After the Average: Evidence and Judgment in Macroeconomic Policy

Abstract

This essay argues that macroeconomic policy becomes more useful when it treats evidence as a discipline of judgment rather than a substitute for judgment. Drawing on empirical work in public finance, labor economics, housing, education, monetary transmission, trade, and household finance, it asks how policy should move from aggregate outcomes to the people concealed inside them. The central claim is that data can clarify mechanisms, expose distributional consequences, and narrow the range of defensible arguments, while still leaving policymakers responsible for deciding which outcomes matter.

What Evidence Permits

I once expected macroeconomics to operate at the scale of countries, where growth, inflation, employment, and trade seemed to belong to a language of national totals (a tidy expectation, and, like most tidy expectations about the economy, one the evidence felt no obligation to honor); yet evidence steadily moved the subject closer, until a change in output had become a foreclosure in one county, a shift in the price level a transfer from a retired saver to a young mortgage borrower, a labor market recovery two separate events, one for a college graduate in a growing city and another for an older production worker in a rural town, and a fiscal transfer a different matter for a liquid household than for a constrained one. What I had taken for a conclusion came to look instead like a beginning, an aggregate whose meaning depended entirely on the people concealed inside it.

That shift gave data-driven policy a stricter purpose than I had once assigned it, for data force a preferred argument to disclose its mechanism and the limits of its reach, subjecting every serious causal claim to the same interrogation: what changed, compared with whom, through which channel, and for how long? The order of those questions matters, because a persuasive estimate can still support an unpersuasive policy when the mechanism beneath it has been misread, and several expensive failures began at exactly that point, with policymakers who observed a relationship, supplied an intuitive story, defended the story as a mechanism, and built an intervention around a margin that turned out to be secondary. Disciplined doubt, I came to think, is less a mood than a method.

Bruce Sacerdote’s study of Korean American adoptees gave that discipline an unusually clear opening, since adoption separates inherited traits from family environment through an assignment process that approximates random placement, and the design thereby converts a familiar quarrel about nature and nurture into an empirical question about outcomes across households. Its force comes from institutional detail, from the fact that randomness has to be defended through the actual procedures by which children entered families and that the relevant population has to be understood before any estimate can travel elsewhere. What made the paper concrete for me was precisely this: identification rested not on abstraction but on placement decisions, administrative rules, sample boundaries, and the gaps that no control could close (every credible design has a few of these; the trick is to name them out loud), so that the human setting stayed visible inside the design (Sacerdote 2007).

The toolkit grew outward from that example, for randomized experiments, regression discontinuities, difference-in-differences designs, and event studies each manufacture a comparison that ordinary observation withholds, and though their technical differences are real, their shared moral is plain: a credible counterfactual has to be built. Because the world rarely shows what would have happened to the same person under a different policy, the researcher hunts for another source of variation able to carry that weight, and pretrends, placebo tests, bandwidth choices, and institutional knowledge then decide whether the comparison earns trust. I began to read every figure with this anatomy in mind, knowing that a flat line before treatment could harden a result, that a suspicious break could dissolve an elegant one (elegance, in this literature, being the first thing a robustness check comes for), that a clean placebo could starve a rival explanation, and that a narrow design could secure a local truth while leaving the national question wide open (Cunningham 2021).

Nathaniel Hendren and Ben Sprung-Keyser’s unified welfare analysis makes an unusually broad attempt to compare unlike policies, dividing beneficiaries’ willingness to pay by the government’s net cost, later fiscal consequences included, so that the resulting ratio can let an intervention in childhood health stand beside an adult tax change inside a single accounting system. Its most memorable pattern is age: money directed at disadvantaged children often yields benefits that persist through schooling and work, and the earnings that follow can return part of the original outlay through higher taxes, whereas many adult programs deliver immediate support with far smaller long-run offsets. In giving a number to a proposition that public argument usually leaves sentimental, the framework makes a quiet but consequential point, which is that the timing of assistance can decide its social return (Hendren and Sprung-Keyser 2020).

I trusted the paper because it exposed the judgment lodged inside its own precision: long-run earnings must be projected, willingness to pay must sometimes be inferred, and published estimates arrive attached to different populations and designs, so that when the authors test alternative assumptions and confront selective publication, the final rankings come to read as structured evidence with its seams left visible. I still hesitate, even so, to treat an infinite marginal value of public funds as a literal quantity (a policy that pays for itself forever being the kind of thing one should want to be true badly enough to check twice), since infinity usually means only that estimated fiscal gains cover measured costs under one particular set of assumptions; the category illuminates, yet its air of certainty can outrun the evidence beneath it. The paper is strongest, in the end, as a discipline for thought, one that asks policymakers to hold future tax receipts and avoided public costs in the same field of vision as the initial appropriation.

