
Valerie Ramey on Stimulus and Multipliers 10/24/2011
What this covers
This is a conversation between Russ Roberts and Valerie Ramey of UC San Diego about how economists measure what government spending does to output and jobs. Ramey's own work treats wartime buildups as a source of exogenous variation. The talk keeps returning to why that evidence travels poorly to a modern recession. World War II and Korea came with a draft, rationing, and price controls, all of which distort measured consumption and employment. She also notes that most of the job gains from spending are government jobs, which shifts the policy question from GDP to employment.
The range is wide. Ramey points to the postwar demobilization, when spending fell 60 to 70 percent and unemployment still rose only to about 5 percent, against Keynesian forecasts of depression. She reads falling consumption after military spending as a resource constraint rather than a multiplier story, and Roberts brings in Robert Higgs on the New Deal. On taxes, she discusses Christina Romer and David Romer's very large tax effects, how anticipated tax increases change behavior early, and Edward Prescott's account of the US–Europe work gap, as refined by Richard Rogerson through the extensive margin.
The last stretch turns to method. Ramey explains how vector autoregressions can mislead when misspecified, and why the counterfactual is so hard to establish. She and Roberts then move to the Lucas critique, Sargent's and Sims's contributions, and Hayek's pretense of knowledge. Ramey defends modest, parsimonious models as better than shooting from the hip.
Ramey argues that government spending multipliers are imprecisely estimated (roughly 0.8 to 1.2 for military spending, with much wider ranges across studies), that anticipation and measurement problems distort the estimates, and that stimulus is therefore a poor bet, especially once future tax increases are counted.
- Estimates of the spending multiplier vary widely within and across papers because of sample period, specification, and data source choices
- Anticipated spending and tax changes shift behavior before the spending occurs, so standard baselines measure the wrong thing
- Consumption and private investment tend to fall after military buildups, which fits classical crowding-out better than Keynesian stimulus
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
After World War II, government spending fell by 60 to 70 percent and military employment dropped from about 12 million to 3 million, yet the civilian labor force grew by 5 million and unemployment rose only to about 5 percent, which contradicts the Keynesian forecasts of a severe depression.
“government spending falls by sixty seventy percent the Keynesian of the day ... predict a horrible contraction a depression”
Americans work far more than Europeans over a lifetime, and Edward Prescott attributed this gap to higher tax rates in Europe; Richard Rogerson later found that the difference shows up mainly in movements in and out of the labor force for people in their fifties and sixties and for the young, not in prime-age hours.
“Edward Prescott first made this point why do Americans work so much more than Europeans and he argues that it's because the difference in the tax rate”
Robert Lucas argued that policy effects cannot be estimated by regressing historical data without accounting for whether a policy is expected to be permanent or temporary, and Tom Sargent developed methods to estimate deep structural parameters that allow policy analysis outside historical experience.
“previously Robert Lucas who won in a well-priced several years ago argued that it did not make sense to analyze the effective policy by simply running a regression”
Robert Higgs argues that the New Deal did not end the Great Depression; the economy recovered in 1939–41 as the government mobilized for war, but private consumption fell because wartime forced people into the army and rationed goods.
“what Higgs argues is that private consumption went down and I think it's very important to emphasize that this endless claim that that war is good for the economy”
Estimating the effect of stimulus requires a counterfactual of what the economy would have done without it, and that counterfactual is extremely hard to establish; the same data are consistent with stimulus raising unemployment, having no effect, or reducing it.
“the unemployment rate would have been 12% yeah or it could be that the stimulus package didn't do anything and the unemployment rate is exactly what it would have been”
Government spending in wartime is a resource constraint: more tanks mean less butter, so a classical argument explains consumption falling after military spending, without needing a neoclassical or Keynesian multiplier story.
“government spends more or less there's fewer resource available for other things and so by definition people are going to have to consume less have more tanks there's less butter”
Prime-age full-time male workers do not change their hours much in response to tax rate changes, so large tax effects must come from the extensive margin (entry and exit from the labor force), particularly at the ends of the life cycle.
“if you look at a typical full-time employed male we know that they do not shift their hours much in response to changes in tax rates”
Estimated spending multipliers vary as much within individual papers as across papers, because of sample period choices (such as including or excluding the Korean War) and specification choices, so a single study can produce a very wide range.
“the range within papers is often as big as the range across paper explain it's a very important planet yes really because it often depends on the sample period”
Most of the employment increase from government spending is government employment, not private employment; the policy question has shifted from GDP multipliers to whether government spending can create jobs.
