Having reached the unlucky number of 13, it is fitting to talk about multipliers and model estimates.For much of this lecture, we are going to leave the world of economics behind. We are going to start out in the world of what Greg Mankiw calls the macroeconomic engineer.

Below is a table that gives the output of a hypothetical “model run.” It forecasts for 12 quarters (think of this as starting in the first quarter of 2009 and ending in the the last quarter of 2011) the value of GDP in a baseline scenario (b) and in a scenario (s) in which government spending (G) is higher by 100 every quarter.

Quarter G(b) G(s) GDP(b) GDP(s) Multiplier
1 200 300 1000 1100 1.0
2 200 300 950 1100 1.5
3 200 300 900 1100 2.0
4 200 300 900 1120 2.2
5 200 300 950 1140 1.9
6 200 300 950 1140 1.9
7 200 300 1000 1160 1.6
8 200 300 1050 1180 1.3
9 200 300 1100 1200 1.0
10 200 300 1150 1200 0.5
11 200 300 1150 1200 0.5
12 200 300 1200 1200 0.0

Let me try to explain what is going on. This economy has a full-employment level of GDP of 1200, but in the baseline scenario it is sliding into recession. If nothing is done, it will pull out of the recession at the very end of the 12-quarter simulation period.

The alternative scenario has government spending go up by 100 from its baseline of 200 in every single quarter. That is, the fiscal stimulus is maintained at a constant rate for all twelve quarters. The multiplier is defined here as the difference in GDP divided by the difference in G in each period. For example, in the first period, GDP is higher by 100 (1100 in the alternative scenario minus 1000 in the baseline), and G is also higher by 100, so that the multiplier is 100/100 = 1.0.

In the example, the multiplier builds up over time, and then it declines. It builds up because of, well, the multiplier. As total spending goes up (and government spending does add to total spending), income goes up. And as income goes up, spending goes up. The process feeds on itself. The exact nature of the process is hidden inside my model, which might have hundreds of equations. Honestly, it doesn’t have any equations, and I am making the numbers up, but you get the idea.

The multiplier is zero at the end (“the long run”), because in the long run the economy will come back to its full-employment equilibrium, anyway. So, in the long run the increase in government spending does nothing but crowd out an equal amount of private spending.

This brings up a question. Why doesn’t the government just increase spending for a few quarters, and then drop it back to the baseline level? It could do that. It makes the meaning of the multiplier less clear. If the alternative scenario has G follow a path of 300, 300, 280, 260, 240, 220, 200, and then 200 the rest of the way, what should we use as our measure of the difference in government spending between the alternative scenario and the baseline case?

How we construct the alternative scenario is a matter of taste. I prefer the permanent increase in government spending, because it can be used to illustrate the way that the multiplier varies over time for the same increase in spending.

Overall, the term “multiplier” should best be treated as shorthand. The more precise terminology would be “response of GDP to a given alternative path of government spending.”

I said that in the baseline scenario, the economy returns to full employment on its own in 12 quarters. Most macroeconomic models have the property that the return to full employment comes about, although the length of time may vary. If a model instead has a tendency to deviate from full employment forever, the model is said to be “not well behaved,” which is not a good thing. However, if you go back to Lecture 11 on Leijonhuvfud, you will find that his view is that the economy is only well behaved when it is close to full employment. Far from full employment, it can be badly behaved, and if you buy that, you would not want to impose good behavior on your engineering model when it gets far from full employment.

These models, you see, are a combination of empirical estimates and imposed structure. The empirical estimates are derived from past data, often using a statistical technique known as regression. The structure is based on what the human in charge of the model thinks is reasonable based on theory.

Greg Mankiw’s essay, alluded to earlier, makes the point that academic macroeconomists approve of neither the empirical method nor the structural constraints used by model practitioners. I don’t believe that any major economic journal has published a paper based on a conventional macro-econometric model in the past twenty-five years. You might just as well try to get into the American Economic Review by submitting porn.

There are academic macroeconomists (CEA chairperson designee Christina Romer is one) who have come up with ways to estimate multipliers and still get past the censors at economics journals. This requires really clever technique to conform to the latest methodological standards of propriety. Whether the resulting estimates have any redeeming social value is open to question.

In my opinion, the multiplier (or, more precisely, the response of the economy to a change in the path of government spending) is far from a constant. It should vary over time, for both cyclical and structural reasons.

The cyclical issues concern the state of the economy. I have alluded to the idea that the economy is likely to behave differently when it is close to full employment than when it is far away. That seems relatively noncontroversial. However large the multiplier might be at the bottom of a recession, one would expect it to be lower when the economy is close to full employment. Near full employment, one expects to see more crowding out and more inflation as a result of fiscal stimulus.

In addition, there is the behavior of monetary policy. Fiscal stimulus that runs into a headwind of rising interest rates would be expected to be less expansionary than fiscal stimulus that is accompanied by monetary policy that maintains steady interest rates.

Also, there is the Minsky effect. Hyman Minsky argued that if corporate balance sheets are weak, investment depends on profits. In that case, government deficits generate profits that will strengthen those balance sheets, which will remove a major constraint business investment. Thus, the multiplier will be larger at a time when balance sheet constraint are operative than at a time when firms are not constrained to begin with.

The structural issue concerns fundamental characteristics of the economy. For example, in a small economy, an increase in government spending can “leak out” into spending on foreign goods. In an economy where government debt is less trusted than that of the United States, an increase in government spending could cause adverse financial effects. There are those of us who are not convinced that the United States itself is immune from adverse financial effects.

Another structural issue that I have discussed in previous lectures is the nature of the unemployment that exists. The standard recession of the post-war United States found the economy with excess inventories of durable goods, and most of the unemployed were production workers temporarily laid off from manufacturing facilities. More recently, it seems to me that a lot of our imbalances are structural. In financial services, most of the people who are leaving jobs related to mortgage origination and securitization are not coming back. In autos, some of the decline in production is cyclical, but worldwide there is too much capacity in the industry, so that many unemployed auto workers are also not coming back.

I doubt that I would want to apply the same multiplier analysis to structural unemployment that I would to cyclical unemployment. To my knowledge, neither the old-fashioned macro engineers nor the modern academics have addressed this issue.

Finally, I should point out that the basic logic of the multiplier is at the same time both simple and devoid of economics. The idea is that more spending leads to more income, and more income leads to more spending.

To understand why this is devoid of economics, ask yourself when the process stops. When do people stop earning more income and spending more income? One way to answer is to say that people aim for a balance among labor, leisure, consumption, and saving. In classical economics, people find that balance by responding to market signals. Our belief in the multiplier is a belief that (a) the market is unable to provide that balance and (b) that an increase in government spending can help them to find that balance. For some economists, that faith is so strong that they take seriously an exercise like the “model run” that I made up in the table above.

My own view is that, as of early 2009, there may be enough cyclical unemployment developing to warrant a modest attempt at fiscal stimulus. Also, I suspect that the Minsky effect may be operative, which may make fiscal stimulus particularly effective at reducing cyclical unemployment and perhaps even speeding up structural adjustment, as firms use profits to finance expansion. However, in general, I think that structural adjustment will take a long time, and creating a big fiscal imbalance will cause more problems than it will solve.

Previous lectures:
12. Herd behavior, Procrastination, and Business Cycles
11. Leijonhufvud and the corridor
10. Money or Credit?