This tension between a clean estimate and a messy reality returns across questions that otherwise seem unrelated: flat pretrends strengthened the case for localized housing effects, a falsification exercise made the China trade instrument more credible, two mortgage modification rules pried liquidity apart from principal wealth, and linked account records tied employer shocks to household spending, each paper winning confidence through a different design and each drawing a different boundary around its conclusion. The sheer size of a dataset began, accordingly, to impress me less than the source of variation inside it (size being the cheapest thing a dataset has to offer), for millions of observations can trace a pattern with exquisite accuracy and still leave the original uncertainty about cause entirely intact.

External validity drew a second boundary, since Norwegian evidence can reveal channels of monetary transmission even as American mortgage contracts form a different landscape, Swedish labor records can pick out vulnerable displaced workers even as mobility and social insurance differ across borders, a Japanese earthquake can expose production networks even as the next disaster may strike firms holding other inventories and other suppliers, and euro area inflation can reveal balance-sheet redistribution even as national energy systems shape the outcome. Qualifications of this kind locate a study’s authority rather than diminish it, and a paper grows more useful to me the moment I can state the conditions under which its conclusion should travel.

Administrative data also changed my sense of what the state can know, for tax records, card transactions, payroll deposits, and mortgage files let researchers watch policy at a scale no survey reaches; but that power carries a governance problem, since coverage follows the institutions that generate the records, and people outside formal systems can therefore turn least visible at exactly the moment policy most needs to find them. A card dataset measures available credit with rare precision while staying blind to the cash borrowed within a family (it knows your limit to the dollar and has never heard of your brother-in-law), and tax records trace earnings while missing the unpaid care that surrounds them, so that linked records can deepen knowledge and reproduce an administrative absence in one and the same gesture.

Data infrastructure came, in turn, to look like part of policy design rather than a backdrop to it: secure linkage protects privacy, clear access rules decide which questions can be answered at all, public audit shows whether a program reaches the population it names, and continuous measurement exposes effects that diverge from the mechanism used to justify enactment. Revision then becomes an institutional capacity rather than an afterthought, since a learning system keeps evidence active through implementation, and data-driven government accordingly collects its evidence during the action rather than only before it.

My understanding of uncertainty changed as well, for where I once read a wide confidence interval as a confession of weakness (youth preferring a narrow interval the way it prefers a confident friend, and for about the same reasons), I now read disciplined uncertainty as part of the finding itself, recognizing that a narrow estimate resting on a fragile comparison deserves less trust than a wide range resting on a transparent one. Policy obeys the same rule: a government that admits uncertain incidence can build adjustment into a program from the start, whereas a government that treats its favored mechanism as settled can lock households into an intervention whose costs surface only once the law is in force. Empirical humility, then, has practical content, favoring pilots, automatic review, simple eligibility, and administrative data able to reveal its own failures.

A research paper marks the boundaries of what can be inferred about social obligation: it can clarify who receives a benefit and who pays for it, estimate behavior under a specific rule, drag into view consequences that political language has hidden, and expose the uncertainty a slogan suppresses. These are substantial achievements, for they shrink the range of defensible claims and make certain forms of indifference harder to sustain; yet the deeper lesson is that evidence and judgment remain bound together, evidence disciplining judgment while judgment decides which outcomes enter the analysis at all. That bond shows most plainly in the evidence on opportunity, where a map of place becomes a map of inherited possibility.

Opportunity Has an Address

The allocation of talent paper makes one of the boldest claims in this literature, for Hsieh and his coauthors estimate that the falling barriers facing women and Black men account for a sizable share of the growth in market output per person between 1960 and 2010. Their model treats occupational choice as an allocation problem, in which the economy assigns talent to the wrong work whenever capable people are shut out of professions by discrimination or by blocked access to human capital, so that wider inclusion comes to raise fairness and productive efficiency together (an arrangement rare enough in policy that one is tempted to read the proof twice) (Hsieh et al. 2019).

The idea is powerful because it seats equality inside the growth account, making exclusion itself a major source of waste, for a society that steers people away from the occupations in which they would be most productive forfeits output across decades. By assigning economic weight to lives that standard growth narratives keep in the background, the estimate also rewrites the counterfactual, until the question becomes how much richer the country might have been under institutions that read talent more accurately.

The framework leans, however, on a demanding assumption about how innate ability is distributed across groups, and although the authors marshal the available test evidence for stability, the model still needs a common benchmark to identify its wedges. I stayed uneasy, because childhood conditions shape measured skill long before anyone chooses an occupation, and because school quality, family resources, exposure to role models, and neighborhood safety all feed the capacities that later pass for natural, so that the line between innate talent and developed talent blurs in practice (the word “innate” doing a great deal of quiet work in any sentence that contains it). The worry leaves the model’s contribution intact even as it suggests how much of the estimated wedge may already have formed before the labor market comes into view.

