The Great Multiplier Debate, New Keynesian Edition

Greg Mankiw cites a study by Cogan, Cwik, Taylor, and Wieland to buttress his arguments that fiscal multipliers are small, especially when considering New Keynesian models. He also provides a startling graphic showing the dynamic multipliers from Romer-Bernstein versus the Taylor (1993) model, incorporating model consistent expectations; this graphic motivates Wieland et al. to remark:

We first show that the assumptions made by Romer and Bernstein about monetary policy — essentially an interest rate peg for the Federal Reserve — are highly questionable according to new Keynesian models. We therefore modify that assumption and look at the impacts of a permanent increase in government purchases of goods and services in the alternative model. According to the alternative model the impacts are much smaller than those reported by Romer and Bernstein.

Cogan et al. use a New Keynesian dynamic stochastic general equilibrium (DSGE) model, specifically the Smets-Wouter model (Working Paper version of AER paper here).

In contrast, here’s some remarks based on results from the IMF’s New Keynesian DSGE (The Case for Global Fiscal Stimulus).

…the effect on U.S. GDP of investment expenditures is 3.9 when there is global fiscal expansion and only 2.4 when the United States acts alone. Similarly, the effect on Japanese GDP of targeted transfers is 1.5 when there is global fiscal expansion and only 1.0 when Japan acts alone. Differences in multipliers across regions relate to the size of leakages in the different areas, including leakages into saving and imports. [italics added]

Here’s a graph summarizing what happens when monetary policy is held fixed for two years, and then is allowed to revert to a Taylor rule thereafter.

gimfmult.gifFigure 3 from The Case for Global Fiscal Stimulus. Assumes monetary accomodation in the two years of the fiscal stimulus, and stimulus equal to 1 ppt of GDP in year one, and 0.5 ppt of GDP in year two.The obvious question — why the disjuncture? They’re both New Keynesian DSGEs? (documentation for the IMF model, GIMF, here).

While I don’t dispute the conclusion that in the Smets-Wouter model implies small fiscal multipliers when monetary policy follows more Taylor-rules, I do dispute the conclusion that this is a general conclusion relevant for all New Keynesian DSGEs. After all, in the IMF simulations, monetary policy reverts back to a Taylor rule after two years, and yet they obtain large multipliers.

Hence, I think New Keynesian DSGEs — like any other large models incorporating many equations — can yield differing results depending on the assumptions made, some of which might seem inconsequential or conventional; what I have in mind include those assumptions regarding the nature of asset markets. (The Smets-Wouter model has a completely standard assumption of complete markets, a generic one-period bond that all households can access. The IMF model incorporates an overlapping generations framework. The differences in the two approaches yield drastically different implications for fiscal policy, as has been discussed in the context of fiscal consolidation — see Chinn (2005).)

Now, I don’t believe that DSGE’s are useless for policy analysis, as Willem Buiter has argued. And I count myself a New Keynesian. But I also believe that just because a model has microfoundations, intertemporal optimization, model consistent expectations, and so forth, doesn’t mean it necessarily provides more plausible estimates. It matters what assumptions are made (is “complete asset markets” a good assumption, for instance? And if they were three years ago, are they still now? And I’ll bet the proportion of “rule of thumb consumers” or “liquidity constrained consumers” is probably higher now than three years ago…).

Link to previous installment of The Great Multiplier Debate.

Originally published at Econbrowser and reproduced here with the author’s permission.