Or, on reading those who can do math, and those who can’t (i.e., yet more from Heritage)
The CEA released its most recent assessment of the impact of the ARRA. Some popular accounts, from the Heritage Foundation, from the Weekly Standard, and from Ted Nugent, argue that those estimates of output and employment effects are outlandish. They also cite incredibly high cost-per-job figures. Here are some thoughts on the those arguments (analyses would be going too far).
The basic findings of the CEA report are recounted in Tables 7 and 8 of the report. Those tables also include comparable estimates from the CBO and private firms. I’ve used data from Table 7 to generate the counterfactual GDP. Figure 1 presents the CEA approaches and the CBO low and high effect estimates, while Figure 2 presents the CEA model approach estimate against the private sector estimates.
Figure 1: Log GDP (blue bold), and implied counterfactual (no stimulus) GDP under CEA model (multiplier) approach (red bold), CEA projection (VAR) approach (light green), CBO low and CBO high estimates (gray), all in Ch.2005$ SAAR. NBER defined recession dates shaded gray. Source: BEA, 2011Q1 3rd release, CEA, NBER, and author’s calculations.
Figure 2: Log GDP (blue bold), and implied counterfactual (no stimulus) GDP under CEA model (multiplier) approach (red bold), Macroeconomic Advisers (chartreuse), IHS-Global Insight (light purple), Goldman Sachs (green), Moody’s (teal), JP Morgan/Chase (orange), all in Ch.2005$ SAAR. NBER defined recession dates shaded gray. Source: BEA, 2011Q1 3rd release, CEA, NBER, and author’s calculations.
Two observations
- The CEA model approach (which incorporates using multipliers) is bracketed by the CBO low and high estimated effects.
- The CEA model approach yields a counterfactual GDP in 2011Q1 that is virtually indistinguishable from the private sector forecasts.
Once one views these graphs, it is somewhat surprising to see the ridicule heaped upon the CEA estimates, implying that they are outlandish. The Weekly Standard writes:
When the Obama administration releases a report on the Friday before a long weekend, it’s clearly not trying to draw attention to the report’s contents. Sure enough, the “Seventh Quarterly Report” on the economic impact of the “stimulus,” released on Friday, July 1, provides further evidence that President Obama’s economic “stimulus” did very little, if anything, to stimulate the economy, and a whole lot to stimulate the debt.
Figure 2 demonstrates that the CEA estimates are in line with private sector estimates indicating a substantial impact on GDP.
What is truly hysterically funny (and demonstrates the innumeracy so typical of so much writing these days) is the following assertion:
… The council reports that, using “mainstream estimates of economic multipliers for the effects of fiscal stimulus” (which it describes as a “natural way to estimate the effects of” the legislation), the “stimulus” has added or saved just under 2.4 million jobs — whether private or public — at a cost (to date) of $666 billion. That’s a cost to taxpayers of $278,000 per job.
This seemed like a familiar argument, and then I realized Casey Mulligan performed a similarly (incorrect) calculation two years ago, which involved a similar stock-flow mismatch calculation. My rebuttal was here; to quote:
Note that $787 is spent over several years (a flow). 3.5 million jobs is a stock. But wouldn’t we want to incorporate how long those jobs would be around? That suggests we should use job-years instead of jobs, to make the numerator and denominator comparable. The Administration estimated the number at 6.8 million [2]. That works out to a cost per job-year of $116 thousand.
Since innumeracy abounds, here is a reprise, applied to the Weekly Standard calculation. So far, $666.3 billion has been spent over the past two years (a flow). The cited 2.4 million number applies to an instant, 2011Q1 — a stock. In previous quarters, the number of jobs created was higher (and lower). What one wants to do is to compare a flow to a flow. The number of person-years (equal to one job over one year) is 3.911 million (a flow). Doing the math correctly leads to an estimate of $170,366.
To further highlight the stupidity of the Weekly Standard calculation, suppose in 2015, when almost all funds are expended, only one job was still being supported. Then, using their preferred methodology would indicate each job cost $787 billion…
In addition, back in my 2009 critique of Mulligan’s approach (which he implicitly acknowledges as correct in his blog), I observed that one would also want to know what had happened to profits (i.e., returns to both labor and capital matter).
