Repairing Macroeconomics

Here are some vague thoughts on the state of academic macroeconomic research and how it might be improved*. The current mainstream macroeconmic research program has recently been severely criticized by so many extremely emminent econonmists, that it is hard to consider it the main stream. However, the many efforts to present an alternative approach (including this one) are vague and tentative. I am one of those who think a good first step is to go back to the old Keynesian research program, to research done 40 or more years ago, and pick up where Tobin, Solow and Samuelson left off.

To publish, researchers need to address questions which are neither too easy nor too hard. Answering an easy question is not significant original research, while answering a very difficult question is very difficult. I think macroeconmic research has suffered because the questions to be addressed are either too easy or too hard. I think it is very hard to answer any question about behavior which depends on long term expectations. I think it is too easy to understand macroeconomic issues which don’t depend on long term expectations. This is the view expressed 80 years ago in “The General Theory of Employment, Interest, and Money”. I think subsequent research has tended to demonstrate it to be valid.

The appeal of assuming rational expectations is that models with rational expectations are tractable but not trivial. I think the resulting advantages for the academic made the assumption irresistable. Unfortunately, I think it leads to a sterile research program. I think it might be wise for all macroeconomists to try at least once to understand economic aggregates without even any appeal the rational expectations assumption. I think this is a good way to realize how central it is. This isn’t my first attempt to do so, but I aim here to consistently avoid the concept of model consistent expectations.

Excessively Easy Problems

I think the too easy problems include the determination of aggregate consumption and short term fluctations in employment. The old model of consumption is that it is determined mostly by current disposable income, but somewhat smoother than one would guess looking at current disposable income, and, finally, also higher when the ratio of financial wealth to current income is higher. This fits a model in which people mostly consume from income, are myopic, and have habits so their consumption slowly adjusts to new levels. It is a brief summary of the Keynes’s discussion. I think it fits the data amazingly well (old pdf warning). There are theoretical arguments for why such a primitive model shouldn’t work, but it does. I think the fatal weakness of this story is that it was written 80 years ago. The old Keynesian model of consumption also has the advantage that it is so tractable that adding it to a model of other macroeconomic variables adds almost no difficulty. I see no reason not to use it (really 0 (zero)).

The average non economist assumes that employment is determined by labor demand because there are unemployed workers looking for jobs. There are theoretical arguments that this can’t be true. However, it is very hard to fit the data unless one assumes that there is excess supply of labor. The currently dominant new Keynesian (NK) dynamic stochastic general equilibrium (DSGE) models imply that there is excess supply of labor. Worker’s desire for leisure affects future wages, but doesn’t prevent them from supplying all the labor demanded in the present. I fear that this paragraph isn’t obvious to some working academic macroeconomists (at least 2 in the world). The models aren’t transparent. The fact that their equation for total hours worked is old school, isn’t obvious.

Similarly, the average person and most working mainstream macroeconomists assume that current sales of goods and services are determined by demand — that firms supply the amount demanded and that their costs and profit maximizing decisions about prices affect future prices not current sales. This assumption is now main stream but it wasn’t either 15 or 30 years ago. Again, I think the extent to which current NK models are similar to old Keynesian models isn’t universally recognized.

I have not written that the level of production is demand determined. Total output is equal to sales plus net inventory investment. One very odd thing about NK DSGE models is that there are no inventories. This is very odd, because net inventory investment is highly cyclical and they are models of business cycles. Instead all investment is treated as non residential fixed capital investment.

Investment of all sorts will be discussed under the heading of “too hard problems”.

In NK DSGE models, government consumption plus investment is treated as a policy variable. The idea is that it isn’t optimized now but can and should be. I think that’s a safe assumption.

Closed economy models are oddly appealing even to economists who work in small open economies. Here a very large part of the problem is that exchange rates are very hard to model. Given exchange rates, I think macroeconomists can get away with the old assumption that imports depend on GDP and the real exchange rate, so exports depend on foreign gross production and the real exchange rate.