“don't you find that most of the employment increases government employment not private yes”
Tax multipliers estimated by Christina Romer and David Romer are between -2.5 and -3, meaning a one-dollar tax increase lowers GDP by two and a half to three dollars, far more than the effect of spending increases.
“the leading estimates are say the ones by Romer and Roemer who find tax multipliers of between minus 2.5 and minus 3”
The main criticism of military-spending multiplier studies is that they rely on World War II and Korea, periods with a draft, rationing, and other concurrent shocks, so their historical averages may not apply to a policy of two years of stimulus during a recession today.
“the key criticism is we don't want to know what the multiplier is during a period when there's a big draft”
Measured consumption during wartime is unreliable because of price controls and rationing: goods appear artificially cheap in official price data, and some goods were unavailable at any price, so raw GDP and consumption series overstate wellbeing.
“the personal we have price controls there's a lot of stuff that looks artificially cheap there's stuff that wasn't actually available you couldn't even buy it because of rationing”
Economics can build parsimonious models that are reasonable approximations of the economy and give more insight than shooting from the hip, even though it cannot be as precise as physics.
“we cannot be as precise as we would like say as a physicists are however we can come up with parsimonious models that are reasonable approximation”
Alesina and coauthors find that fiscal contractions expand the economy, suggesting that the fiscal multiplier can be negative, consistent with the post-World War II demobilization experience.
“recent work of ala Sina and others that actually finds that government contraction expands the economy”
After a military spending increase, consumption falls in the short run and then recovers as government spending declines; this pattern is consistent with the classical model and inconsistent with the traditional Keynesian model.
“I find that in the short run consumption actually Falls a little bit it's for the most part statistically significant”
Vector autoregressions can produce misleading results if they omit the proper variables or lags: simulation tests show that a VAR run on artificial data from a known model can be very far from the true response, so VARs are not reliable without careful specification.
“if you don't estimate the vector autoregression correctly if you don't include the proper variables or the proper lags in these you can get results that are very misleading”
Per-head taxes would have little differential effect on the economy compared with deficit financing, but actual taxes on income and profits are distortionary and so the future tax cost of deficit-financed spending is large.
“if taxes were the kind of thing that we're sort of per head taxes where if you're alive you pay $100 sort of tax it's not clear that there would be much difference”
During World War II and the Korean War, much of the measured employment increase reflects the draft and the rise in military personnel rather than productive private jobs.
“when I look at the number who were in the military a big part of the increase in employment is just an increase in the number of people in the military”
Whether the Korean War is included in the sample can change the estimated multiplier, because the same period of rising spending may have had a small multiplier or may have occurred alongside tax increases, which makes the two explanations hard to separate.
“the problem is does Korea have a smaller multiplier because the multiplier is small or because taxes were increasing”
Applying historical multipliers to the 2009 stimulus requires that the structural relationships of the economy today match those in the past, and CBO's own counterfactual forecasts were grossly wrong, suggesting that the recession was unusual because of debt overhang and destroyed household and business balance sheets.
“the CBO made their own estimates ... those were grossly wrong they confront that too because they're honest”
Anticipated future tax increases can depress the economy today: a planned sales-tax increase leads people to pull durable-goods purchases forward, and businesses respond even more strongly, so the expectation alone changes behavior before taxes rise.
“you would get thinking that taxes are going to be higher next year can actually burn the economy this year”
Hayek's 1974 Nobel address argued that economics suffers from a 'pretense of knowledge': the economy is too complex to model precisely, so confident quantitative policy estimates are illusory.
“the economy is not of course you can master in college to think otherwise is the pretense of knowledge”
The standard approach uses the change in government spending as the shock and therefore picks the wrong baseline; measuring from the date when news of a future spending change arrives yields a very different and more accurate estimate.
“the standard way of doing this takes the change of government spending itself to be the the dues it about the few you know the news”
Government spending decisions are simultaneous with monetary policy, tax policy, and regulation, and causation runs in both directions (a weak economy can prompt stimulus, and stimulus can be mistaken for the cause of a weak economy), so correlation between spending and output cannot be read as causation.
“you've got causation running in both directions you could have the economy doing really poorly encouraging stimulus leading you to a false conclusion”
Research by Christopher House and Matthew Shapiro suggests the phase-in of certain business-related Bush tax cuts slowed the recovery from the 2001 recession because firms delayed investment while waiting for the scheduled cuts.
