Household Wealth Differences Change How Fiscal and Monetary Policy Work

A transfer of $500 does not mean the same thing to every household. Some can save it or spread its value over time; others may spend a substantial share quickly because they have little liquid wealth to draw on. That difference is central to Greg Kaplan’s account of why macroeconomic models need to represent households’ different resources and constraints. It also changes how economists should interpret fiscal stimulus and monetary policy: the effects depend not only on the size of a policy, but on who receives it and what those households can do with it.

The average marginal propensity to consume—the share of an additional dollar that a household spends—conceals a wide range of responses. Kaplan puts the average at roughly 20 to 25 percent. That is much lower than the 80 or 90 percent associated with the Keynesian consumption function as he describes it, but much higher than the small response implied by a household that can borrow and save freely and spreads a small payment over the rest of its life. Some households look closer to the Friedman–Hall account of consumption smoothing: they report that a little extra money would change very little. Others say they would spend a large share of it relatively quickly.

The same difference matters for stimulus payments. A payment arriving today may raise spending more than one expected next month, even when a household knows the later payment is coming. Kaplan and Gianluca Violante tried to reproduce evidence on that timing difference in a model. The evidence came from research by David Johnson, Jonathan Parker, and Nicholas Souleles on stimulus payments in 2001. Their comparison was not simply between recipients and nonrecipients: it compared households that had received a payment with households that expected to receive one in a month or two.

If households had resources readily available, it was not obvious why the date on which the payment arrived should make much difference to their spending. One explanation was that some households had too little wealth to spend before the payment came. But Kaplan says the share with very low total wealth was not large enough to account for the spending responses. A household might instead have considerable wealth in a home or retirement savings, but little money available for immediate use. Adding this illiquidity to the model made it possible to match the evidence.

The model’s implication surprised its authors. Its simulations suggested that many households had almost no liquid wealth. Kaplan recalls that he and Violante initially thought the result could not be right, then checked the data. They found that roughly a third of US households fit the category they called “wealthy hand-to-mouth”: households with wealth in forms such as housing or retirement savings, but little liquid wealth. Kaplan says the 2025 Survey of Consumer Finances still showed roughly a third or more in that category, with another 10 percent or so having essentially no wealth at the end of a pay period.

This distinction is not just a way to describe household balance sheets. It helps explain why timing and the form of a policy matter. A household that cannot readily use its assets to meet current expenses may respond differently to a payment than one that can borrow against its wealth or move money between accounts. A model centered on an average household can miss that difference, and so produce a different account of what a transfer will do.

The same distributional perspective changes how interest-rate policy looks. In a representative-agent New Keynesian model, the central mechanism is intertemporal substitution. A higher interest rate increases the return to saving, so a household shifts spending from today to tomorrow. Kaplan describes this as the Euler-equation channel: the household chooses how to balance consumption across time, assuming it can smooth consumption and is not borrowing constrained.

In a model with different households, that is only one part of the picture. Some households have little liquid wealth or face borrowing constraints; others hold savings, government bonds, or large mortgages. Kaplan says the direct effect of interest-rate changes on spending is weaker and less uniform in heterogeneous-agent models. The effects can even run in opposite directions. A household with a large mortgage is a net borrower and may reduce consumption when rates rise. A retired household holding government bonds may receive higher returns and benefit.

Mortgage structure also affects the link. In countries with adjustable-rate mortgages, a change in rates directly changes many households’ payments. Kaplan says that is one of the strongest direct connections between interest rates and spending. In the United States, where many households have fixed-rate mortgages, the channel is weaker. These differences are part of why a single, direct line from a policy rate to household spending can misrepresent the effects of monetary policy.

Kaplan contrasts this account with the older IS-LM and Phillips-curve explanation many people have in mind: lower interest rates stimulate aggregate demand, more spending raises output, and the additional output generates inflation. In that account, the central mechanism resembles the Keynesian multiplier, with household spending responding strongly. In representative-agent New Keynesian models, by contrast, monetary policy works primarily through intertemporal substitution. HANK models put household spending responses back into a setting where people have different assets, constraints, and propensities to consume.