Research on inventors deepened the point, for children who grow up near innovation, or who meet inventors from backgrounds like their own, become more likely to patent in adulthood, and a family tie to inventive work supplies another dose of the same exposure. Because access is unequal, many potential innovators never reach the pipeline at all, and the lost inventor turns out to be a person whose ability never found a setting in which it could become legible (Bell et al. 2019). Set beside Hsieh and his coauthors, this evidence suggests that allocation begins in the imagination, and that occupational barriers work not only through formal exclusion but through the earlier absence of a believable path.

The Opportunity Atlas translated this argument into geography, estimating, by linking tax records to census data, the adult outcomes of children raised in individual census tracts, and its striking feature is the scale of the local difference it reveals, since adjacent neighborhoods can produce sharply different earnings or incarceration outcomes for children from comparable families. I kept returning to two places a few miles apart in Los Angeles, where adult incarceration rates for Black boys diverged dramatically, a contrast that made national mobility statistics feel almost evasive; for a country can run a single tax code while childhood opportunity shifts block by block (Chetty et al. 2018).

The atlas gains its force from an exposure design in which children who reach a higher-opportunity place at younger ages capture more of its apparent benefit, so that the gradient between years of exposure and later outcomes lends support to a causal reading of place, support that Moving to Opportunity reinforces with experimental evidence strongest for children who relocated young (Chetty et al. 2016). These findings convinced me that neighborhoods shape development over time even as they sort families by existing advantage, and that the open question is the mechanism, for although safety and schools are plausible, so too are peer environments and the employment patterns of nearby adults, and the tract estimate bundles these forces together while leaving most of their internal structure unresolved (it knows the neighborhood matters and is studiously vague about why).

Historical evidence from the Great Migration complicates any simple instruction to move, for the Black families who left the South sought higher wages and different institutions only to find that destination cities had grown new forms of segregation that fenced off the very opportunity on offer, so that migration redrew the geography of racial inequality without spending much of its force (Derenoncourt 2022). The history matters because mobility policy can sound frictionless when read off a map, whereas families carry relationships, obligations, housing constraints, and local knowledge that do not transfer, and a destination can answer its newcomers with higher prices or political exclusion. The causal benefit of a neighborhood is real; access to that benefit is an equilibrium problem.

That tension shaped how I read place-based policy, for Kline and Moretti show that local subsidies can raise welfare when they correct agglomeration failures or labor market frictions, a temporary push helping firms coordinate around shared infrastructure and a thicker labor market; yet the same subsidy can also lift land values or draw new residents, leaving the intended population with the smaller share of the gain, so that incidence turns finally on housing supply and mobility. I grew warier, in consequence, of proposals to conjure a technology cluster through public spending (every region, asked, would like to be the next Silicon Valley; the supply of Silicon Valleys has not kept pace with the demand), since the rationale needs a specific market failure and the design needs a credible account of who will in the end capture the return (Kline and Moretti 2014).

The Empowerment Zone evidence supports a conditional case for the same approach, since Busso and his coauthors find meaningful gains in employment and wages inside treated areas, with little sign that rising housing costs swallowed the whole benefit, and their design, by using rejected applicants as the comparison, gives the policy a sharper empirical footing than most place-branding programs ever earn (Busso et al. 2013). The study moved me toward conditions rather than conclusions, toward the recognition that place-based intervention can work when eligibility has institutional logic, when local labor demand is genuinely weak, when housing can respond, and when administrators can watch for capture, while a single label continues to hide policies with very different mechanisms.

Affordable housing showed the same duality, for Diamond and McQuade estimate that Low-Income Housing Tax Credit developments raised nearby property values in lower-income neighborhoods and lowered them in affluent ones, that crime fell around projects in the poorer settings they studied, and that the mix of nearby buyers shifted as the estimated welfare effects moved with neighborhood income. Context thus becomes central to interpretation, since a subsidized development can read as investment where the housing stock has been neglected and as an amenity loss, capitalized into lower prices, where residents prize exclusivity (neighborhood sentiment having a way of discovering deep principles exactly at the property line) (Diamond and McQuade 2019).

The event-study evidence convinced me, especially the calm before construction and the speed with which the effects localized; the welfare calculations, by contrast, unsettled me, for the paper infers rents from property values and rebuilds renter preferences from demographic information (the econometric equivalent of guessing what someone wants for dinner from their zip code), so that its large benefit estimates lean on assumptions reaching well past the narrow area in which the price effect is actually identified. The mechanism runs into data limits as well, since building quality and density may matter, tenant composition and project management may matter too, and a price response captures only the combined effect while leaving the separate channels in shadow. That gap matters because design depends on the source of the gain, and a well-managed mixed-income project and a generic increase in subsidized units carry very different lessons.

