Not A DSGE Model

I will try again to answer Larry Summers’s question “”What wouldn’t be a Dynamic-Stochastic General-Equilibrium model?” I won’t present a formal macroeconomic model in a blog but will try to describe what I might do if I got down to work.

I think the first step back to not DSGE starts with the simplest standard 3 equation New Keynesian DSGE model. That is, in this post, I will make lots of absurd simplifying assumptions to keep things very simple, even though I think the main point of not DSGE modeling is that one doesn’t have to make these assumptions (or even the weaker absurd assumptions in cutting edge NK DSGE models).

The assumptions are crazy. First there is no capital of any kind — people just produce goods by the sweat of their brows without using plant or equipment. There is no investment, There is also no Government consumption so aggregate demand is equal to private consumption. This is determined by intertemporal optimization yielding a first order condition or Euler equation which looks a lot like an old IS curve. Demand increases in the expected value of future GDP and decreases in the real interest rate. There is a micro founded forward looking Phillips curve so inflation increases in GDP and increases in expected future inflation.
Finally there is a Taylor rule with the nominal interest rate increasing more than one for one in lagged inflation and decreasing in lagged output.

This model doesn’t fit the data at all and is used mainly for teaching.

For an old Keynesian alternative model, I would stick with a Taylor rule, even though it doesn’t fit the data very well

I do not think the Euler equation for consumption (aslo known as the Permanent Income Hypothesis) has any useful role in macroeconomics. The simple equation is radically rejected by the data. It can be modified so that it no longer has the tested implications which have turned out to be false. They are basically all of the implications that have been tested. I don’t think it adds anything of value. I have discussed this at length here here, here, here here and in a long pdf here.

A model in which consumers have habits and are myopic fits the aggregate data better. As far as I know it explains all the stylized facts which convinced economists that the old Keynesian approach should be replaced first by the simple original PIH then by modified versions in which roughly one free parameter is added to fit every summary statistic. In this model, consumption is a linear function of current personal disposable income and a few lags of personal disposable income. I stress that the DSGE approach has lead economists to replace personal disposable income with GDP when explaining consumption. The data tend to suggest that this was moving away from the truth. A key difference between the PIH and older models (which I propose resurrecting) is that the PIH can imply Ricardian equivalence. There is almost no evidence of Ricardian effects let alone full Ricardian equivalence. Notably applied work measures the fiscal stance with the structural primary deficit — this is further from Ricardian equivalence even than the IS-LM model. Policy discussions often treat tax cuts and spending increases interchangeably — that is they have moved from IS-LM analysis in the direction opposite of academic models.

The forward looking Phillips curve is inconsistent with the data. To come close, economists add lagged inflatin terms talking vaguely about inflationary momentum and appealing to indexation. However, they insist that the true conditional mean of future inflation (rationally expected inflation) matters and should not have been left out by old Keynesians. I know of no evidence supporting this claim. Both survey expectations and TIPS breakevens are well fit by a simple autoregressive model. The equation does not fit achieved inflation at all well as it should if expectations were rational. The old approach fits actual data relative to expectations better than the new one. A Key episode which allegedly show the importance of a more sophisticated approach than the static autoregressive model — the effect of Volcker’s dramatic regime change, has an implication for a coefficient of the sign opposite the statistically significant coefficient estimate with actual data. The old Phillips curve, with coefficients on a few lags of inflation used to estimate expected inflation, works better than the new one. The alleged silly Phillips curve with no expected inflation effect was not, in fact used by old Keynesians.

The case against the old approach to the Phillips curve is that it had implausible implications for the effects of policies which were not being considered. Here Friedman argued that a model was not useful for evaluating policies within a given set, because it had implausible implications for policies not in that set (that is he said models must be true to be useful and that while he can use his methodology for positive economics, Keynesians can’t).

Now models which use the assumption of rational expectations do not have much effect on the policy debate. A Phillips curve subject to the extremely similar critiques of Friedman, Phelps, Lucas, Hicks, Samuelson, Solow and is used all the time.

However, it is not the one which fits the data well. It has been decided that the best crude simple Phillips curve relates unemployment and the acceleration of inflation — this means that a coefficient on lagged inflatin is set to one instead of being estimated with data. This means that policy makers are advised that the Natural rate of unemployment in Spain is over 20%. Current applied work has little to do with current academic research and has moved further from the data than 1960s Keynesian models were.

Unfortunately, bad as the NK approaches to modelling consumption and price formation are, they are much better than the approach to modeling investment. Current models use the ad hoc assumption that it is constly to adjust the rate of investment even though no one found this plausbile a priori. They do not distinguish between housing investment, non-residential fixed investment and inventory investment. They do not perform as well as the flexible accelerator (which says investment increases if real GDP growth is high and decreases if the real interest rate is high). I don’t think I have to argue that ignoring housing is problematic. It is also obviously extreme to Ignore inventories when modeling the business cycle.

I think the old approach of using flexible accelerators with different coefficients for housing, non residential fixed investment and inventories is better than the NK DSGE approach. For one thing, it would remind economists (other than Paul Krugman as usual) that interest rates affect aggregate demand mainly through residential investment and, to a much lesser extent through investment in plant but not statistically significantly through investment in equipment or inventories.

I guess I should write about net exports to finish off my proposal to resurrect old Keynesian macroeconomics, but I don’t have anything to write.