2008 Stimulus Checks Raised Saving More Than Consumption

Valerie Ramey treats the 2008 U.S. rebate program as a useful test because the payments were large, concentrated in time, and plainly visible in aggregate disposable-income data. Enacted in February 2008, the program distributed roughly $100 billion, which she describes as about 11% of aggregate disposable income in the preceding January. The average check was $1,000. Payments went out mainly from April through August, with about half arriving in May.

Disposable income showed a large, temporary spike when the checks arrived. Aggregate consumption, Ramey says, did not show a comparable increase—at most, a small “gopher hill.” The saving rate rose sharply instead.

Martin Feldstein and John Taylor reached a similar conclusion from the macro data soon after the rebates. As Ramey recounts it, they regarded the program as a failed stimulus effort and invoked Milton Friedman’s permanent-income hypothesis: recipients of a temporary payment may spend some of it, but are likely to save much of it and spread their spending over time rather than treat it as a lasting increase in resources.

The later household-level evidence appeared to tell a different story. Jonathan Parker and collaborators added questions to government and private surveys asking households whether they had received a rebate, when it arrived, how large it was, and how they spent it. The data included large government-survey samples and spending records from sources such as Nielsen. Using econometric methods that were standard at the time, the studies estimated marginal propensities to consumeThe share of an additional dollar of income that a household spends rather than saves., or MPCs, between 0.5 and 0.9 for the 2008 payments.

That is the contradiction at the center of Ramey’s argument. If households spent 50 to 90 cents of each rebate dollar promptly, the program should have left a much clearer footprint in total consumption than she sees in the aggregate record.

In New Keynesian models with heterogeneous households, high MPCs imply that temporary transfers can have substantial macroeconomic effects. One household spends a check; that spending becomes income to another household; additional spending rounds can amplify the initial payment. In the simplest Keynesian formulation, an MPC of 0.8 implies a multiplier of five, though Ramey notes that modern Keynesian models generally produce smaller multipliers because they include other relevant forces.

Her work with Jake Orchard and Johannes Wieland uses historical “plausibility case studies” to ask whether the aggregate implications of high household MPC estimates can actually be seen. Their conclusion on the 2008 rebates is that the multiplier was below 0.2.

Ramey says researchers and policymakers gave the micro estimates substantial weight partly because of the “credibility revolution” in applied microeconomics. Household-level identification strategies came to be viewed as more credible than macro time-series analysis, she says, and the apparent strength of those methods made it easier to discount the aggregate evidence. In her account, that also helped make much larger transfer programs seem defensible when COVID arrived.

She has not yet applied this framework specifically to COVID-era payments. The problem, she says, is disentangling transfers from vaccination rollout, reopening, and people becoming more willing to leave home without the same fear. That is a separate historical-identification challenge.

Ramey and her coauthors test the micro-macro gap by constructing a counterfactual path for consumption. They take the actual timing of aggregate rebate payments, multiply it by an assumed MPC—first 0.5 and then 0.9—and subtract the resulting implied rebate-financed spending from observed consumption. The result is an estimate of what consumption would have looked like had the rebates not occurred.

This initial exercise is accounting rather than a complete Keynesian model. It does not include secondary effects from one household’s spending becoming another household’s income. But Ramey argues that the limited exercise is already difficult to reconcile with conditions in the summer of 2008.

With an MPC of 0.5, she says, the implied no-rebate consumption path forms a pronounced V: consumption would have fallen and then mostly recovered by September, when Lehman Brothers was failing. With an MPC of 0.9, subtracting a larger rebate-financed spending response produces an even bigger V.

Ramey says the researchers compared the implication with pessimistic Goldman Sachs and Federal Reserve forecasts and could not find forecasts bleak enough to make the implied counterfactual plausible. They also accounted for the summer’s rise in oil prices, which could have dampened consumption, but she says that adjustment still did not yield anything resembling the decline implied by the high-MPC estimates.

Adding Keynesian multiplier effects would make the mismatch larger in her account, not smaller. Those mechanisms magnify the initial spending response, so the counterfactual V-shapes become more pronounced. Her claim is therefore not merely that aggregate data are noisy. It is that payments of this scale, delivered over a short period, should have generated a more conspicuous consumption response if the reported household MPCs had been as large as 0.5 to 0.9.