The folk economics of housing explains why even well-identified supply evidence struggles to enter politics, for many respondents worry about high prices while predicting that new construction will push those prices higher, and their answers drift across repeated questions in a way that suggests housing beliefs often lack a stable internal model (supply and demand being, evidently, optional once the topic is one’s own street) (Elmendorf et al. 2025). The instability struck me as both encouraging and unsettling, since weakly held beliefs may yield to clear information but may equally yield to whichever narrative arrives first, and a luxury tower in a booming district can seem to prove that building causes high rents simply because visible development and expensive housing show up together (the cause and the cure looking, from the sidewalk, identical), while the regional counterfactual stays invisible.

Housing turns causal reasoning into a public problem, for voters see the cranes where demand is strong but rarely see the rents that would have prevailed under a tighter supply constraint. Rent control offers a story just as immediate, in which sitting tenants feel the protection while the contraction in rental supply unfolds over a longer horizon and a wider area, and evidence from San Francisco shows real benefits for covered tenants alongside landlord responses that shrank the rental stock (Diamond et al. 2019). The policy thus delivers genuine insurance to one group while raising the barrier for future residents, and both effects belong in the welfare account.

Successful housing policy, I came to think, needs a narrative able to stage an absent counterfactual, and “Build more” gains political meaning only when the argument ties supply to the specific households that scarcity excludes and pairs construction with protection during the transition (a slogan being, on its own, no match for a luxury tower one can point at). Zoning reform, housing vouchers, public development, and tenant assistance can act together precisely because each addresses a different margin, and the political task is to make their relationship visible before scarcity hardens into a defense of the status quo.

Higher education forms another geography of opportunity, one the mobility report cards map by splitting access from outcomes, pairing students’ parental incomes with their later earnings. Many selective colleges produce strong outcomes for low-income students once those students are inside, so that their thin contribution to mobility comes mostly from low enrollment shares, whereas several broad-access public institutions combine scale with substantial upward movement, a combination that lends them a national importance prestige rankings tend to hide (Chetty et al. 2020a).

The result changed the question I ask about universities, for academic quality matters, but access decides who reaches the production process at all, and the evidence favors wider access for qualified low-income students at selective institutions precisely because the inequality is often largest at the door. The failure, in other words, sits at recruitment, preparation, admissions, or affordability, and targeted outreach can shift college choice among high-achieving students who would otherwise enroll at less resourced institutions, which shows that information and application support can move outcomes even where formal eligibility already exists (Dynarski et al. 2018).

The concentration of mobility in a handful of public colleges creates a vulnerability of its own, for state disinvestment, rising tuition, capacity limits, and geographic inequality can weaken the very institutions that carry the most students, and generous aid at a few private universities offers too little scale to replace lost access across an entire public system. I once held an elite-centered picture of higher education policy (the picture most of us are handed, in which a nation’s mobility runs through a dozen famous campuses), but the data pulled my attention toward institutions whose importance lies in volume, until it became clear that a college can transform the lives of those it admits and still add little to national mobility for as long as its door stays narrow.

College policy also showed why an enrollment count has to be tied to each student’s educational and financial trajectory, since admission creates opportunity when the institution delivers a credible return and inflicts financial harm when students borrow for programs with weak labor market value and low completion. The same dollar of debt means one thing to a graduate with rising earnings and quite another to a dropout holding a credential worth little, which is why default clusters among borrowers with modest balances, and especially among those who attended poorly performing institutions (Looney and Yannelis 2015).

The pattern changed my view of headline debt statistics, for a large average balance can sit behind a professional degree with strong expected income while a small balance can mark an unfinished program and a damaged credit record (the borrower in the magazine photograph, with six figures and a medical degree, being almost never the borrower in trouble), so that policy aimed at distress should track repayment capacity and institutional quality, and policy aimed at the price of higher education should address financing and public capacity, two objectives that a single cancellation rule serves unevenly.

Evidence on marginal admission supports the value of access under the right conditions, for Zimmerman, studying students near an admissions cutoff at a public university, finds substantial earnings gains from getting in, with the strongest pathways running into professional work (Zimmerman 2014). The result backs admission for students at the margin even as it shows that the return can hinge on major choice and employer networks, a reminder that a college is at once a site of instruction and a gateway into occupations.

Predictive analytics opens another tension, for universities can mine prior records to flag students who may need support, yet the same model can steer recruitment away from applicants judged costly or risky, and because a prediction encodes historical patterns shaped by unequal preparation, acting on it can reproduce those patterns inside a system that presents itself as neutral. Data improve policy when they trigger assistance and grow troubling when they turn inherited disadvantage into an automated reason for exclusion.

I came away with a two-part standard for higher education policy: institutions should widen access wherever causal evidence shows strong returns, and they should answer for outcomes once public financing follows the student, with transparency about completion and earnings sustaining the public oversight, and with aid eligibility tracking educational value while protecting fields whose social contribution salary measures poorly.

