### What's Useful about DSGE Models?

George Evans responds to the recent discussion on the usefulness of DSGE models:

Here is what I like and have found most useful about Dynamic Stochastic General Equilibrium (DSGE) models, also known as New Keynesian (NK), models. The original NK models were low dimensional – the simplest version reduces to a 3-equation model, while DSGE models are now typically much more elaborate. What I find attractive about these models can be stated in terms of the basic NK/DSGE model.

First, because it is a carefully developed, micro- founded model incorporating price frictions, the NK model makes it possible to incorporate in a disciplined way the various additional sectors, distortions, adjustment costs, and parametric detail found in many NK/DSGE models. Theoretically this is much more attractive than starting with a reduced form IS-LM model and adding features in an ad hoc way. (At the same time I still find ad hoc models useful, especially for teaching and informal policy analysis, and the IS-LM model is part of the macroeconomics cannon).

Second, and this is particularly important for my own research, the NK model makes explicit and gives a central role to expectations about future economic variables. The standard linearized three-equation NK model in output, inflation and interest rates has current output and inflation depending in a specified way on expected future output and inflation. The dependence of output on expected future output and future inflation comes through the household dynamic optimization condition, and the dependence of inflation on expected future inflation arises from the firm’s optimal pricing equation. The NK model thus places expectations of future economic variables front and center, and does so in a disciplined way.

Third, while the NK model is typically solved under rational expectations (RE), it can also be viewed as providing the temporary equilibrium framework for studying the system under relaxations of the RE hypothesis. I particularly favor replacing RE with boundedly rational adaptive learning and decision-making (AL). Incorporating AL is especially fruitful in cases where there are multiple RE solutions, and AL brings out many Keynesian features of the NK model that extend IS-LM. In general I have found micro-founded macro models of all types to be ideal for incorporating bounded rationality, which is most naturally formulated at the agent level.

Fourth, while the profession as a whole seemed to many of us slow to appreciate the implications of the NK model for policy during and following the financial crisis, this was not because the NK model was intrinsically defective (the neglect of financial frictions by most, though not all DSGE modelers, was also a deficiency in most textbook IS-LM models). This was really, I think, because many macro economists using NK models in 2007-8 did not fully appreciate the Keynesian mechanisms present in the model.

However, many of us were alert to the NK model fiscal policy implications during the crisis. For example, in Evans, Guse and Honkapohja (“Liquidity traps, learning and stagnation,” 2008, European Economic Review), using an NK model with multiple RE solutions because of the liquidity trap, we showed, using the AL approach to expectations, that when there is a very large negative expectations shock, fiscal as well as monetary stimulus may be needed, and indeed a temporary fiscal stimulus that is large enough and early enough may be critical for avoiding a severe recession or depression. Of course such an argument could have been made using extensions of the ad hoc IS-LM model, but my point is that this policy implication was ready to be found in the NK model, and the key results center on the primacy of expectations.

Finally, it should go without saying that NK/DSGE modeling should not be the one and only style. Most graduate-level core macro courses teach a wide range of macro models, and I see a diversity of innovations at the research frontier that will continue to keep macroeconomics vibrant and relevant.