When Fiscal Multipliers Look Bigger Than They Really Are

Survey-Based Evidence Can Overstate the Power of Fiscal Stimulus

A new working paper by Martin Eichenbaum, João Guerreiro, and Jana Obradović has been making the rounds in macroeconomic circles. The paper, Ricardian Non-Equivalence, asks a simple but important question: When the government sends households money, do people really think about the future taxes needed to pay for it?

Their answer is mostly no, and that’s an important contribution.

Using a large survey of 6,000 U.S. adults, the authors show that people plan to spend roughly the same amount of a government transfer whether it is framed as an individual rebate or as a universal, deficit-financed payment. Only when respondents are explicitly told that their taxes will rise in the future do planned spending levels fall. In short, many people behave as if government debt is “free money,” at least in the short run.

That behavioral insight is both plausible and well-supported by the survey evidence. The authors then embed this finding into a modern macroeconomic model and conclude that fiscal stimulus, both transfers and government spending, can have much larger effects on economic output than standard models predict. In some cases, they even find government spending multipliers above one, meaning an initial change in spending can lead to a larger overall change in economic output.

That’s a significant finding, but before we update our beliefs about the multiplier effects of fiscal policy, it’s worth slowing down and asking a careful question: How much of this result reflects real-world behavior, and how much reflects the methods and assumptions used to measure it?

A Plausible Mechanism, Carefully Shown

To be clear at the outset, the paper makes a valuable contribution. The core mechanism is intuitive: If a significant share of households fails to internalize future taxes—that is, to factor future tax payments into today’s decisions—then government transfers will feel like an increase in wealth, leading to higher short-run spending. That raises aggregate demand and amplifies the effect of fiscal stimulus.

The authors also deserve credit for directly measuring this mechanism rather than assuming it. Their survey design is thoughtful, and the finding that stated spending plans barely change between individual and universal transfers is genuinely informative. At a minimum, the paper adds strong evidence against the strict textbook version of Ricardian equivalence, which is the idea that households fully account for future taxes when government debt rises.

The concerns begin not with the mechanism itself, but with how far we can push it quantitatively.

Hypothetical Spending Is Not Actual Spending

The most important methodological issue is hypothetical bias. The survey asks respondents how they would allocate a hypothetical $1,400 transfer. There is a large literature showing that stated intentions in hypothetical scenarios do not reliably map into real-world behavior. Sometimes people overstate spending; sometimes they understate it. Either way, planned spending is not the same as actual spending.

This matters because the authors calibrate their macroeconomic model directly to these stated spending responses. If responses to hypothetical scenarios overstate real consumption, even modestly, the model will mechanically overstate the fiscal multiplier. The authors acknowledge this literature and cite recent work suggesting survey marginal propensities to consume (MPCs) can be informative, but hypothetical bias remains a first-order concern, not a minor technical detail.

Who Is in the Survey Matters

There is also the issue of representativeness. The survey was conducted on Prolific, an online platform that is not a probability sample. The authors do impose demographic targets, but the sample still skews younger and includes a disproportionate number of unemployed respondents. Adults over 65 (one-in-five US adults) are excluded entirely.

This matters because spending behavior, financial sophistication, and exposure to fiscal policy all vary substantially by age and income. Older households, in particular, tend to have lower marginal propensities to consume and more experience with taxation. If such households are underrepresented, the estimated average spending response will be biased upward.

Low compensation and short completion times also raise the risk that some respondents were not carefully processing the scenarios presented to them. When the object of interest is “attention” to fiscal details, inattentive survey responses become especially problematic.

Framing, Salience, and What “Taxes” Mean

Another issue is framing and salience of the survey question. In one treatment, respondents are explicitly told: “your taxes will rise by $1,400 next year.” That is far more concrete and attention-grabbing than how taxes actually rise in practice: gradually, diffusely, and often indirectly.

Conversely, in the universal-transfer scenario without explicit tax information, respondents may reasonably interpret the payment as something that will be financed by “the government” or “future Congresses,” not by them personally. These framing effects can easily change stated behavior without reflecting a deep or stable belief about government budgets.

The survey itself shows that respondents dramatically underestimate government debt and spend very little time following fiscal policy. That respondent behavior supports the authors’ mechanism, but it also implies that many respondents may not have a coherent understanding of what “future taxes” really entail. Changes in survey answers may reflect confusion or anchoring i.e., respondents adjusting answers around salient reference points created by question framing rather than genuine intertemporal reasoning.

Calibrating a Model to One Moment Is Risky

Another concern is quantitative and comes from how the behavioral parameter is calibrated. The authors choose the model’s “cognitive cost” parameter—meant to capture how costly it is for households to pay attention to future taxes—so that the model exactly matches the survey’s planned spending response in one scenario.

This is effectively a one-moment calibration, that is, the parameter is pinned down using a single survey response in a single scenario, of a parameter that governs attention to taxes, income, interest rates, and future dynamics more broadly. If the survey MPC (the share of an extra dollar that respondents say they would spend) is biased, even slightly, the entire model response scales up or down accordingly. This approach may overlook the influence of other parameters, such as time spent on economic news, stated tax expectations across horizons, and evidence from actual stimulus payments.

Without that, the risk of overfitting (that the model is tailored to one survey outcome rather than underlying behavior) is substantial.

Assumptions That Push Multipliers Up

Finally, several modeling assumptions tilt the results toward larger multipliers. Debt is assumed to be repaid slowly, which increases the perceived distance of future taxes. Capital and investment are excluded, ruling out standard crowding-out effects. And the analysis relies heavily on linearized dynamics, which approximates large, complex policy changes using small, simplified relationships and can overstate responses to large shocks.

There is good reason to be cautious here. Other work has shown that once nonlinearities, capital accumulation, and realistic financing are introduced, fiscal multipliers tend to fall, sometimes substantially.

A Balanced Takeaway

All said and done, the paper convincingly shows that many people do not naturally think about future taxes when asked about government transfers. That is an important behavioral fact, and macroeconomists should take it seriously.

But translating that fact into large, reliable estimates of fiscal multipliers requires stronger evidence than hypothetical survey responses and a tightly calibrated model can provide. The mechanism is plausible, but the magnitude is uncertain.