Taken together, these cases make place a form of accumulated policy, in which a neighborhood holds the consequences of zoning and school boundaries, a college records funding choices and admissions rules, a labor market carries the history of the firms that arrived or left, and a household carries the balance-sheet residue of those same local institutions. Opportunity has an address, then, though the address explains only so much, for families sort into places, institutions react to migration, prices capitalize public action, and governments rewrite rules; and if the empirical task is to separate these forces, the political task is to improve a place while keeping its gains within reach of the people who gave the intervention its purpose.

Research on racial disparities in housing returns further complicates homeownership as a universal road to wealth, for the return depends not only on purchase price and neighborhood appreciation but on sale timing and the odds of distress, and Kermani and Wong find large racial gaps in realized housing returns, with foreclosure and short sales doing much of the work (Kermani and Wong 2026). A household can own an appreciating asset and still miss the market return when liquidity pressure forces a sale at the worst possible moment, so that housing wealth comes to link neighborhood history to the very balance-sheet fragility that runs through household finance.

Data on race and economic opportunity show disparities that persist across generations even among children raised at similar parental incomes, with downward mobility especially severe for Black men (Chetty et al. 2020b), and the racial pattern reaches past class, for although place and family income explain part of it, exposure to discrimination and the criminal justice system can redirect outcomes from the same broad starting point. The atlas makes these differences measurable while leaving policy the harder task of naming the institutions that produce them.

The reparations model adds a long-run warning, for a one-time transfer can close a measured wealth gap on the day it is paid and yet leave untouched the processes that govern investment and returns, so that as historical exclusion continues to shape beliefs about risk and access to profitable opportunities, the gap can reopen (Boerma and Karabarbounis 2021). The argument is useful because it joins redistribution to structure without subordinating either: a transfer can repair an inherited deficit of resources, but lasting convergence may also require access to investment and protection against the shocks that destroy wealth again and again.

Together, these findings reshaped what equal treatment means to me, for a formally common rule lands in households with very different histories of appreciation and liquidity: mortgage subsidies can reward owners in already rising neighborhoods, retirement incentives can reward people with spare cash, a race-neutral foreclosure process can produce unequal realized returns when one group reaches distress more often, and college aid can reward enrollment while leaving unequal access in place (a surprising amount of social policy rewarding, with quiet efficiency, the people who needed it least). Evaluation should therefore trace both the path into the rule and the path back out of it, recognizing that equality at the point of administration can sit beside deep inequality in everything that surrounds it.

Inequality as Structure and Policy

Pareto analysis at the upper tail exposes processes hidden inside national averages, and Chad Jones’s account gave me a new grammar for concentration, since when the tail follows a power law, a small change in its parameter can swing the share held by the richest households dramatically, and a figure that first looks like a mere descriptive ratio becomes instead evidence about the process generating extreme outcomes. The framework ties concentration to growth at the top and to the rate at which fortunes are reset: compounding thickens the tail, and turnover thins it (Jones 2015).

The mechanism satisfied me because it puts persistence at the center of inequality, for although a fortune can grow fast when its owner saves at a high rate or earns exceptional returns, how long it survives depends on taxes, inheritance, business failure, and institutional turnover, so that the same top share can descend from very different histories, one economy rewarding repeated innovation where another merely preserves rents across generations. A Pareto parameter, in pinning down concentration, leaves its moral and productive character open, and while the model sharpens the object that needs explaining, the explanation itself still takes empirical work.

Jones makes the welfare stakes vivid by setting growth for the extreme upper tail against growth for everyone else, an exercise that shows how national prosperity can conceal separate economic histories and that works precisely because it turns an abstract top share into divergent living standards. It also made me cautious, for income recorded at the top can be labor earnings, business income, capital gains, or movement across tax bases, each component answering policy differently, and a thickening tail may register greater productive reward, stronger market power, more effective avoidance, or some blend of all three, so that the aggregate movement always needs institutional interpretation.

Diamond and Saez attack that problem through sufficient statistics, deriving a formula for the optimal top marginal rate that turns mainly on the shape of the income distribution and the elasticity of reported income, and the appeal is immediate, since a forbidding optimal tax problem reduces to quantities researchers can actually estimate (the closest thing public finance has to a magic trick). Their methodological point matters just as much, for several famous results in public finance rest on assumptions too restrictive for real policy, above all the claims that the top rate should fall to zero or that capital income should go untaxed over the long run (results with a long half-life in op-eds and a notably shorter one in the seminar room) (Diamond and Saez 2011).

The composition of the reported-income elasticity became the policy object I cared about most, even as its size continued to matter, for a taxpayer can work less, shift income across years, change legal form, or hide income from the authorities, and these responses carry very different welfare implications: a real cut in productive activity is a social cost, whereas timing and relabeling merely expose weaknesses in the tax base. Because enforcement can move that second margin, the elasticity becomes partly a policy outcome rather than a constant, and I left the paper more skeptical of any argument that treats an observed response as a fixed law of behavior.

