Comments on: The Case Against VARs http://www.arnoldkling.com/blog/the-case-against-vars/ taking the most charitable view of those who disagree Mon, 21 Dec 2020 14:00:34 +0000 hourly 1 https://wordpress.org/?v=4.0.32 By: ScottA http://www.arnoldkling.com/blog/the-case-against-vars/#comment-407239 Sat, 22 Feb 2014 19:14:28 +0000 http://www.arnoldkling.com/blog/?p=2738#comment-407239 A lot of academics have the strange idea that making regressions more complex can eliminate the problem with regressions. Some of the problems can be mitigated (some problematic assumptions, distributional issues, auto-correlation, etc) but it’s always good to remind everyone that the basic nature of the process (non-experimental data with massive quantities of unknown or latent influences) can’t be mathematized-away. Any thoughts on simulation/agent-based modeling as a substitute? Particularly where grounded in observed data.

]]> By: Matt Young http://www.arnoldkling.com/blog/the-case-against-vars/#comment-389541 Mon, 10 Feb 2014 16:34:06 +0000 http://www.arnoldkling.com/blog/?p=2738#comment-389541 I can invent a term. Vector auto entropy. Replace time with probability of occurance. Replace regression with entropy. Works pretty well and is nothing but structural.

]]> By: S http://www.arnoldkling.com/blog/the-case-against-vars/#comment-389492 Mon, 10 Feb 2014 15:26:40 +0000 http://www.arnoldkling.com/blog/?p=2738#comment-389492 I use VARs to project loss severity and frequency for insurance exposure. Mostly it they me some kind of covariance to work with, and I have found that they are usually more skillful than someone just arm waving a selected value (which is what is usually done). There are exogenous factors that the models cannot possibly know about, so I try to do my best to adjust the data before the model is fit, that and, dummy variables help.

Anyway, I have no idea how broad your dislike of the method is, but I was curious if there was any instance where you thought they might be appropriate, or at least better than nothing?

]]>