Last week I did a post based on a chart that I saw in an article in the New York Review of Books by Paul Krugman. Relying on an earlier paper by Robert Hall on the empirical evidence about the effectiveness of fiscal stimulus, Krugman used the chart to illustrate the efficacy of government spending as a stimulus to economic recovery. While Krugman evidently thought his chart was a pretty compelling visual aid in showing that fiscal stimulus really works, I didn’t find his chart that impressive, because there were relatively few years in which changes in government spending were clearly associated with large changes in growth, and a lot of years with large changes in growth, but little or no change in government spending.
In particular, the years in which government spending seemed to make a big difference were during and immediately after World War II. The 1930s, however, were associated with huge swings in GDP, but with comparatively minimal changes in government spending. Instead, changes in GDP in the 1930s were associated with big changes in the price level. The big increases in GDP in the early 1940s were also associated with big increases in the price level, the rapid rise in the price level slowing down only in 1943 after price controls were imposed in 1942. When controls were gradually lifted in 1946 and 1947, inflation increased sharply notwithstanding a sharp economic contraction, creating a spurious (in my view) negative correlation between (measured) inflation and the change in GDP. From 1943 to mid-1945, properly measured inflation was increasing much faster than official indices that made no adjustment for the shortages and quality degradation caused by the price controls. Similarly, the measured inflation from late 1945 through 1947, when price controls were being gradually relaxed and dismantled, overstated actual inflation, because increases in official prices were associated with the elimination of shortages and improving quality.
So in my previous post, I tried to do a quantitative analysis of the data underlying Krugman’s chart. Unfortunately, I only came up with a very rough approximation of his data. Using my rough approximation (constructing a chart resembling, but clearly different from, Krugman’s), I ran a regression estimating the statistical relationship between yearly changes in military spending (Krugman’s statistical instrument for fiscal stimulus) as a percentage of GDP and yearly changes in real GDP from 1929 to 1962. I then compared that statistical relationship to the one between annual changes in the price level and annual changes in real GDP over the same time period. After controlling for the mismeasurement of inflation in 1946 and 1947, I found that changes in the rate of inflation were more closely correlated to changes in real GDP over the 1929-1962 time period than were changes in military spending and changes in real GDP. Unfortunately, I also claimed (mistakenly) that that regressing changes in real GDP on both changes in military spending and inflation (again controlling for mismeasurement of inflation in 1946-47) did not improve the statistical fit of the regression, and did not show a statistically significant coefficient for the military-spending term. That claim was based on looking at the wrong regression estimates. Sorry, I blew that one.
Over the weekend, Mark Sadowski kindly explained to me how Krugman did the calculations underlying his chart, even generating the data for me, thereby allowing me to reconstruct Krugman’s chart and to redo my earlier regressions using the exact data. Here are the old and the new results.
OLD: dGDP = 3.60 + .70dG, r-squared = .295
NEW: dGDP = 3.26 + .51dG, r-squared = .433
So, according to the correct data set, the relationship between changes in government spending and changes in GDP is closer than the approximated data set that I used previously. However, the newly estimated coefficient on the government spending term is almost 30% smaller than the coefficient previously estimated using the approximated data set. In other words a one dollar increase in government spending generates an increase in GDP of only 50 cents. Increasing government spending reduces private spending by about half.
The estimated regression for changes in real GDP on inflation changed only slightly:
OLD: dGDP = 2.48 + .69dP, r-squared = .199
NEW: dGDP = 2.46 + .70dP, r-squared = .193
The estimated regression for changes in real GDP on inflation (controlled for mismeasurement of inflation in 1946 and 1947) also showed only a slight change:
OLD: dGDP = 2.76 + 1.28dP – 23.29PCON, r-squared = .621
NEW: dGDP = 3.02 + 1.25dP – 23.13PCON, r-squared = .613
Here are my old and new regressions for changes in real GDP on government spending as well as on inflation (controlled for mismeasurement of inflation in 1946-47). As you can see, the statistical fit of the regression improves by including both inflation and the change in government spending as variables (the adjusted r-squared is .648) and the coefficient on the government-spending term is positive and significant (t = 2.37). When I re-estimated the regression on Krugman’s data set, the statistical fit improved, and the coefficient on the government-spending variable remained positive and statistically significant (t = 3.45), but was about a third smaller than the coefficient estimated from the approximated data set.