The paper’s numerical recommendations still demand caution, for long-run career choice can respond to taxation, inventors and firms can move across borders, pre-tax compensation can shift through bargaining or corporate governance, and corporate location can move as well; indeed, research on inventor mobility confirms that highly skilled workers respond to international tax differences, with the largest effects among the most mobile (Akcigit et al. 2016). These margins argue for a progressive system built for an international setting, one in which residence rules, profit allocation, information exchange, and base definition belong next to the statutory rate itself.

The makeup of top income blurs the line between labor and capital, for Smith and his coauthors show that much of it flows through closely held businesses whose owners are actively at work, so that their earnings fuse entrepreneurial labor with returns to ownership, and the tax system keeps asking the researcher to separate components the firm itself prices as one (a request the firm regards as none of the tax system’s business) (Smith et al. 2019). This matters because a rate aimed at passive wealth can strike active business income, while a labor tax can invite relabeling through corporate form.

The evidence left me with a more tangled picture of top earners, since closely held owners often combine labor with control, the two sources feeding each other, and ownership can throw off a return through market power or organizational capital even as it rewards the owner’s ongoing work, so that tax policy has to operate across that ambiguity rather than around it.

The measurement problem runs on into wealth estimates, for capitalizing reported income assumes a fixed relationship between an asset and its yield, whereas households earn different returns on similar measured wealth and private businesses are especially hard to value, so that a low observed yield may flag either an overvalued asset or a strategy that defers taxable income, and a high yield may reflect either risk or rare managerial skill (wealth being the one quantity its owner has both the means and the motive to misstate). Measured wealth concentration therefore depends in part on the very rate assumptions used to convert flows into stocks.

The ambiguity sharpened my interest in information reporting, since a tax base leans far less on heroic inference when administrators hold ownership records and valuations, and it sharpened, too, my caution about point estimates, for a fight over the exact top share can consume attention while the broad fact of extreme concentration stays perfectly secure. Measurement uncertainty should shape rate design and enforcement; it should not be allowed to obscure the distribution itself.

The wealth tax proposal pushed administration further toward the center, for Saez and Zucman argue that extreme concentration creates a large potential base and that modern information reporting can fix the failures that sank the European wealth taxes, their case resting on broad coverage, third-party reporting, credible valuation, and serious enforcement. The measurement work persuaded me, since wealth estimates built from capitalized income and from surveys diverge precisely because each sees particular assets through particular errors, so that reconciliation becomes part of the policy argument rather than a footnote to it (Saez and Zucman 2019).

My confidence fell when the paper crossed from feasible design to political durability, for valuation disputes will concentrate among the households with the sharpest incentive to contest them, exemptions can arrive through lobbying, closely held firms can reshuffle ownership until market value grows hard to observe, and enforcement budgets can shrink once the tax is law (a tax being written once and defunded annually); and these are economic responses, since together they set the effective base. The deepest uncertainty, then, is state capacity over time, for a technically coherent statute can erode into a porous one across a single decade of administrative neglect.

Wojciech Kopczuk’s response sharpened the worry by asking what problem a wealth tax is actually meant to solve, given that taxing a stock of wealth hits normal returns and economic rents with the same instrument, whereas reforms to capital income or estate taxation may reach some of those margins more directly. The criticism struck me as important even as I kept my sympathy for the proposal, since existing systems leak badly too (the choice rarely being between a clean instrument and a leaky one, but between two leaks of different shapes), but the response shifted the burden of proof, so that a recurring wealth tax now needs an explicit edge over repaired alternatives, especially once valuation cost and political fragility enter the ledger (Kopczuk 2019).

Retirement saving incentives offer a quieter version of the same design problem, for tax preferences and employer matches reward contributions, and households with the liquidity to save collect the larger subsidies, while households facing immediate expenses contribute less and often pull their funds out early. Matching formulas only widen the original gap, since the public or employer benefit climbs with the worker’s own deposit, so that the system aims its strongest reward at the very people already able to defer consumption (a subsidy for saving being, in the end, a subsidy for having money left over) (Choukhmane et al. 2024).

The paper changed how I think about incentive-based social policy, for a subsidy tied to desirable behavior hands out benefits in proportion to each household’s capacity to perform that behavior, and retirement security rests not only on income and plan access but on liquidity risk, family obligations, health shocks, and inherited support. Since two households with identical annual earnings can differ sharply in their ability to lock money away, a policy keyed to voluntary saving can reproduce precisely the constraints that income-based progressivity would catch.

