Mark Thoma has assembled a set of useful discussions of multipliers. Econbrowser has added a handy new category “multipliers”, that compiles entries on the topic. In addition, Ethan Ilzetzki, Enrique G. Mendoza and Carlos A. Vegh provide a very useful cross-country (including emerging market economy) survey here and here [pdf].
This week’s Economic focus in the Economist (Sept. 26th – Oct 2nd, p. 90) also has an interesting discussion about fiscal multipliers.
Here is the link to the on-line version of the article. However, I’m not sure if nonsubscribers can access it.
Much ado about multipliers
The article also has a link to various papers on the subject:
multiplier articles.
–SMG
I found the Itzetzki et. al. piece a bit puzzling. They seem to be making calculations based on the assumption that the multiplier is constant, at least within a subset of nations. But isn’t it zero by necessity when the economy is at full employment? Does that not imply that, when the stimulus is actually needed, spending will be multiplied by more than they calculate?
…I had thought this all had one upside — that the macro I had learned back in school was not so outmoded as my new friends from Minnesota had been telling me. Should I think again?
Short question: do multipliers have predictive value?
Interesting WSJ op ed piece in yesterday’s (Oct 1) wall street journal.
“Stimulus Spending Doesn’t Work
Our new research shows no evidence of a Keynesian ‘multiplier’ effect. There is evidence that tax cuts boost growth.”
By ROBERT J. BARRO AND CHARLES J. REDLICK
[snip]The bottom line is this: The available empirical evidence does not support the idea that spending multipliers typically exceed one, and thus spending stimulus programs will likely raise GDP by less than the increase in government spending…However, there is empirical support for the proposition that tax rate reductions will increase real GDP.[/snip]
tj: If you want to get your insights from the WSJ op-ed page, and from a person who doesn’t recognize breaks in data series (see here for his famous data gaffe), then more power to you.
Yeah, that was an unhelpful comment, Menzie.
After today’s unemployment report, and in light of your past running blog fight with Posner, I’m thinking that Q3 GDP is going to come in much, much less than you would have predicted based on stimulus spending. You won’t take this as a repudiation of your multiplier theories, but I will.
We’ve talked about an alternative history where Lehman was bailed out. I wonder about one where a payroll tax holiday was substituted for this “stimulus”, the one where spending was front end loaded into “shovel ready projects”, yet heavy construction employment STILL fell 5%!
For the sake of thoroughness, here’s a tool that does Kevin Murphy’s math on the effectiveness of fiscal spending stimulus, which allows you to incorporate different multipliers, as well as estimates of spending efficiency and deadweight losses.
Menzie,
I agree, it’s much easier to attack the source rather than refute the methodology.
Do you hold the same contempt for the NBER that you do for the WSJ op ed page?
http://www.nber.org/papers/w15369.pdf
Do you put any credence in ANY empirical studies that find fiscal spending multipliers
tj,
You should not attack someone’s religion.
Isn’t it interesting that spending has a positive multiplier effect but confiscating resourcs to fund the spending are simply out of sight, out of mind? Negative multipliers anyone?
tj: Actually, citing the NBER working papper is much better, because then you can see how Barro-Redlick cavalierly dismiss alternative identifying approaches (such as in SVARs), cautions against the difficulties of using time series to infer causality, and then proceeds to make equally incredible identifying assumptions. You will also note that no sample starts after 1950; that is suspicious once one examines Figure 1 in the paper.
Finally, I will admit to a natural wariness of Barro results of this sort. Those of us who are old enough to recall the Barro/Mishkin debates (see Barro, JME, 1976) will recall how fragile his results (relying on a seemingly plausible identification scheme) were.
This is why I attach greater import to surveys and ranges rather than individual studies of multipliers.
By the way, I don’t see a rejoinder on the data issue. I do believe not seeing data breaks (i.e., not looking at the data) is indicative of how one treats data in general…It’s definitely something I warn my PhD students against.
Even Krugman is questioning spending multipliers now: a few weeks ago he proclaimed the death of Reaganomics, yet yesterday he called for tax credits to employers for job creation!
Reaganomics is dead, long live Reaganomics!
Thank you Menzie. DickF (chuckle). Is this what you are regerring to …
Brad delong
http://delong.typepad.com/sdj/2009/04/the-great-ricardian-equivalence-misunderstanding.html
“Robert Lucas: would a fiscal stimulus somehow get us out of this bind…? I just don’t see this at all. If the government builds a bridge, and then the Fed prints up some money to pay the bridge builders, that’s just a monetary policy. We don’t need the bridge to do that… the only part of the stimulus package that’s stimulating is the monetary part…. But if we do build the bridge by taking tax money away from somebody else, and using that to pay the bridge builder — the guys who work on the bridge — then it’s just a wash…. [T]here’s nothing to apply a multiplier to. (Laughs.) You apply a multiplier to the bridge builders, then you’ve got to apply the same multiplier with a minus sign to the people you taxed to build the bridge. And then taxing them later isn’t going to help, we know that… ”
This gets really interesting when you consider the planned increase in G over the next decade. It’s also obvious from recent talk of VAT and Cap’n Trade that higher taxes are in store for everyone.