Finally benchmark NK DSGE models assume that there is one interest rate and that it is determined by a monetary authority using a simple backwards looking Taylor rule. It is also assumed that all agents are free to lend or borrow as much as they want at this interest rate (that they aren’t liquidity constrained).

This is not adequate and a huge amount of the research of the past 8 years has attempted to address risk premiums, term premiums and liquidity constraints. I am quite certain that benchmark NK DSGE models have fatal faults which are not addressed by this literature.

So far, I’ve mentioned but not addressed six difficult topics: investment, wage inflation, price inflation, exchange rates, interest rate differentials, and liquidity constraints. They are all difficult because they are determined by agents who worry about the medium and long term future.

Excessively Hard Problems

Some investment doesn’t not depend crucially on long term expectations. In particular, inventory investment doesn’t. It is easy to see why this is true of work in progress inventories — investing in half finished products is just a cost of business. The level of inventory investment can be fit pretty well assuming that finished goods inventories are retailing work in progress. Basically, a very simple model which works quite well is the target inventory model in which desired inventories are a constant times final sales and inventories are adjusted to hit the target next period. This means that short term forecasts are enough. I strongly suspect that the model would work about equally well whether expectations were assumed to be rational or it it were assumed that agents assume that next period’s final sales are equal to current sales (plus a Christmas season effect).

Investment in equipment and software also might not involve long term forecasts. Equipment wears out and becomes obsolete. Software is perceived to become obsolete (I am typing this on a brand new computer running Windows 7, but I am eccentric).

I think that inventory, equipment, and software investment can be modeled with a super simple accelerator in which such investment depends on the growth of GDP. This is even simpler than 1960s era models of investment, because the (problematic) subjective expected real interest rate doesn’t matter.

Investment in structures (whether residential or not) is much more problematic. Investors (househods in the case of residential investment) care a lot about what the structure’s price will be in future years and decades. People buying houses have to try to forecast their future income and other future needs (their future health) to decide how much wealth to tie up in a house. Firms building structures have to predict demand for their products and operating revenues years into the future.

I think it is possible to usefully summarize all of this uncertainty as if it were uncertainty about the resale value of the structure. I also think that decades of painful experience have taught us that we are nowhere near to being able to forecast agents’ long term forecasts of resale values. I am very confident that there are irrational bubbles which grow and burst. I am equally confident that I can’t predict when they will inflate, explain how to prevent them, or even recognize all of them while they are inflating. I think macroeconomists have to deal with a failure to model bubbles. This does not mean that I think they can be treated as exogenous. That would be appropriate only if one were sure that policy can’t affect the frequency and magnitude of bubbles. I think we have to prepare to deal with bubbles (and with our mistaken impressions that a sustainable asset price increases are bubbles). The problem is very hard to solve, but I don’t think we can pretend either that we have solved it or that it doesn’t matter.

There has been a similarly heated debate about the effect of future price expectations on wage and price inflation. This is the rock on which old Keynesian macroeconomics was perceived to have smashed.

The old model of wage inflation was an expectations augmented Phillips curves in which wage inflation depended on unemployment and expected future price inflation. It was perceived that simple primitive formulas for expected price inflation failed when confronted with data (and still more when confronted with thought experiments). Oddly the primitive formulas work very well (an old PDF on expectations warning and one on the Phillips curve). In fact, current discussion of monetary policy by the Federal Reserve Open Market Committee (FOMC) sure sounds extremely old fashioned. I think it is very hard for an agents to forecast inflation during a hyperinflation and very hard for an economist to understand the flailing efforts of such agents. I don’t think that is a relevant problem for first world monetary authorities.

update: replying to a comment by Bakho, I realize that I didn’t mention downward nominal wage rigidity — that is the claim that it takes extremely high unemployment to convince workers to accept fewer dollars per hour (but it is much easier for them to accept the reduction value of fixed nominal wages caused by inflation). I think it is extremely important. I think it isn’t discussed because it is unusually hard to reconcile the clear pattern in the data with any sort of conventional micro foundations.
end update:

I think the problem with the Phillips curve is the other variable unemployment. There is strong evidence that cyclical unemployment can become structural (that there is hysteresis). Wage inflation depends on lagged unemployment as well as current unemployment and is fit a bit better using the change in the unemployment rate not the level. This is very important, because it implies huge long lasting costs of high unemployment. It does not make models intractable. It is very odd that not so very much is written about the issue even in countries (such as Italy where I type) where it is very clear it is gigantic problem.