It’s also important to recall that around a third of the ARRA was tax related. Had more been devoted to infrastructure investment and transfers to the states, as I’d suggested [1], then the cost-per-worker-year would have been lower. It was only at the insistence of Republicans that a larger share of the stimulus was devoted to tax cuts. (By the way, where are all those critics who argued the recession would be too brief for infrastructure spending to matter?).
good and i generally agree
results would have been really bad w/o it, and like anyone reasonable i think fiscal works some
But not the last modeling word.
I have seen no counterfactual that adequately explains how much help was supplied by commercial paper bailout and Money market guarantee, both massively effective in my view. and TARP and fed QE where i think the latter was key in getting treaury 10 yr yields down to 2%.
so the counterfactual w everything else is likely better than shown
all this math and graphs..to argue with people who just repeat, beligerently, you can’t spend your way out of debt..
If I may say so, politely, you are making the same mistake obama made in selling this dry boring longwinded.
take a lesson from krugman or JK Rowling; give a compelling human anecdote (Marge, a retired housewife who lost her savings to madoff, facing eviction, got a job…) and then have a simple summary: we spent x amount, and hired so many people…
Menzie
The crux of your argument is the belief that central planning is the superior economic way.
You are wrong.
Menzie
What is relation of Log GDP to GDP (real or nominal)?
You’ve posted half the results. Yes, GDP is higher than under the counterfactual forecasts assuming no stimulus. However, why didn’t you show the forecasts that did incorporate the stimulus package? Then you’d see GDP is much lower than those models prognosticated. So it’s not clear to me how one can say those (large-scale structural) models “worked” all that well. I think John Taylor and his co-authors have done the best academically rigorous work on the stimulus package (and JDH, of course).
Brian: My research has looked at the effects of QE2 but not the fiscal stimulus.
Take out the growth of war spending and personal transfers, then adjust for price effects and population, and what are the results? Hint: They ain’t what neo-Keynesians, Monetarists, and QEn . . . n+1 proponents want to believe or to tell you.
If one argues that it is not useful, or is misleading, to remove war spending and transfers, then one must conclude that escalating imperial wars, growth of an aging, less-healthy population, and higher unemployment are good for GDP growth.
However, were one to concede that the resulting growth of overall aggregate demand is what is important, irrespective of the source of the growth or its sustainability, then one is, well, merely making the Keynesian case for increasing aggregate demand at any or all costs.
On a December 31 2009 post you wrote:
Moving Forward
At the end of the post, Professor Taylor concludes:
…Moreover, in my view, the models have had their say. It is now time to look at the direct impacts using hard data and real life experiences.
I think one should always look to the data; whether we have sufficient data yet (given that the stimulus started in 09Q2, and we have not yet seen the advance 09Q4 release) makes me less than optimistic that we can yet tease out the effect of the stimulus using, for instance, SVAR approaches.
Is this Taylor’s attempt at looking at the direct impacts?
http://www.stanford.edu/~johntayl/Cogan%20Taylor%20multiplicand%2010-25.pdf
As a daily reader I would be interested in reading your thoughts on this paper.
Menzie-
Never mind – found my answer.
Brian-
As I poked around to find out how the stimulus package was sold originally and how the CEA evaluated it after the fact and it is difficult to tie the numbers into anything concrete. Most of the analysis simply tells you the delta due to the stimulus. Turning the log GDP into GDP (I assume it is constant dollar GDP), I get a growth of about 4.6% trough to peak, or about 2% per year. Nothing to write home about and it appears pretty anemic compared to past recessions. See this analysis from the St. Louis Fed
Any way you massage the numbers, the bottom line conclusion is the same. Economic stimulus dollars are much less cost efficient as a way to create jobs than simply hiring people to do whatever tasks you can find without much regard to what benefit those tasks produce in a New Deal style makework program.
At $170K, you can create at least six tolerable jobs, and ten lousy ones. At $278K, you can create proportionately more. Either way, stimulus is a horribly costly and inefficient way to create jobs.
“The crux of your argument is the belief that central planning is the superior economic way.”
Whoa! Central Planning! I call smear tactic!
Why didn’t you call him a “fellow traveler” or, better yet, just come right out with it and call him a “communist.”
In rebutting Casey Mulligan, Menzie says:
“3.5 million jobs is a stock.”
I do not think this is correct. What is a stock is the amount of jobs in an economy at a certain point in time. But the 3.5 million figure is the *change* in the stock; this is the amount that the stimulus fans said that would be created new during a certain period of time; thus, it is a flow, not a stock.
