This blog has moved.

Redirecting you to in seconds.
Or, go now

Sunday, February 21, 2010

Macro Model Disagreements and Reality

Last Friday Macroeconomic Advisers (MA), a forecasting firm, posted a blog entry responding to empirical work by me and others on the “stimulus act” of 2009. I welcome the discussion, but unfortunately the blog seems to have missed the main points of my work along with John Cogan, Tobias Cwik, and Volker Wieland, which showed that (1) findings that the stimulus would have a large impact were not robust and (2) the data so far indicate that the actual effects on GDP or employment are not significant. Here is our working paper of a year ago and an oped.

First, we did not say or imply that “macro modelers such as MA were unaware of the ‘modern’ life-cycle (or permanent income) theory” or anything else of a personal nature. Instead, we showed that the temporary tax and transfer payments, which were a large part of the stimulus packages of 2008 and 2009, did not jump start consumption contrary to the claims of many. When you look at the evidence from 2009 (or 2008) you see that the theory holds up remarkably well. I am glad that the MA model incorporates that view.

Second, the MA blog claims that we relied on “small multipliers often found in reduced-form models” and that we used an “a-theoretical” approach. I see no way one could come to that conclusion from reading my work with Cogan, Cwik, and Wieland. Instead of reduced-form models we stressed the importance of using empirically-estimated structural models. We focused on the kind of models that have been taught in most graduate schools in economics in recent years and about which there has been a consensus in research, as emphasized by Michael Woodford of Columbia. In particular, we used a structural model estimated by Frank Smets, Director of Research at the European Central Bank, and his colleague Raf Wouters. We also looked at my own structural model, but we did not focus on that model so as to be more objective. Our structural approach predicted an impact of the stimulus which was only 1/6 what the administration claimed.

Third, the crowding out of investment and consumption by the stimulus in our analysis does not depend on the Fed increasing the interest rate in the short run. Our model simulations held the interest rate at zero for one or two years.

Fourth, the MA blog seems to miss the main point of my criticism as emphasized on Economics One for the past six months: There is no consensus among structural models about the forecasted impact of the stimulus: some models, such as ours, forecasted little or no impact; others models, like MA, forecasted a larger impact. The results of MA are thus not robust to different modeling assumptions. Menzie Chinn over at Econbrowser is right to say that “Something like [the Brookings Model comparison project of the 1980s] is desperately needed these days, so that those people who are concerned with seriously considering policy problems can sort out why the simulation results differ.” Indeed Volker Wieland has assembled a model database for exactly this reason.

In the meantime, I have argued that it is time to go beyond the models and see what actually happened rather than repeat the same model forecasting exercise over and over again with the path of government purchases, taxes, and transfers in the stimulus. In fact, many in the press perceive that these repeated forecast simulations are new evidence, even though they are essentially the same evidence provided before the stimulus was passed a year ago. I illustrated this misperception with a New York Times article from which I copied a chart from MA and other models. (Incidentally, this indirect reference to the Times’ MA chart is the only mention of MA in all our work). I have no disagreement with the MA blog about what the actual path of government spending, transfers, or taxes was. It has been very close to what Cogan, Cwik, Wieland and I assumed a year ago when we first circulated our analysis.

The MA blog does not refer to such evidence. For example, it does not analyze the contributions of the various components of real GDP growth in the past year to see if they are consistent with the stimulus having an effect. I have provided evidence using those numbers that government purchases did not contribute a noticeable amount to the change in real GDP growth. Likewise the evidence shows that the temporary tax changes or one-time transfers did not impact consumption or GDP significantly.