We are focused on the magnitude of the multiplier associated with recent stimluli. However, if we could move ahead in time 10 years and look back we might observe that the stimulative effect from today’s fiscal policy was offset by slow GDP growth and a permanent decrease in disposable income. (The permanent decrease in disposable income related to the difference between the temporary stimulus and permanent decrease in income from higher taxes).
This discussion is very frustrating because it fails to consider the theoretical foundation for multipliers and the empirical implications of the theory.
The “standard” expenditure multiplier comes from the old Keynesian model Y=C+I+G, where there is no production function and consumption is a function of current income. Then we get a multiplier greater than 1 because an extra dollar of spent by G raises C by MPCxdG, which raised C again by (MPC)^2xdg, etc. That seems to be the model the federal government economists have in mind when they say the stimulus plan will have a multiplier of 2.
That view of the macroeconomy was invalidated by the mid-1970s (*35* years ago) by the permanent income/life cycle theory of C (which dates back much farther) and the recognition that the model was needed an aggregate production function, Y=F(K,L). Those two features were incorporated into a consistent theory of G’s effect on Y by Barro in some articles around 1980. Here are two illustrations emerging from that theory.
Transitory increase in G (that is, a change in the path of G that leaves the present value of the path unchanged): Suppose current G goes up in a transitory way. Suppose for the moment that L is fixed. K is a state variable, so it also is fixed. That means Y *cannot* rise, so the G multiplier is *zero*. The equilibrating mechanism is an increase in the interest rate r.
Now suppose L responds positively to r, which it actually does, though apparently only slightly. For the sake of argument, make the extreme assumption for the moment that L is so responsive to the interest rate that the aggregate supply curve is flat in Y-r space. Then the G multiplier is 1. Output goes up by the shift in G. If the aggregate supply curve has any positive slope, the G multiplier is necessarily less than 1. The multiplier is positive, but *less than 1*. There is no extra kick from the usual multiplier process because the increase in Y is only transitory. Life-cycle households know that, and so have no wealth effect inducing them to raise C.
I don’t see any way out of the constraint imposed by the existence of the production function and life cycle behavior, except to return to the old and clearly inadequate Keynesian model. Nobody seriously advocates using a model with no production function and current income consumption functions.
How about a permanent increase in G (one that changes the present value of the path of G by the same amount as the current increase in G)? In that case, permanent taxes rise, reducing permanent disposable income by exactly the same amount as the increase in G. So far, no change at all in the aggregate demand curve, so no change in r. We get a positive multiplier by noticing that household wealth is now lower (permanently higher taxes), so labor supply presumably rises (assuming leisure is normal). However, C also is normal, so households pay some positive fraction of the increase in G by reducing C demand. That implies that the amount of extra labor is sufficiently small to create extra income that is only a fraction of dG (because the other fraction is being paid by a negative dC). There is no change in r because all shifts in aggregate demand are matched by shifts in aggregate supply. The multiplier again is less than 1.
Barro’s original empirical work 30 years ago supported multipliers less than one. The work I have seen that finds multipliers greater than one ignores either the production function constraints, the implications of life cycle income, or both, making the results useless.
The foregoing theory can be extended to include distorting taxes rather than the lump sum taxes implicitly assumed above, with no change in results. The only possible large stimulus through fiscal policy seems to be through an increase in the deficit, matched with a reduction in current taxes. Because all real-world taxes are distorting, a reduction in current taxes typically means a reduction on current distortions and so more productive activity. Of course, that comes at the expense of even less productive activity in the future, when the debt must be repaid by imposing higher distorting taxes. Because distortion costs are convex, the present value of output is reduced by this scheme, but at least it is conceivable that the current tax multiplier exceeds 1. The current stimulus plan, however, does not reduce taxes and so does not reduce current distortions.
tj,
Understand that GDP is growing. Government spending is a part of GDP. Just imagine, once the entire economy is socialized the GDP can go through the roof and the FED will conveniently supply all the liquidity needed to fuel that growth. There won’t be any consumer goods but there will be huge GDP growth.
I’ll try to keep this brief (for me) as I’ve been chided by JDH via Menzie for being long winded.
The paper by Ilzetzki et al is very interesting. The question for the US is which estimate is most relevant.