Until recently, I would have thought that it was easy to pass from wage inflation to price inflation with prices adjusting towards marginal cost times a constant markup factor. It is easy to calculate marginal cost if one assumes a Cobb-Douglas production function, so the marginal product of labor is the constant share of labor times the ratio of output to labor. However, recently, there has been reduced pass through from wage to price inflation (basically prices kept going up even with very low wage inflation — in other words, the share of labor declined). I still think the really hard part of the problem is that firms attempt to forecast future wages, materials prices and prices charged by competititors. So expected future inflation affects current price inflation as well as current wage inflation.

Here I am confused enough that I have to appeal to actual data. I don’t claim to understand why this graph looks as it does.


I think any attempt to model exchange rate fluctuations is foolish. It might be possible to understand them some day, but not usefully soon. Most foreign exchange transactions are speculative — forecasts of future exchange rates are the whole point not just one factor. I think that they have to be treated as an endogenous but not understood variable. In any case, most macroeconomists can say they aren’t our problem.

Similarly, I think it is very hard to understand risk premiums. They are much too important to ignore. Risk premiums and liquidity constraints are financial frictions. There is currently an extremely active effort to put financial frictions into macro models.

As with long term house price expectations and exchange rates, I think varying risk premiums must be considered too important to ignore and too difficult to model. I don’t think it is OK to treat them as exogenous shocks. I don’t think it will be possible to model them any other way for a good long while (too long for impatient policy makers).

I think it is possible to largely dodge the question of when and why people are liquidity constrained, because I think it is possible Liquidity constrained consumers consume all of their current income. They do not create much of a problem in a model of aggregate consumption largely determined by current income. Myopic consumers do not attempt to solve the very hard problem of optimal consumption by a consumer who is not liquidity constrained now but might be in the future. similarly, liquidity constrained firms invest their free cash flows. Their behavior is simple. Liquidity constraints make simple inter-temporal models messy. But I think that such models have worse problems.

It is possible to understand the effect of home equity on liquidity constraints — people with negative home equity (underwater mortgages) can’t get home equity loans. Unfortunately, the same mysterious wave of fear which causes high risk premiums causes tight lending standards. I think we can understand a little about who is liquidity constrained when, but I doubt it is enough to be useful when modelling aggregates.

So, in sum, my proposal is to admit that many problems are very simple and were solved decades ago, and that others won’t be solved any time soon. I think the hard problem of how people really form expectations is still unsolved and must be solved before macroeconomics can amount to a social science. I think (as I have for over 30 years) that we have to listen to psychologists and (shudder) sociologists. I am pleased to note that many very famous economists do just that (cough Akerlof cough). I do think that wide spread distress about the fruits of the rational expectations assumption make

We can still ask if there is anything useful we can say any time soon. I think it is clear that there is. NK and old Keynesian economists have fundamental methodological disagreements (which means most macroeconomists suffer from fundamental methodological ambivalence). However, we agree that fluctutations in nominal aggregate demand can affect output and welfare, that demand is currently still slack in most of the developed world, and that temporary government spending increases are effective fiscal policy.

That isn’t all that much (and adds nothing at all to Keynes). However, it implies policy recommendations which are too radical to be seriously considered by policy makers and supported by the majority of macroeconomists (the vast majority of macroeconomists who don’t work near great lakes).

*Sorry I can’t resist the temptation to write this post. I have posted most of these thoughts many times already. (Almost) no one ever asked me. The post is above my pay grade — I should first earn some standing in the profession before presuming to tell it how to reform itself. I am also sorry for this apology (also typed many times before) so I put it down here where no one is likely to read it.