Jim Devine wrote:
For the record, I did NOT reject time-series analysis as "crude empiricism." Instead, I wrote the following: >>> isn't ARMA nothing but crude empiricism, a "modern" version of factor analysis? <<<
No. For the record, you wrote: "that's [time-series analysis is] a dynamic form of what's called factor analysis. Nowadays, it's called time-series analysis, if I recall correctly. Crude empiricism." http://archives.econ.utah.edu/archives/pen-l/2007w28/msg00065.htm And that is a categorical assertion.
what's the biz, then, about lumping me with "you guys"? there are one or two dingbats on pen-l that I do not want to be associated with.
Again, you wrote this: "that's a dynamic form of what's called factor analysis. Nowadays, it's called time-series analysis, if I recall correctly. Crude empiricism." And, following that, Michael Perelman wrote: "Before I face the firing squad, let me first agree with much of what Jim Devine said." That's why I wrote "you guys." I wasn't attributing to you Michael's ideas. I was attributing to Michael agreement with *your* ideas. If anything, I should apologize to Michael -- since "agree with much" doesn't mean "agree totally with." (Sorry Michael.)
(Of course, to quote Marx, I don't want to belong to any club that would accept me as a member!)
Yes, Marx Twain.
huh? If I reject theory- minimization, why is this saying that "the amount of theory is irrelevant"? or that we should maximize it instead? I agree with neither of these.
If I should be considerate with the intent and context of your statements, the same rule should apply in the other direction. Please go back and re-read the text to which you replied. I was talking about inference, econometrics, time-series analysis. The starting point is a set of empirical observations, data. In drawing inferences, every extra postulate you impose on your data is -- logically -- a restriction on the estimate. Robustness means that, for a given amount of information extracted from the data, the theory (restrictions) imposed is (are) only the minimal necessary. That is theory minimization. It's equivalent to robustness. In inference, theory minimization makes sense and what you wrote doesn't.
Instead, the cost of using theory should be balanced against its benefits (and vice-versa).
Not "instead." But after the clause, what you say sounds right to me. The marginal cost and benefit of inference should be equalized. That means that, for a given amount of information extracted from the data, you are minimizing your cost function (the amount of theory, the loss of robustness). Or that, for a given amount of theory (robustness), you are maximizing your gain function (amount of information).