Minimum wage research carries the design problem from saving incentives over to labor income, for Dube and Lindner, surveying local increases, find little evidence of large employment losses at the levels studied, and the border comparison around San Jose stays with me, since restaurants a short distance apart faced different wage floors and kept their doors open all the same (the burrito, it emerges, does not cross a county line to chase a cheaper cook). Local product markets, on this evidence, look more tightly bounded than the simplest relocation story assumes, and firms can adjust through prices, turnover, productivity, or thinner profits, so that employment turns out to be one margin among several (Dube and Lindner 2021).

On balance the evidence supports higher local wage floors, though the welfare question runs through payroll records and local living costs alike, for a nominal raise can be eaten by rent in a supply-constrained city, commuting costs can reshape access to the covered jobs, workers outside employment gain no direct raise at all, and firms may rework scheduling or benefits in ways hourly wage data never see. Research on firm adjustment stresses exactly these nonemployment margins, which deserve the same scrutiny as head counts (Clemens 2021), and the strongest case for a minimum wage, in the end, joins labor market power to local cost conditions.

The fight over Seattle taught me to resist a familiar temptation, for a weak control group can warp an estimate, and even a refined synthetic comparison can rest on unobserved differences (the synthetic control being a flattering composite of cities that are not, in the end, Seattle), so that one city remains a case with limited national reach. The accumulated literature gains its force because several designs converge on small average employment effects within the range studied, yet that conclusion still leaves room for variation by place and cycle, and it leaves the optimal level open, since evidence gathered around one wage says little about every higher wage.

I came to see redistribution as part of production itself rather than a stage that follows it, for taxes, benefits, retirement rules, wage floors, and enforcement move bargaining power and opportunity long before measured income ever appears, and pre-tax and post-tax accounting therefore belong inside a single causal account of how public institutions generate the distribution they later adjust. Policy enters the market at its foundations.

Technology and the Organization of Work

Technology poses a question that keeps defeating prediction: when a machine performs part of a job, what happens to the people who used to perform that part? The automatic teller machine is David Autor’s most useful example, for cheaper machines lowered the cost of running a branch, which led banks to open more branches, and tellers kept their jobs as their work tilted toward customer relationships and sales (the cash drawer automated, the human promoted to selling things). The machine, in other words, displaced a task and enlarged the organization around it, so that a forecast fixed on the cash window would have missed entirely the spread of the branch network (Autor 2015).

The example taught me that a reliable forecast has to track the responses that gather around a visible substitution, for technology moves prices, prices move demand, organizations redesign work, and workers retool their skills, and those responses can invent new tasks or widen the market for existing services even as they degrade job quality where employment survives. A teller who joined a bank for stable clerical work may well experience a sales role as a loss, a judgment that employment counts cannot settle, since the post-automation job can demand more social skill while offering less autonomy and more pressure. The task framework explains survival; worker experience decides whether survival counts as progress.

Autor’s account of polarization named something I had seen without quite naming it, for employment growth has drained away from routine work toward both ends of the occupational distribution, a pattern that computerization fits, since codifiable tasks are the easiest to automate. The wage evidence is murkier, however: fast growth in low-skill service jobs ran alongside falling pay, as a swelling supply of displaced workers could push wages down, and although the paper makes that mechanism plausible, I wanted more direct evidence linking the occupational shift to the fall in compensation, for a framework grows stronger when it identifies the institutional process that turns task change into wages.

Acemoglu and Restrepo extend the task approach by casting automation and task creation as opposing forces, the one shifting activities from labor to capital, the other rebuilding labor demand wherever humans hold a comparative advantage in emerging work; and the balance between them is historically contingent, since institutions shape which technologies attract investment, and firms may chase labor displacement even when a different innovation path could raise productivity through complementarity (the cheapest way to do without a worker and the best way to use one being different projects, and management not always choosing the second) (Acemoglu and Restrepo 2019). The point made technological change feel less autonomous, a process steered after all by research incentives and managerial choices.

Power decides whether productivity gains turn into wages or into profits, with bargaining and strong labor demand pushing toward the first and market concentration toward the second, so that a technology which complements workers in production can still weaken them inside the firm once monitoring tightens or outside options narrow. The rise of superstar firms and the falling labor share only sharpen the concern, for scale can reward productive efficiency while funneling the surplus to owners and a thin layer of employees, and the distributional result depends in the end on antitrust policy, labor institutions, tax rules, and ownership, the same technical improvement yielding different social outcomes according to those settings.

Education works on a long institutional horizon, and Autor points to the high school movement of the early twentieth century, when broad public investment readied workers for a changing economy, an analogy at once hopeful and sobering, since that transformation took decades, and today’s institutions have struggled to redesign community colleges and vocational pathways at the pace digital change demands. A worker displaced at fifty bears an immediate loss while a new educational system slowly matures (“retrain” being a brisk word for a process that, up close, takes years a fifty-year-old may not have to spare), which is why long-run skill formation and short-run adjustment call for separate policies, and why treating education as a universal cure can become a way of deferring responsibility for people whose losses have already landed.