That does not mean that interest rates have no effect or that stimulus payments do not raise spending. Kaplan’s argument is about the mechanisms and their scale. The familiar story in which rate cuts reliably prompt a representative household to spend more is not the only account, and the effects of policy are distributed unevenly. Models that represent household differences make those mechanisms visible.

Kaplan argues that fiscal stimulus should be evaluated as redistribution, not treated as a free way to raise aggregate demand. He does not deny that payments can provide valuable relief. His point is that the stronger case for them lies in the uneven costs of recessions and the hardship concentrated among some households—not necessarily in the claim that giving money to households will produce a large aggregate-demand multiplier.

A representative-agent framework has difficulty representing a recession that is severe for some households and less damaging for others. A model with a distribution can ask who receives support, how much they spend, and how the policy shifts resources among households. That distributional account, Kaplan says, provides a stronger rationale for payments than the standard multiplier story.

But the resources have to come from somewhere. If the payments are financed by higher taxes or lower transfers in the future, those costs should be considered alongside the relief they provide. The analysis should include the incentive effects of the taxes and the consequences of moving resources between households. Kaplan’s point is not that redistribution is necessarily undesirable; it is that stimulus payments should be assessed by the same standards as other redistributive policies.

He adds that in the last two recessions, the payments have not appeared to be matched by lower future transfers or higher future taxes. In Kaplan’s account, their cost has instead, so far, been met through inflation. Inflation, too, redistributes: it imposes costs on some households and benefits others.

Kaplan’s concern about emergency measures is that their urgency can push these tradeoffs out of view. Officials focus on getting payments out quickly and describing them as aggregate stimulus, rather than confronting who benefits, who pays, and what incentives the policy changes. He therefore favours considering more durable safety-net programs and automatic stabilizers, which require policymakers to make those choices more explicitly. The aim is to move toward a system in which emergency payments are less necessary in the first place.

This is not an argument that emergency payments should never be used. Kaplan acknowledges their role in alleviating hardship, and Jon Hartley raises the possibility that people may want to borrow against future income during a recession, even if that means paying higher taxes later. The issue is how to make the tradeoff visible. A more durable program could still involve costs and incentive effects; its advantage, in Kaplan’s argument, is that those choices have to be considered in advance rather than obscured by the urgency of a crisis.

The distinction also clarifies why different policies can have different effects even when each is described as “stimulus.” A transfer to a household with little liquid wealth may relieve an immediate constraint and produce a different spending response from a transfer to a household with ample available resources. A policy’s aggregate effect cannot be understood separately from its distribution and financing. For Kaplan, models that make those features explicit are useful not because they settle every policy dispute, but because they help identify the tradeoffs a representative-household story leaves out.

Kaplan treats forward guidance and quantitative easing as distinct policy questions. The case for forward guidance grew in part out of representative-agent New Keynesian models. If households are strongly forward-looking and interest rates have a powerful effect on their current spending, then a commitment to lower rates in the future can affect consumption today. Kaplan says this logic is in tension with micro-level evidence and with models that give households shorter effective horizons or constraints on borrowing. In his view, those features cast doubt on the strength of forward guidance as a tool for aggregate-demand stabilization.

The strength of that conclusion depends on the mechanism being discussed. Kaplan is addressing the classic argument that a central bank can stabilize demand by promising future interest-rate changes, in a framework where households respond strongly to rates through their intertemporal choices. He notes that models also contain other assumptions, including implicit policy responses. His point is not that every form of communication about future policy is ineffective, but that the representative-agent rationale does not carry over unchanged when households cannot all adjust spending in the same way.

Quantitative easing, by contrast, cannot be assessed without specifying what assets are bought and what the central bank is trying to accomplish. Kaplan says the familiar form of QE—changing the duration composition of government debt—can move asset prices, but he sees less evidence that it produces large real effects. He says this conclusion is also consistent with many models.