IMO it’s the fixed exchange rate estimate. Yes, we have a floating rate. But the authors explain the relatively high fixed-rate number by pointing to Mundell-Fleming, which says that fiscal policy is effective under fixed rates because it doesn’t drive up interest rates (capital flows in). We’re in a similar position for a different reason: ZIRP. The fixed rate multiplier is around 1.5 (remember that figure for now).
And I’ve pointed out via analysis by Glenn Rudebusch we’re likely to have ZIRP through FY 2012 by which time about 99% of the stimulus will be spent. Some say that’s an awful long time to maintain ZIRP. Here’s another way to look at it:
It took 35 and 32 months each from the end of the last two recessions (in order) to the beginning of the fed funds tightening cycle. If the recession trough this time was in June that would at least take us into the 1st or 2nd quarter of 2012 and FY 2012 ends after the 3rd quarter. Also, with the exception of 13 scattered months ZIRP was maintained in the Great Depression for a period of 9 years 4 months (July 1932 through October 1941).
I’ve also found this IMF analysis of the medium-term effects of financial crises in advanced, emerging, and developing economies over the past 40 years to be very interesting:
http://www.imf.org/external/pubs/ft/weo/2009/02/pdf/c4.pdf
I took the following from the IMF report:
1) Following a financial crisis the effectiveness of monetary policy is usually greatly impaired (as in a liquidity trap).
2) GDP usually falls below trend growth for a period of three years but then resumes growing at previous trend growth (but only trend growth) for a period of another 4 years (seven years from start of crisis constituting the medium term). Thus GDP never returns to previous trend in the medium term and a gap remains.
3) The gap between between previous trend growth and current trend growth in the medium term is smaller in proportion to the increase in government consumption. Each 1% of GDP in increased government consumption in the short term results in a 1.5% (gee, this number looks familiar) smaller gap between medium term trend and the previous trend.
This has the following (back of the envelope) implications for the US over the medium term:
1) We should expect GDP growth to be less than trend (less than 2.7%?) for the next two years.
2) Our discretionary fiscal stimulus is being spent at a rate of $100 billion a quarter in the short term. But only about two thirds of this is increased government consumption. This is roughly less than 1.8% of GDP so the gap between trend growth in the medium term and the previous trend growth will be about 2.7% less than contrafactual.
3) Thus GDP in the medium term (between three and seven years out or a total of four years) will be about 10.8% of annual GDP higher. This represents about $1.5 trillion in increased output.
4) Because output will be $1.5 trillion higher in the medium term, and if we assume the aggegate marginal federal tax rate is about one third, federal revenues will be roughly $500 billion higher in the medium term.
5) Given Krugman’s estimate that revenues will be higher to the tune of 40% of the stimulus in the short term and the stimulus was about $787 billion this comes to about $300 billion higher in the short term. Add that to the $500 billion in 4) and you come up with a figure of $800 billion in increased revenues in the short to medium term total. This constitutes a virtual “free lunch” from simply a fiscal point of view.
6) If we had devoted 100% of the stimulus to temporary increased government consumption instead of just two thirds this figure would have been higher by about $400 billion or about $1.2 trillion which is even better than a “free lunch” (if Laffer were a Keynesian he would be proud).
7) From an output point of view this situation is even better. If Krugman and the IMF is right the stimulus should add about $2.4 trillion in output over the short and medium term or about three times what it cost.
This also has implications of the long run. The CBO estimated that the stimulus would reduce real GDP between 0.0% and 0.2% by 2019 because of “crowding out.” However, if the stimulus is essentially a fiscal “free lunch” this effect will be zero.
P.S. On the subject of Barro’s paper, if you count Federal Emergency Relief workers as employed (they would be considered so today according to current BLS definitions: wage received for work done) then unemployment wasn’t really that high at the start of WW II. The unemployment rate was 6.0% in 1941 and fell to 3.1% in 1942 (4th edition Historical Statistics of the United States). My impression is there was a lot less slack in labor and physical capital when WW II started than most people believe.
Fiscal multipliers are dependent on the size of the output gap, as even Barro’s paper concludes. Barro’s paper estimates the defense spending multiplier. The biggest boosts in defense spending during the period covered by Barro (1912-2006) occurred during WW I, WW II, the Korean War and the Vietnam War. Actual GDP exceeded potential GDP throughout both of the latter two wars. Similarly unemployment was very low throughout both world wars. And there were draconian building restrictions in effect during WW II, in fact, the end of those restrictions helped set off the postwar housing boom, and new cars weren’t being produced, because the factories were making tanks instead (and if you did manage to acquire a car somehow, gasoline was rationed). Given this I’m surprised that Barro’s defense spending multiplier estimates were as high as they were. He seems to have made it his life’s work to prove fiscal multipliers are zero and yet the best he can come up with is something in the range of 0.6-0.8.