Fiscal Stimulus and Christina Romer: Once More into the Breach

The debate over the effectiveness of fiscal stimulus continues to rage. I have previously given my views on the likely efficacy of fiscal stimulus (Does the multiplier have to be one? and Another stab at fiscal neutrality): the multiplier is almost certainly less than one, is likely close to zero, and may even be negative.

Christina Romer, the head of the President’s Council of Economic Advisers, disagrees and she gave a forceful defense of fiscal stimulus in a recent speech. She has many reasons supporting her view, but basically she believes that any increase in government spending boosts output. This belief is predicated on the notion that the government can increase its demand for goods and services without any negative effect on other parts of the economy: she believes changes in government spending do not change prices, neither goods prices nor interest rates. Further, she appears to believe that the composition of government spending does not matter for this result to hold: any fiscal expenditure boosts output.

There is a broad literature on this topic. Many economists have attempted to estimate the multiplier. And, the vast majority of these studies, including several by the IMF, have found very small and often statistically insignificant effects. That is, most studies have found fiscal multipliers close to zero.

Romer is aware of this literature but attributes the empirical findings to omitted variable bias. In simple terms, fiscal policy is only implemented when the economy is expected to weaken. The results then simply fail to account for how much worse the economy would have been in the absence of the fiscal stimulus. Omitted variable bias is an endemic problem in empirical studies; I lack Romer’s faith in the sign of the bias.

Romer takes a three-pronged approach to defending her position: she highlights the magnitude of the increase in government demand, she draws inferences from her recent work with David Romer, and she uses the results of large-scale macro models. I take each of these in turn.

The Magnitude of Government Spending

Romer notes that the administrations stimulus plan is the large (unless noted, all quotes are from Romer’s March 3 speech:

It is simply the biggest, boldest countercyclical fiscal stimulus in American
history. One way to see this is to compare it with Franklin Roosevelt’s New Deal. In the biggest year of the New Deal, 1934, the fiscal expansion was about 1½% of GDP. And this expansion was followed the very next year by a cutback of almost the same size. In contrast, the act that was just passed provides fiscal stimulus of close to 3% of GDP in each of 2009 and 2010.

And her assessment of the program includes none of the financial rescue packages. With the Fed spending about $1.3 trillion (having already spent close to 2) and the Treasury implementing a new trillion dollar toxic asset program, total spending on stimulus is remarkably large, arguably closer to 25 percent of GDP rather than 3. In Romer’s view, this government spending cannot help but stimulate economic output.

Romer clearly believes that all government spending is stimulus. In her introduction, she makes the case over the timing of the bill rather than composition. I have already spoken on the content of the bill. But, here I find an inconsistency in Romer’s views. If all government spending increases output more than one-for-one, rebate checks should also have a larger multiplier.

After all, what is the difference between sending somebody a check for $1,000 and hiring them for 10 hours at $100 an hour? Both cases entail the same transfer of resources. Romer, however, differentiates these cases. In the former, case she believes the multiplier is close to 0.3 in the latter it is close to 1.5 (Bernstein and Romer 2009). Yet, while both examples entail a transfer of $1000 from the Federal government to private individuals, in the latter case, the government has reduced the total labor supply available to the private sector by 10 hours, placing upward pressure on wages. Higher wages mean less private employment.

Take this another way, Romer states “all of an increase in government purchases goes into spending, whereas only some fraction of a tax cut is spent.” This is a partial equilibrium statement. Think of the government’s budget constraint. To spend money the government must either raise current taxes or it must borrow money; either way, it must take cash from the private sector to spend (either in the form of a bond or in the form of taxes). Assume it is debt finance, the private investor (who has his own budget constraint) must either reduce consumption or reduce investment in some other project: He has less money net of his purchase of the government bond. This indicates a multiplier no greater than one.

Now think, the government uses the case to purchase something. Where does the government get the item? It must have bought the good and it must have paid a positive price (otherwise spend away, I don’t care). But this increase in demand for the particular good must raise its price (as the government has to bid away the good from some other user). This is a real change in relative prices. If the economy is subject to any frictions at all, the multiplier must be less than one.

In this case of the current situation, the price effects are likely to be much larger than in the historical record. There is never enough slack in the economy to spend 3 percent of output and not see a difference. And if the CBO’s new estimates are anything close to accurate the crowding out is going to get a lot bigger before it gets any smaller.

Large Scale Macro Models

In her speech, Romer states the “policy multipliers [derived in large scale macro models] are surely more accurate than the simple calculations Barro suggests because big macro models try to take into account other factors driving output.” The suite of modern macro models used by and monetary policy influence output: the models are hard-wired to find beneficial effects of policy. This is an assumption in modeling not a result driven by data. Forecasting models without these features perform just as well in forecasting output but find no effect of policy.

Nevertheless, even taking the models as an accurate tool, Bernstein and Romer force the models to produce even greater effects of fiscal policy. I have already written on Bernstein and Romer’s misuse of the macro models (here). In summary, they force the Fed’s interest rate to remain at zero forever. And, since relative prices are all relatively fixed in these models, they force all prices to remain unchanged forever. Since there is no budget constraint in the models (they absorb differences in income in consumption in an external sector), they have essentially assumed that the government does not displace any other source of demand.

[By the way, whenever you hear someway say they have produced results from a large-scale macro model assuming no monetary policy response, they are effectively fixing prices in the model. So, if you hear these words, use caution in interpreting the results.]

A new paper by Cogan, Cwik, Taylor, and Wieland shows this argument using the same suite of models. Cogan et al assume that the Fed’s interest rate remains zero for two years and then responds optimally via a Taylor rule. They use exactly the same models used by Bernstein and Romer. This one change pushes the fiscal multiplier down to 1 in the first quarter (prices are sticky in these models) and pushes the multiplier all the way down to 0.2 after four quarters. Imagine if they were able to let all other prices in the model adjust at the same time. I am thinking the model would be close to 0.2 in the first quarter.

The Romer and Romer Results

Christina Romer’s work with David Romer uses a narrative style to discuss the likely effectiveness of fiscal policy through the historical experience with tax cuts. (I discuss the details of this paper here) In short, I find their results just as biased (in an omitted variable sense) as other studies in the literature.

I object to two characterizations she makes in reference to this work. First, she states that doing a narrative analysis for government spending would be difficult. I refer her to this post and this post. In the first post, I discuss the rise in defense expenditures in the late 1970s in the United States and in the second I discuss Japan’s experience. Combine these with Barro’s wartime narrative and I think we have done her work for her.

Second, she says “the usual relationship between tax and spending multipliers would be maintained. That is, measured correctly, I would expect eh spending multiplier to be larger than the tax multiplier.” But, this sentence comes right after she tells us that a one percent decrease in taxes boosts GDP by 3 percent. That means she actually believes the multiplier is greater than 3. I think we should be able to measure a multiplier of that size.

Summary

Nothing I have said is proof that I am right. I do, however, feel that the preponderance of evidence is tilting sharply in my favor. The current economic cycle should once and for all settle the debate on the efficacy of fiscal policy. If the economy is indeed currently in the midst of recovery or if it recovers over the next few months, I will take that as evidence in favor of stimulus; if, on the other hand, the economy continues to fester, I will take that as contrary evidence. I do not believe we can say, as Krugman has argued, that the response is too timid.

But that is for future research.