“some of the work by Christopher house and Matthew Shapiro shows that the phase-in of some of the Bush tax cuts the particular parts on business actually led the recovery from the 2001 recession to be slower”
Even with a range of multipliers from 0.5 to 2, people can draw reasoned policy conclusions, because the maximum plausible multiplier is about two and even that does not make a short-run stimulus worth its long-run tax costs.
“most people don't believe that we think that probably the maximum is two and even with the multiplier of two it is not clear that it's worth passing a stimulus”
Investment rises initially after a military spending increase because firms anticipate higher government demand, but then falls as input prices (such as steel) and labor costs rise and crowd out private investment.
“investment actually goes up at first and then Falls and that's actually also consistent with the neoclassical model”
Assumptions and specifications drive most of the variation in fiscal multiplier estimates, so a range spanning four or six-fold cannot support confident policy, unlike a physics equation that engineers rely on.
“once you know one view one thing this teaches us of course is that assumptions matter a lot”
Labor supply responses to taxes must be assessed over a whole lifetime rather than a single quarter: high-productivity workers work very hard while employed and retire earlier, so they look unresponsive in a short window but still respond to incentives.
“it's in many ways its life time hours that matter not this particular quarter”
Auerbach and Gorodnichenko find a spending multiplier near zero in expansions and about 3 in recessions, but this result depends on a regime-switching specification: when the model lets a spending increase move the economy out of recession, the recession multiplier falls to about 1.5.
“when they modify their model to change that effect then suddenly their top multipliers during recessions are 1.5”
Military spending is the best candidate for an exogenous fiscal shock, because large buildups driven by foreign events are not caused by the state of the economy, although normal-year defense budgets do respond to the deficit and must be filtered out.
“during normal years how much they they cut the defense budget is related to how much of the deficit there is so we want to make sure to take those parts out”
Blanchard and Perotti's standard VAR estimates of government spending shocks ignored anticipation, so Ramey augmented the Sims-style VAR with a news variable and found that output responds when news arrives rather than when spending changes.
“Olivier Blanchard who's now the chief economist for the IMF and Roberto Perrotti had used a standard thin vector autoregression”
Because future taxes on income or profits are distortionary, government spending financed by deficits must be unusually productive, and its short-run benefit must be large enough to justify the long-run tax cost, even when unemployment is at 9 percent.
“government spending better be doing something really productive if it's going to be financed”
Anticipated large increases in military spending (such as the lead-up to World War II, the Korean War, Vietnam, the Carter-Reagan buildup, and 9/11) change consumer and firm behavior before the spending actually occurs, so the economy starts responding immediately.
“even though government spendings cannot increase much in the short run people know it's going to increase later and firms start gearing up consumers start responding”
Keynesian economists after WWII, including Paul Samuelson, predicted a depression as government spending was cut, and they explained the job boom afterward with pent-up demand; but the pent-up demand was constrained by scarce steel and goods, not by a lack of consumer willingness to spend.
“the reason ever spent up to me and is because you couldn't you couldn't buy stuff not because of rationing but because the steel wasn't available”
The CBO's claim that the 2009 stimulus created 3.5 million jobs is misleading: the CBO reused its prior multipliers with actual spending levels, effectively redoing the forecast rather than measuring what happened, and it openly acknowledged that it did not estimate the counterfactual.
“the CBO found that the stimulus worked and they estimated created say one point x2 3.5 million jobs”
Economists are not very good at estimating the fiscal multiplier, and the profession's estimates are heavily shaped by ideological priors, which is awkward for a field that claims empirical grounding.
“we're not very good at this as economists we should admit it but the fact is you have people who think the multiplier is really big”
Rationing and the switch of auto plants to tank production meant consumers could not buy cars at any price, and Ramey's consumption growth remained slower even before rationing started in spring 1942; excluding World War II but including Korea still shows falling consumption.
“even when I omit World War two and just to have Korea and on I'm finding the decrease in consumption as well”
Fiscal policy research was considered a backwater in economics from the 1970s until the late 1990s, but a resurgence of studies with conflicting multipliers now provides a learning process that may narrow the range over time.
“from the 1970s until just a few years ago studying fiscal policy was considered a backwater in economics”
Using a news variable dated from Business Week reports of anticipated spending changes, Ramey found that the economy started responding right away, and she identified military spending episodes driven by external events as the cleanest natural experiment.
“I went back to Business Week for the most part but also New York Times and look at where Business Week was predicted a major change”
Consumption fell during wartime even after excluding World War II and including only the Korean War period, so the consumption decline is not an artifact of the rationing period alone.
