l