One consequence of asset purchases can be redistribution between households with different exposure to the assets being bought. If those households have different marginal propensities to consume, the shift in resources may affect total spending. Kaplan expects this effect to be relatively small as a way to boost aggregate demand. But the distributional effects remain important: a policy can move resources between households even if its aggregate spending effect is modest.

Kaplan objects to using “QE” as a catch-all term for expansionary monetary policy. A central bank acting as lender of last resort to contain contagion in a frozen market is doing something different from purchasing government bonds and issuing reserves. Kaplan agrees that intervention to stem contagion can be important. The label alone does not identify the policy’s purpose or its likely effects.

He draws a sharper distinction around direct central-bank purchases of newly issued government debt. In Kaplan’s account, this begins to look like monetary financing of the deficit: using newly created money to pay for government expenditure under the label of QE. He expects that policy to have significant inflationary consequences. It should be considered as part of the interaction between monetary and fiscal policy, he says, rather than treated as an ordinary instrument of conventional central banking.

The distinction matters because an asset purchase is not defined by its label alone. A response to a frozen market, a change in the maturity composition of government debt, and direct purchases of newly issued debt may have different purposes and consequences. Kaplan’s broader point is that policy debates become less clear when a common name is used for interventions that work through different channels.

Kaplan traces the move toward heterogeneous-agent models through changing theories of consumption. The sequence is not simply a progression toward ever more elaborate models. It is an exchange between theory and data, organized around questions that the existing account could not answer.

Keynes’s consumption function linked household spending to income. As Kaplan describes it, households spent a fraction of their income—perhaps 70 or 80 percent, or more—and saved the remainder. Early household budget studies appeared to fit one feature of this account: higher-income households spent a lower fraction of their income. But a different pattern emerged in aggregate time-series data. Incomes grew substantially, while the saving rate did not seem to rise as the cross-sectional relationship might suggest.

Friedman tried to reconcile the difference by distinguishing temporary income from permanent income. It was not only the amount of income that mattered, but whether it was expected to persist. Franco Modigliani and Richard Brumberg independently developed a related life-cycle hypothesis. Both approaches asked how a household would optimally spend and save in an imagined environment, rather than treating current income as a sufficient guide to current consumption. Kaplan notes that Friedman made much of his argument verbally, but the move was still theoretical: specify a world and reason through what households would do in it.

Robert Hall brought the permanent income hypothesis into a rational-expectations framework. If households make plans based on expected lifetime resources, they should respond to news when they learn it, rather than wait until the income arrives. Someone told today about a raise next year should increase consumption today, because the news changes the person’s expected lifetime income. This prediction gave researchers a way to test the theory against data: do households respond to information they already had, or only when income actually changes?

The evidence on marginal propensities to consume posed a further challenge. A household that can borrow and save freely should spread a small additional payment over the remainder of its life, consuming only the annuity value now. That implies a small immediate response. But the measured average response Kaplan describes—roughly 20 to 25 percent—was between that prediction and the much larger Keynesian spending share. And the average did not describe everyone. Some households smoothed consumption; others spent much more of an additional payment quickly.

Models incorporating precautionary saving helped bridge the gap. A household may wish to smooth consumption but hold a buffer because future income is uncertain. In Kaplan’s account, work by Angus Deaton, Christopher Carroll, and Stephen Zeldes brought precautionary saving into the permanent-income framework and helped establish the household-level foundations of modern heterogeneous-agent models. Those models retained the question of optimal spending over time while allowing household circumstances and responses to differ.

Kaplan encountered this approach as a graduate student at NYU. He had studied actuarial science in Australia, then completed a master’s degree in economics at the London School of Economics and worked at the Institute for Fiscal Studies in the UK. The institute’s use of microdata to think about macroeconomic policy helped draw him toward the field. At NYU, Gianluca Violante’s course on macroeconomics and distributions introduced him to heterogeneous-agent macro. Kaplan says it was the first time he could see himself inside a model—not as a representative household, but as one of many households in an imagined economy.