Valerie Ramey says later econometric innovations exposed a source of bias in the earlier micro estimates. The methods standard when the high MPC estimates were produced had, she says, implicitly used the wrong control group. Borusyak and Jaravel identified that issue and, using improved techniques on one broader set of estimates, found much smaller MPCs for nondurable spending.

Ramey and her coauthors re-examined Consumer Expenditure Survey data, where some particularly large MPC estimates had originated. Applying newer methods, identifying additional biases, and developing corrections of their own, they estimated an MPC of 0.3 rather than 0.5 to 0.9. The observed spending response was concentrated in motor vehicles.

The lower estimate reduces the discrepancy but does not resolve it, in her view. Even an MPC of 0.3 generated a counterfactual that still looked implausible, particularly in the motor-vehicle sector.

Her further explanation concerns prices. Real motor-vehicle prices had generally been declining, she says, but their relative price rose during the period in which rebates were paid. Recipients with checks in hand may have increased demand for cars, allowing dealers to raise prices. Those price increases absorbed part of the additional purchasing power. Consumers whose cars had not failed and who could postpone a purchase also had reason to wait rather than buy at temporarily elevated prices.

The fact that the spending response was concentrated in durable goods also affects how Ramey interprets the permanent-income hypothesis. Buying a car with a rebate does not necessarily violate the hypothesis, she says, because a car delivers consumption services over many years and can partly function as an asset. The hypothesis bears more directly on nondurable consumption.

Taken together, lower estimated MPCs and relative-price adjustments are, in Ramey’s account, more consistent with the modest aggregate consumption movement than the original high household estimates. They are also central to her conclusion that the 2008 rebate multiplier was below 0.2.

Valerie Ramey sees a related divergence in Singapore, where an American Economic Review study examined one-time 2011 payments to citizens but not foreign residents. Foreign residents made up 40% of Singapore’s population, according to Ramey, creating a comparison group. Using household spending data from a bank, the study compared citizens’ spending with that of noncitizens and estimated a cumulative MPC of 0.8 within 10 months.

Ramey characterizes the design as the sort economists typically value: rich data, a natural experiment, and an apparently strong control group. But she found that similar citizen payments had been used repeatedly across roughly two decades, often around elections. A few months before an election, the government would announce a dividend payment for citizens—who were also voters—and pay it out around the election period.

That repetition gave her an opportunity for a time-series plausibility exercise using aggregate Singapore data. The payouts rose sharply at the payment date, as did disposable income; income then fell back. One smaller later echo reflected a program in which recipients could cash out only half the payment initially and received the remainder a year later.

If the household MPC of 0.8 described an aggregate consumption response, Ramey says, consumption should have risen visibly. Instead, her estimate for the initial quarters is slightly below zero, if anything, and fairly precise. The saving rate jumps by as much as the payment while consumption shows no corresponding increase.

The Singapore case does not lead Ramey to claim that no individual recipient spent a payment. Rather, it reinforces her view that a large household-level MPC cannot on its own establish a large macroeconomic stimulus effect. In both the U.S. and Singapore cases, she finds clearly visible movements in disposable income and saving but little aggregate consumption response.

Valerie Ramey distinguishes temporary transfers from direct government purchases. A transfer raises a household’s disposable income and depends on the household’s decision to spend. A government purchase—she uses a jet fighter as an example—directly buys output from the private sector. Multipliers on government purchases can therefore be positive, she says, even if multipliers on temporary transfers are zero.

Persistence matters as well. Government purchases can be expected to continue longer than a one-time transfer, and firms deciding whether to expand production care about the anticipated future path of demand, not merely current spending.

The empirical challenge is different, too. Governments may increase purchases in response to recessions, so a simple regression of output on government spending can confound the effect of spending with the weak conditions that prompted it. Ramey says historical military events have been useful because they can provide more plausibly exogenous changes in spending or in expectations about future spending.

Her own research has used military buildups and changes in expected government spending. In joint work with Sarah Zubairy, she studied U.S. data from 1889 to 2015, measuring shocks as changes in expectations about the path of government spending. The material presented here establishes why purchases should not be treated as equivalent to transfers and why military events can help with identification; it does not provide the empirical multiplier findings from that historical work.