More on Models

Of the predictions of the effects of the ARRA, the stimulus package, based on macroeconomic models, I have been most favorably impressed by those of Macroeconomic Advisers. I have found them among the most conservative and also noted that they seem to track actual measured results, both in GDP and employment, most closely. This morning Joel Prakken takes to the company’s blog to defend their predictions:

For starters, parties receiving stimulus funds are asked to file an estimate of the number of jobs saved or created with the money. When these estimates were tallied and audited by the government and press, it was clear the figures were problematic. There were just too many examples of nonsensical results. Perhaps, as some argued, the questions were not well worded. Just as likely, however, is that respondents didn’t really know the answer to the question. Furthermore, even if such estimates are accurate, they reflect only the direct effect on jobs, not the indirect or “multiplier” effects. The point is that it simply is not possible to measure directly (that is, by survey) the total effect on jobs. One needs to estimate the effect with an economic model that can be used to compare a pre-stimulus baseline with a simulation that includes all the effects of the stimulus. This is the only methodology that really makes sense to us.

As I have said before, I’m skeptical of this view. Measuring model results against model results is only meaningful insofar as the models are calibrated against actual results. That actual results are hard to come by doesn’t refute that view in any way nor does it render comparing model results to model results any more worthwhile or accurate.

Read the whole thing.