The financial crisis sharpened his interest in the distance between models and public policy arguments. Kaplan interned at the New York Fed in 2007, completed his PhD in 2009, and then spent a year at the Minneapolis Fed. By then, rates had reached zero, the Fed was using forward guidance, and a major round of fiscal stimulus had occurred. He saw a gap between what he had studied in graduate school and the arguments about policy he encountered in newspapers.

The crisis also brought criticisms of macroeconomics into view: that it had failed to foresee the crisis, used too much mathematics, neglected psychology, or ignored inequality. Kaplan says some criticisms were right, but he rejects the claim that economists had not studied distributions. In his account, a deeper source of disagreement was that macroeconomic debates often depend on different assumptions about how the economy works. People may argue from theories without recognizing that they are doing so. The models make those assumptions more explicit.

Hartley asks Kaplan to weigh inflation targeting against alternatives, including price-level and nominal-GDP-level targets, and raises the Federal Reserve’s flexible average inflation targeting framework. Kaplan does not say which target he prefers. Instead, he turns to how models should be built and used: the appropriate framework depends on the question being asked.

For Kaplan, the aim is not to make a model richer for its own sake, but to build a clearly articulated artificial economy consistent with the microdata relevant to a specific question. The task is to identify which mechanisms determine the answer, rather than add every possible feature.

For a question about household responses to interest rates, differences in liquidity and spending behavior may be central. For another question, the important features might include trade, political power, geopolitical influences, or technological change and its effects on labour-market distributions. A model that omits those mechanisms may be poorly suited to that question, even if it represents household heterogeneity in detail.

The same qualification applies to firms. The models under discussion often have a representative firm, which Kaplan describes as “pretty boring.” For questions about firms’ decisions, frictions, and interactions with monetary or fiscal policy, a more developed account of firms may matter more than a detailed two-asset structure on the household side. There is no requirement, in his view, to integrate every feature into every HANK model. Some features may have to be stripped out when others are added.

Kaplan says he would not use New Keynesian or HANK models to forecast next quarter’s GDP, sales, or even next year’s sales. Those models were developed for questions involving counterfactuals, often at the scale of the whole economy. Their purpose is not to outperform a forecasting model at extrapolating near-term outcomes, but to help reason about a policy or decision whose effects cannot be read directly from past data.

His example is a nationwide wealth tax on billionaires. Such a policy has not been implemented in precisely the form being considered, so researchers would have to draw on related evidence and extrapolate. It also applies to the whole economy. Kaplan describes macroeconomists as having one collective experience of the world to study, rather than many independent versions of the same economy. Data alone therefore cannot directly answer every economy-wide counterfactual.

A structural model supplies a disciplined thought experiment. Researchers identify the mechanisms they believe matter, build an artificial economy that represents those mechanisms, and examine what happens when the policy changes. The model supplements observed evidence; its usefulness depends on whether the environment includes mechanisms relevant to the question.

The same distinction applies outside macroeconomics. A company trying to forecast next period’s sales may not need a structural model. A company considering entry into a market with a product that has never been tested may need a model of the relevant forces. Kaplan says that would not necessarily be a HANK model. The broader point is that a counterfactual about an unfamiliar decision requires some account of the behavior and constraints that shape outcomes, not just a projection from what has already happened.

Kaplan also resists treating more data as automatically decisive. Researchers must decide which data matter and how to use them. If the question is how households respond to interest-rate changes, information about different households’ responses may help discipline the model. The data alone may not answer the counterfactual, but they can give researchers different moments against which to assess the model. Expectations data can serve a similar purpose if they reliably measure what people expect.

He draws a distinction between directly observed evidence and objects such as the natural rate of interest or the natural rate of unemployment. Kaplan says these are outputs of models, not quantities defined independently in the real world. Their definitions depend on the hypothetical economy under consideration. That does not make them useless: they can help explain how policy works within a particular model. But he would look for observable evidence that disciplines the model and its counterfactual result.

This question-specific approach also explains why Kaplan does not see HANK as a universal template. The models were built to study particular questions, and he says they have helped economists learn more about those questions. The next step is to develop appropriate settings for newer counterfactuals, not to make one framework carry every problem.