Furthermore, it seems to me that Menzie and the CEA and others are doing exactly the thing that John Taylor critizes in his paper:
http://www.stanford.edu/~johntayl/JEL_taylor%20revised.pdf
“The problem with using these existing macro models for the evaluation of actual
packages is that they will simply repeat the same prediction story over again. You learn virtually
nothing about the efficacy of a stimulus package if you use the same models to evaluate its
impact ex post that you used to predict its impact ex ante. Indeed, this is one reason for the
disagreement about the impact of the recent stimulus packages. The same models are frequently
being used in policy evaluation studies, which are then referred to in many of the debates about
policy.”
Manfred-
Good cite. If the same models are being used to forecast and evaluate, you get the same answer. Or “it was even worse than we thought.”
You wrote:
Had more been devoted to infrastructure investment and transfers to the states, as I’d suggested
Were there sufficient additional “shovel ready” infrastructure projects available for the budgeting horizon?
Also, transfers to the states may be too broad a category (e.g. “transfers to the states to support education and law enforcement appear to have little effect” per Feyrer and Sacerdote)
I have to disagree with your claim that Professor Mulligan implicitly acknowledges as correct your 2009 critique of his approach in his blog. Here is what he wrote: “We’ll break down the absurdity of that another time, but taking them literally means each job year cost over $100,000.
I rest my case.”
That is, not only he doesn’t acknowledge your critique as correct, he claims it is absurd and, worse, irrelevant. I have no intention of getting into the discussion regarding who is guilty of innumeracy (I do like your argument on how the calculation should be made), but his point still remains that spending $170K to create jobs that on average will pay less than $170K does not make a whole lot of sense. Another critique of Mulligan’s argument (which I prefer) is based on the point you mentioned that he appears to ignore the fact that most of the stimulus is based on tax cuts – e.g., when he writes that “they should consider turning off the first stimulus before more tax dollars are wasted”. It would be interesting to know if Mulligan was in favor of repealing all tax cuts…
For some reason all of the “cost per job” discussions I’ve seen omit the fact that a lot of the projects funded by the stimulus were capital intensive and required lots of construction materials. Some or most of the money did not go to paying workers but to purchasing (or using) equipment and materials. A sophisticated reader might realize this but i’d bet a dollar to a doughnut that most people throwing these numbers around don’t. It is not relevant to compare the overall cost of a project per worker required, to the average salary of comparable workers, but that is clearly what many using these numbers are doing. For example, look at ohwilleke above:
“Any way you massage the numbers, the bottom line conclusion is the same. Economic stimulus dollars are much less cost efficient as a way to create jobs than simply hiring people to do whatever tasks you can find without much regard to what benefit those tasks produce in a New Deal style makework program.
At $170K, you can create at least six tolerable jobs, and ten lousy ones. At $278K, you can create proportionately more. Either way, stimulus is a horribly costly and inefficient way to create jobs.”
Heck of a lot of debt for 9.2% unemployment.
DrJim
Good point. They should have looked for spoon ready jobs, to keep the labor/capital ratio up. Oh, and should have immediately banned ATMs.
Rich Berger: I often log series (the GDP series is real, since it is in Ch.2005$) because then, as I explain to my undergrads, the slope is straight when the percentage growth rate is constant.
Brian: I don’t know what you mean by the most academically rigorous work. After all, Taylor used his model and generated his counterfactual in the same ways as these private forecasting firms did. If you mean using DSGEs, you should (1) read Morley’s piece (you have to understand what a HP filter is though), and (2) recognize that other people (such as at the IMF) have used DSGEs and come to substantially different conclusions regarding fiscal policy.
Manfred: It matters when you do the “delta’s”. If you divide the number of jobs six years into the program by the cumulative spending of ARRA, then you will get infinity…
Marcelo Battisti: If Professor Mulligan had thought his original calculation was appropriate, he wouldn’t have characterized mine as “interesting”. I took his discussion as an implicit acceptance that this was the proper way to go (in contrast to his bizarre house analogy). But that’s just me.
Steven Kopits: I dunno; think of all the debt run up during the Bush Administration, and the net job creation during those eight years.
Menzie: Yes, I know what an HP filter is, and I’ve read Morley’s defense of large-scale macro models. I’m far less than 100% convinced by his arguments, though. Macroeconomic Advisors, Moody’s economy.com, and Global Insight all make their money claiming to be able to produce very detailed forecasts of a massive number of variables. Obviously, DSGE models are not geared toward that kind of work, and so Morley, Zandi, and those people have every incentive in the world to push the large macro models. If those models are as useful and credible as claimed, why is Ray Fair the only well-known academic still pushing them (and even his models are much smaller nowadays)? DSGE’s aren’t the answer to everything, but I still find the microfoundations, careful treatment of expectations, etc. still attractive and, in my opinion, the best we have at the moment.