OLD: dGDP = 2.27 + .49dG + 1.15dP – 13.36PCON, r-squared = .681
NEW: dGDP = 2.56 + .33dG + 1.00dP – 13.14PCON, r-squared = .728
So even if we allow for the effect of inflation on changes in output, and contrary to what I suggested in my previous post, changes in government spending were indeed positively and significantly correlated with changes in real GDP, implying that government spending may have some stimulative effect even apart from the effect of monetary policy on inflation. Moreover, insofar as government spending affects inflation, attributing price-level changes exclusively to monetary policy may underestimate the stimulative effect of government spending. However, if one wants to administer stimulus to the private sector rather than increase the size of the public sector at the expense of the private sector (the implication of a coefficient less than one on the government-spending term in the regression), there is reason to prefer monetary policy as a method of providing stimulus.
The above, aside from the acknowledment of Mark Sadowski’s assistance and the mea culpa for negligence in reporting my earlier results, is all by way of introduction to a comment on a recent post by my internet buddy Lars Christensen on his Market Moneterist blog in which he welcomes the looming fiscal cliff. Here’s how Lars puts it:
The point is that the US government is running clearly excessive public deficits and the public debt has grown far too large so isn’t fiscal tightening exactly what you need? I think it and the fiscal cliff ensures that. Yes, I agree tax hikes are unfortunate from a supply side perspective, but cool down a bit – it is going to have only a marginally negative impact on long-term US growth perspective that the Bush tax cuts experiences. But more importantly the fiscal cliff would mean cuts in US defense spending. The US is spending more on military hardware than any other country in the world. It seems to me like US policy makers have not realized that the Cold War is over. You don’t need to spend 5% of GDP on bombs. In fact I believe that if the entire 4-5% fiscal consolidation was done as cuts to US defence spending the world would probably be a better place. But that is not my choice – and it is the peace loving libertarian rather than the economist speaking (here is a humorous take on the sad story of war). What I am saying is that the world is not coming to an end if the US defense budget is cut marginally. Paradoxically the US conservatives this time around are against budget consolidation. Sad, but true.
I am not going to take the bait and argue with Lars about the size of the US defense budget. The only issue that I want to consider is what would happen as a result of the combination of a large cut in defense (and in other categories of) spending and an increase in taxes? It might not be catastrophic, but there seems to me to be a non-negligible risk that such an outcome would have a significant contractionary effect on aggregate demand at a time when the recovery is still anemic and requires as much stimulus as it can get. Lars argues that any contractionary effect caused by reduced government spending and increased taxes could be offset by sufficient monetary easing. I agree in theory, but in practice there are just too many uncertainties associated with how massive fiscal tightening would be received by public and private decision makers to rely on the theoretical ability of monetary policy in one direction to counteract fiscal policy in the opposite direction. This would be the case even if we knew that Bernanke and the FOMC would do the right thing. But, despite encouraging statements by Bernanke and other Fed officials since September, it seems more than a bit risky at this time and this place to just assume that the Fed will become the stimulator of last resort.
So, Lars, my advice to you is: be careful what you wish for.
PS Noah Smith has an excellent post about inflation today.
Uneasy Money 
asked a related question on Lars’ blog and he pointed me to the following link
Click to access The-Great-Depression-of-1946.pdf
I would welcome your comments on that.paper.
I have a more general question.
Lars suggestion is that QE-type policies should be used to stabilize NGDP in a scenario where the govt deficit is reduced. Your concern is that while in theory this should work, in practice it may have unforeseen consequences that would undermine any recovery we may be in.
My question is : Leaving aside any political (or legal) considerations that would make this impractical – why couldn’t the govt just agree to reduce the deficit then directly create the money needed to fund current govt spending levels (as part of an NGDP targeting regime, where QE takes care of balancing the numbers)?