“even when I omit World War two and just to have Korea and on I'm finding the decrease in consumption as well”
A 2011 economic history paper attributes the rapid post-war job creation to a surge in US export demand because European capital stock had been destroyed, but Ramey is skeptical of ex post explanations because the same outcome could be attributed to other drivers.
“it was export demand because Europe much of Europe capital stock was destroyed the US had a big surge in its export markets”
Roger Farmer argued that if the Pentagon got estimates of 0.5 to 2 for a nuclear submarine's key physics equation, it would not build submarines, so economists should not expect policymakers to act on such imprecise multipliers.
“can you imagine if the Pentagon called over to the physicist and said you know what's the key physics equation”
Romer and Romer's tax multipliers may be large partly because their tax changes mixed labor income taxes with taxes on firms, such as changes to investment tax credits, so the negative effect may run through reduced business formation and job creation.
“some of them could have been taxes on firms variations and investment tax credits so these tax effects could be working through job creation”
Estimates of the spending multiplier can vary with how employment is measured: Ghani-style cross-state work found very different job multipliers depending on whether employment came from an establishment survey or a household survey.
“whether they used one employment survey that looked just at establishments versus a household surveyed”
Ramey's estimates of the spending multiplier from military buildups range from 0.8 to 1.2, with large standard errors, so the estimate is imprecise and could plausibly be well above or below these values.
“I find estimates that range from 0.8 to 1.2 so one is right in the middle there”
Robert Barro's estimate of the spending multiplier is around 0.6, which together with the large future tax costs suggests that the stimulus is not a good deal, while critics who find larger multipliers have strong reasons to challenge the tax estimates.
“let's take the lower bound of your work the 0.8 or the Barrows of 0.6 meaning there's a lot of crowding out”
Economists cannot run controlled experiments on the economy, so they must rely on econometrics to approximate a randomized trial, which makes isolating the effect of government spending very hard.
“the main reason that is difficult to do is that economists are not allowed to run experiments on the economy”
Chris Sims made a major contribution with vector autoregressions, which identify unanticipated changes in government spending or taxes and trace their average effects on GDP and consumption over time, though these methods are difficult and sensitive to specification.
“Chris Sims was also working on this but came up with a nice way to sort of summarize data in something called a vector autoregression”
Hayek criticized 'scientism', the use of the trappings of science such as jargon or statistics on economic questions that may not suit them, a view the economics profession largely rejects.
“Hayek was very critical of what he called scientism the use of what looked like science either in terms of jargon or statistical techniques”
Expectations and timing are two special challenges in measuring spending effects: expectations are never observed directly but shape behavior, and the timing of when expectations form determines the correct baseline for measuring effects.
“expectations which are never observed directly but of course matter people look ahead and the other is timing”
The fiscal multiplier is the change in GDP produced by a one-dollar change in government spending; a multiplier above one is needed for short-run stimulus to be effective, while a multiplier below one implies crowding out of private spending.
“the multiplier is a concept that comes from the good old fashioned traditional Keynesian model”
Roberts challenges the large negative tax effects: if prime-age male labor supply is relatively inelastic to net income, it is unclear why the tax multiplier should be so large.
“given that labor supply is certainly for men it's less so for women but certainly for men is relatively inelastic”
Economists know more than they did eighty years ago, but the 1930s macro debates were frustrating because participants could not agree on basic terms such as investment, and the field has progressed more in identifying what does not work than in settling magnitudes.
“economic debates in the 30s about macroeconomics were so frustrating because people couldn't even agree on what the terms meant”
Economists lack a good understanding of why the labor market sometimes makes job-finding easy and sometimes very hard: in 1999–2000 skilled web workers found jobs easily, but by 2002 and before the 2007 recession, job-finding had become much harder.
“we don't really understand those differences I always used the tech bubble or whatever you want to call it”
Ramey built real consumption and GDP series back to 1939 by deflating nominal series with consumer and producer price indexes, and her independent construction closely matched Robert Gordon's series for the same period, though neither handled wartime rationing directly.
“we shared the data and started plotting them against each other and we're astounded how close the two data series were”
Robert Hall's medicine analogy: economics has learned that bloodletting-style leeches do not work, and like medicine a century ago, economics has progressed through scientific methods even though the economy is as complex as the human body.
“my adviser Robert Hall when he taught would make the analogy to medicine that right now economics we finally figured out that putting leeches on the it's not a good idea”
Real wages, measured as worker compensation divided by firm output prices, fall after military spending increases, and hours worked rise, so the employment gains come with lower real pay.
“hoursworked the increase employment increases but real wages measured at the firm level so it's sort of the way to the worker divided by the price of your output those actually go down as well”