What kind of assessment is this? Who doubted you could have a little bit of effect on short-term GDP when you borrow trillions & the Fed prints like there is no tomorrow?
But what of the costs of paying back the trillions when market interest rates return? The costs of distorted relative prices [investment goods vs. consumer goods]? The costs of a cheapened US$ for everything we import? The costs of the lower efficiency of a larger govt sector & a relatively smaller private sector? The building costs of at the CPI level?
The long run matters much more than the short-run & it is all Keynesian hangover. Are those costs all invisible to you?
Brian: I think you mis-apprehend Dr. Morley’s research; he is not pushing large scale macroeconometric models; rather he is (in line with me) a believer that one has to be extremely careful in the treatment of trends with macro time series. If you examine Morley’s cv, you’ll see that he himself works with small scale models.
I think you have to be careful of what metric you are using when you talk about “best”; microfounded models might be better when speaking of welfare analysis, maybe not so for prediction.
Dr Chinn:
If the micro-founded model that is best for welfare analysis is not consistent with what we observe (i.e., not best for prediction), are we at a loss when discussing the normative benefits of the ARRA? If so, does it make sense to simply acknowledge that we just don’t know because our economic analysis tools are not yet up to the task? Also, in assessing the benefits of the ARRA on dGDP and dJobYears, shouldn’t we include a longer time horizon to accomodate the possibility of the Keyensian hangover? I know we will have to wait a few years for the actuals vs counter-factuals, but shouldn’t we always analyze the whole time horizon, not just impulse-response to date?
@Manfred:
It depends on what you are looking at. The change in the job number per time unit can be seen as a flow. But when you only take the difference of the job numbers disregarding the time period, it’s a stock. A stock divided is still a stock.
Imagine an economy with Zero jobs. Then 3.5 million jobs are created within two years. The flow is 1.75 million created jobs per year. But once they have been created, are those 3.5 million jobs in the economy a flow or a stock? You say yourself the number of jobs in the economy is a stock. And so are the 3.5 million, if the starting point is not Zero.
Professor Chinn how does Professor Taylor’s opinion in the 7-21-2011 edition of the WSJ square with government policies over the past couple years or the CEA assessment, or is this one of the opinion items you warned against citing?
@rootless and @Menzie:
Rootless: in your hypothetical example you start with 0 jobs and end up with 3.5. The flow and the stock are the same, because of the construction of the example (specifically, the construct of the starting point). But that does not mean that the 3.5 claimed by the stimulus are *not* a flow; they are. Because the 3.5 were “created or saved” over a certain period of time, which is the definition of a flow.
Menzie: Honestly, I do not get your comment. I am not sure what the deltas have to do with it. And checking back your original post/criticism to Mulligan’s argument in 2009, there is no mention of deltas.
Bryce, Bryce, Bryce……..we are all dead in the long run (if we ignore the chillen…).
Why are you peeing all over Menzies myopic fairy tales?
I did not intend to defend Bush. I am prepared to praise Clinton, or even Carter, if you like.
Notwithstanding, to run up a couple of trillion dollars in debt, be caught in a serious fiscal crisis and have unemployment firmly stuck at 9.2%–well, it’s difficult to cast that as success. As it is, it’s very sobering situation.
And, today, initial unemp claims at 418k. 15th week at 400k+.
I have been working on some oil numbers, and you know, there could be something of a crunch in the fall. If that’s true, it may well be enough to knock over the economy.
You’ll recall, I suggested some months ago that we need to have a better understanding of the implications of a near-term oil shock wrt to GDP, employment, fiscal balance, and financial system stability. Let me repeat my call for such an analysis now. We can debate thereafter whether that requires additional stimulus or pre-emptive austerity, but I believe we need to get some scenarios on paper so at least we have some sense of what might be headed our way.
@Manfred:
So, according to you, whether it’s a flow or a stock depends on what the offset at the start is. For an offset of Zero, the 3.5 million are a stock, for an offset > Zero, the 3.5 million are a flow.
This doesn’t make any sense.