This policy would avoid falling off the fiscal cliff, while also avoiding tax-increases or the dangers associated with sudden decreases in the govt sector. As the economy recovers we could then gradually (and safely) reduce govt spending (funded by money creation) and replace it with private spending without any risk of jeopardizing the recovery,
LikeLike
David
Am very disappointed in this essay. Munger would label you a man with a hammer.
Defense spending is all about expectations: greater certainty of being paid, fear of losing a war, or, hope of a winning a war. You never consider. Hence your regressions are wholly meaningless.
Feynman would said you were “not even wrong,” as you are so far from being scientific.
Further, what is the point of engaging with any libertarian? Incentive caused bias is reason enough alone to never engage with a libertarian.
LikeLike
Wow. Could this be a beginning of new synthesis – asking for fiscal AND monetary stimulus if conditions are dire enough?
Anyways maybe everybody else already forgot about it, but I reckon that you “promised” to contribute something into the medium of account vs medium of exchange discussion. I would really love to see what you have to say.
LikeLike
@rob,
I think the main problem with your modest proposal is that it would signal the wrong things. Would it indicate that the gov’t will replace future deficits with printed money? It might not be a complete disaster if it is done once, but what if the politicians catch on?
As for John, maybe increased government spendings’ effects are all about expectations.
Mr. Dubois, new synthesis of what? And that makes me think: David, would your model posit a spending multiplier? BTW, I was thinking that military spending is really more of a baseline for the multiplier because (a) there are no returns on investment to speak of, it diverts manufacturing resources from other private-sector uses (particularly in the WWII conversion), it spends quite a bit overseas and any technological advances will only be privately available years later.
LikeLike
Julian,
Yes, there would have to be very clear statement of intent to the effect that a NGDP target was being introduced, as well a target for specified lower level of (future) government spending and a balanced budget.
Unconventional monetary policy would be used to hit the NGDP target.
Money creation to finance govt spending would be a special form of this unconventional monetary policy. It would continue until such time as other forms unconventional monetary policy is no longer needed, at which point govt spending (and the money creation that is financing it) can start to be cut until the target level of govt spending is reached
At that point standard monetary policy could be used to hit the NGDP target.in the future and a balanced budget would be mandated.
LikeLike
David, your comments go astray, but Rob . . .
Where does one start when one reads, “there are no returns on investment to speak of”.
Surely we have learned by now that what a machine represents is knowledge, and little more, in an economic sense.
Surely we have learned the parable of teaching men to fish.
For the United States, tens of millions of men returned knowing about everything one could about construction and motor vehicles.
Millions of women learned extraordinary skills. You should know my aunt who went to Italy for the OSS.
In my hometown, returning soldiers and sailors voluntarily built a factory building so that they would have a factory in which to work.
Have you ever read about people in any third world country with the knowledge to build their own factory building?
I could go on but the narrowness of the POV of view here is
Even our most massive expenditures (Navy ships) were not write offs. Lots of the steel was recycled.
LikeLike
John,
Good points about that.
LikeLike
rob, A quick look at the paper to which you provided a link suggests that there are very acute problems in measuring GDP in the years 1944-47 precisely because of the existence of price controls makes it so difficult to measure nominal GDP much less decompose changes in nominal GDP into changes in real GDP and changes in prices. So I think that the authors are making a useful point, which they may be trying to make more out of than is warranted. But I would have to read the paper more carefully than I have to provide a serious assessment of the paper.
I think that I would generally agree with the idea that you are advancing that the government should be willing to finance current expenditures through money creation rather than impose draconian cuts in spending or increases in taxes. But doing that wouldn’t reduce the measured budget deficit. If the Fed prints money with which to pay for government purchases the deficit is not affected. Dollar creation does not count as government revenue.
John, Sorry to hear of your disappointment. I will try to get over it and suggest that you do the same. I don’t make any great claims for my regressions, but I don’t see how your comments about the defense spending add to or detract from any meaning that my regressions might have. I have the utmost respect for the genius of Richard Feynman, but I don’t think invoking his authority advances your position in the slightest. Further, I am not now a libertarian, and don’t think that I ever was one, certainly not for longer than 15 minutes, but libertarians are people and I think that a better reason than “incentive caused bias” (whatever that might mean) never to engage with them.
J.V., I am just trying to assess what a certain set of data show. I am not proposing a new synthesis. I am just suggesting that we not assume too casually that abruptly tightening fiscal policy would not have any affect on an economy that is operating at or near the zero lower bound, or in what Hawtrey characterized as a “credit deadlock.” I am hoping to post something about the medium of account v. medium of exchange discussion in the next few days. Thanks for not forgetting. Stay tuned.
Julian, I have no single model that I am using. I use different models in different situations as seems appropriate. My regressions were pretty atheoretical, simply attempts to summarize a small body of data in a useful way.
John, Your comment seems to have been directed toward Julian not toward rob. Which is why Julian responds to you.
LikeLike
David
Being from Missouri perhaps I am overly sensitive about Twain’s wisdom (lies, damn lies, and statistics)
FWIW, your regressions, to me, are statistics. Comparing the impact of one year’s defense spending to another’s strikes me as beyond folly.
Munger charges, The Psychology of Human Misjudgment, that this is man with a hammer syndrome. You have the skill to do the regressions so you mistakenly conclude that they show something, that they have to be defended, etc.
My POV is that they don’t show or tend to show anything.
I keep returning to fundamentals. The task and charge of finance is to provide the means and methods for putting idle people and resources to work. It is a confidence game, as best explained by Minsky.
Finance, today is broken. We have idle people and resources beyond imagination to most of us. They remain idle because no one can finance putting such to work, especially because no on can figure out how to put consumers in a position to finance the purchase of what is made or produced, if such idle people and resources were put back to work.
Today’s WSJ had a story on a major part of the problem. Millions with overwhelming college debt. It would have been more useful for you to have done a regression study on rising college debt/economic performance, say over the last 20 years, than defense spending before and after WWII.
But, enough of me ranting. You get my point.I am pushing you to be productive.
LikeLike
David,
Thanks for your answer.
I agree that financing govt spending out of newly created money wouldn’t reduce the measured budget deficit but would stop the level of govt debt increasing. I was seeing it as an alternative solution to many of the “financial cliff” laws that will come into effect in 2013, that would taper the level of govt spending over time within a NGDPT type framework.
LikeLike
I have to say I found Lars’ post compelling.
Monetary policy is the key to stimulus, not fiscal policy.
I also wonder if federal agency spending is the worst kind of stimulus. On other words, transfer payments go to people who will spend the money, directed by their choices in free markets.
Agency spending is just a black hole (mostly), improving no one’s economic lot, except those directly benefitting. Really, we could not eliminate the USDA, Commerce and HUD without ill effect?
As to defense spending, everything Eisenhower feared has come true, in spades.
Bush jr’s reaction to 9/11 was to double defense outlays in real terms, and guess who the defense establishment votes for?
We have a Cold War military on steroids to fight a few stateless terrorists…..
LikeLike
John, I revere Mark Twain, but that doesn’t mean I have to go through life asking myself whether would Mark Twain would approve of my regressions. I don’t claim that my regressions prove anything, they just provide a convenient way of summarizing a jumble of data to see if there is any discernible pattern that emerges from the data. We always do that, a regression is just one way to do it a little more systematically. By the way, it takes very little skill to run a regression nowadays. I hope future blog posts will be more to your liking, but I offer no guarantees.
rob, OK, thanks.
Benjamin, I also think monetary policy is key, but that doesn’t mean we shouldn’t be concerned about a tightening of fiscal policy in the current environment. That’s all I was saying. About defense spending, that’s not a discussion that I want to get into. I am in favor of a strong national defense, and an willing to pay for it. But all the defense spending in the world will not help us if the people in charge are clueless.
LikeLike