As I said, it depends on what you are looking at. If you are looking at the change per time unit, then it’s a flow, but if you are only looking at differences between job numbers w/o considering the time period, it’s a stock. A fraction of a stock variable is still a stock variable.
MarkOhio: We use models for many reasons, sometimes for welfare analysis and sometimes for prediction. A VAR-based forecast might have a smaller RMSE than a DSGE based prediction, but maybe not so much as to invalidate the use of the DSGE.
Not sure how to do the formal welfare analysis of debt. (Actually, depending on how money is incorporated in the model, welfare analysis can be difficult — money in the utility function for example.)
Rootless: (and Manfred) Think about the calculation again. Suppose we evaluate the ARRA 10 years down the line; the number of jobs created relative to baseline is zero. The cumulative expenditure was $787 billion. $787 billion divided by 0 jobs equals infinity per job. All Mulligan (and Weekly Standard) did was to do the calculation 2.5 years in, rather than 10 years in. Does that simple math make the problem with the calculation clear?
AS: I much prefer Taylor’s intermediate macro textbook (w/Robert Hall, then w/David Papell), to his op-eds. And his academic papers.
DCramer: My thoughts on the Taylor macroeconometric assessment of the stimulus are here. Other discussion of the Smets-Wouter model is here, and here. The latter highlights how modifications change the results.
@rootless and @Menzie:
rootless: it is perfectly possible for the flow variable and the stock variable to be of the same magnitude, if the starting point of the stock variable is zero, as in your hypothetical. Another example: if my capital stock is 0 and I invest 3.5 during a certain amount of time, I end up with 3.5 in capital stock. My flow variable (investment) and my stock variable (capital stock) are of the same value. There is nothing mysterious or wrong about it.
Menzie: if, hypotheticaly speaking, the ARRA created 0 jobs during a certain period of time, then, yes, the cost of *per job* of the ARRA is infinite. Again, nothing mysterious about this. In economics we do have infinite costs. For example, if something is not technically feasible to produce, the modelling equivalent is to say that the cost is infinite to produce it. Another example is an autarchic economy; the modelling equivalent is that it has infinite tariffs.
If zero jobs were created by the ARRA, the cost was infinite, and thus, the worse the case in favor of the ARRA, but nothing mysterious.
Of course, you may have computation complications; that’s too bad, but that is just a complication in the empiricist’s life, not a mistake, as you argued.
I still do not see this as a refutation of Mulligan. A complication in computation? Perhaps. But not a refutation.
@Manfred:
I’m not aware that I said it wasn’t possible. So I don’t know what you are trying to refute.
What I said was that not just the total of a stock variable is a stock, also any fraction of the stock variable is a stock. Once they have been created, the 3.5 million jobs are a fraction of the total number of jobs in the economy. The total number of jobs is a stock. So are the 3.5 million. Or, the other way around, if you say the 3.5 million can’t be a stock because they have been created over a time period, then, consequently, also the total number of jobs can’t be a stock, since the total number has been created also, although over a much longer time period. Either the total number of jobs in the economy and the 3.5 million are both a stock, or none of them is, if you don’t want to violate logic, because it’s the same type of variable.
@rootless:
I was trying to refute your statement “This doesn’t make any sense.” It makes perfect sense to have the flow variable and the stock variable of the same magnitude, when the starting point of the stock variable is 0.
Yes, at the end of the period, the amount of jobs created during that period become part of the stock; in the same way that investment becomes part of the capital stock. And if 0 is my starting point, both the flow and the stock variable are of the same amount (as I am not getting tired of repeating). And it does not matter what unit of time period you take; could be 6 months, 2 years, 10 years; just be careful in using comparable data for the same time frame.
Menzie was trying to rebutt Casey Mulligan’s argument by saying that the 3.5 million jobs are a stock. I say, no, they are not a stock, they are a flow, because those are jobs created during a certain period of time, and thus, ARRA (a flow) and the 3.5 million (also a flow) can, in principle, be compared.
Dr Chinn:
I would appreciate any links to research you (or others) have evaluating the predictive validity of DSGEs. Can’t we evaluate debt by modeling it as reduced future consumption, using an appropriate discount factor? I am sure it is complicated, but seems like incorporating debt into a formal welfare analysis is something that any overlapping generations model could accomodate. Putting the formal welfare analysis aside, I still think it is important to divide the integrals at time T, not time T-n. You showed Cumulative ARRA Cost per JobYear for T-n, not T.
Ted nugent. Palin should pick him for veep.