Generative artificial intelligence made these questions immediate, for Brynjolfsson and his coauthors, studying a conversational assistant rolled out among customer support workers, find that productivity rose most for less experienced employees, a result consistent with a system that encodes the practices of stronger performers. The large novice gains first looked almost built into the design, since a tool trained on successful interactions should help most the people who have yet to learn those patterns; but the revealing part was persistence, for workers who followed the useful recommendations seemed to improve beyond the immediate assistance, which suggests that the system can double as on-the-job instruction (Brynjolfsson et al. 2023).

The learning channel changes the welfare calculation, for a tool that merely speeds a conversation during one shift produces a flow benefit, whereas a tool that helps a worker build transferable skill changes future productivity as well, a distinction that matters for pay and for career design, since firms may capture the immediate output gain while workers keep part of the human capital. The balance turns on whether the learned practices hold their value outside the original platform: customer support may share enough structure for that transfer, while more specialized occupations may breed knowledge bound tightly to one firm’s data and workflow.

The fate of top workers deserves equal attention, for an assistant assembled from common successful responses can distract people whose judgment already runs ahead of the template, and generic recommendations can lower quality whenever an unusual case demands a departure. The feedback problem cuts deeper still, since pressing expert workers to conform may erode the very source of the organization’s future training data, their eventual exit thinning the range of examples the model learns from, so that a productivity tool can end by consuming the expertise it had set out to reproduce (a model trained on its experts can, by nudging them all toward the average, slowly eat its own dinner). The original experiment captures only an early stage of adoption and leaves this slower organizational cycle well beyond its horizon.

The macroeconomic extrapolation is shakier still, for although customer service supplies abundant text records and repeated interactions, with an objective that can be read off resolution time and satisfaction, many jobs run instead on sparse data or moving goals: physical trades depend on local environments, nursing involves bodily care and ethical judgment, management often turns on conflict that wears no stable label, and legal strategy can confront a situation with little precedent. No chatbot has yet unclogged a drain, held a dying patient’s hand, or talked two vice presidents off a ledge. Such settings may gain useful assistance, yet the customer support estimate is a narrow foundation on which to build a general productivity parameter.

Daron Acemoglu’s macroeconomic analysis puts this scaling problem at the center, for aggregate gains depend on the share of tasks artificial intelligence can perform and on how far it cuts costs within those tasks, so that striking demonstrations can sit beside modest national effects whenever the applicable activities cover only a slice of production (a dazzling demo and a rounding error in GDP being entirely compatible) (Acemoglu 2024). General equilibrium adds yet another layer, since firms may hire less experienced workers once the tool raises their productivity, wages may move, product demand may expand, and new services may appear, the micro estimate remaining real even as its incidence shifts with the economy’s reaction.

Early labor market evidence counsels caution about speed, for survey and administrative work in Denmark finds rapid adoption of generative tools with little measurable effect on earnings or hours during the initial period (everyone has the tool open in a browser tab; the paycheck has not yet noticed) (Humlum and Vestergaard 2025), a result that may describe a technology whose organizational complements have not yet arrived. Electrification, after all, delivered its largest gains only once factories had rebuilt their layouts and workflows, and artificial intelligence may need a comparable redesign, so that the delay itself carries policy weight, buying time for training and institutional adjustment even as it clouds the case for investment.

The earthquake study uncovered a different form of technological organization, one in which production is laced together by supplier relationships that carry a local shock far past the prefectures actually damaged, and Carvalho and his coauthors find meaningful propagation in both directions along the network, a firm losing output when an input supplier fails and a supplier losing revenue when a customer stops producing. The effects fade with network distance, a fact that lends the pattern a coherence which a simple regional correlation would lack (Carvalho et al. 2021).

What surprised me was the size of the national effect set against the small share of output directly hit, for a geographically narrow disaster became a macroeconomic event precisely because production hung on specialized links, and the finding changes the meaning of efficiency, since a tightly optimized network both lowers ordinary costs and magnifies rare disruption. Inventories look wasteful in calm periods and valuable after a shock, a second supplier looks redundant until the first region fails, and geographic concentration looks efficient until transportation fails (redundancy being waste right up until the morning it is the only thing still working); resilience, in short, is a productive asset whose return arrives only at irregular intervals.

The data also set a boundary, for the study sees only whether firms trade, while transaction values have to be imputed from industry information, so that a minor supplier and a critical one can enter the network looking alike when real exposure depends on input shares and substitutability. A firm may swap in a generic part within days, whereas a specialized component can halt an entire line, and better transaction data would let the model tell these cases apart and point policy toward the genuinely systemic links.

Taken together, this evidence changed my view of automation, for the unit that matters is the machine set inside a workplace and a market, where tasks shift, organizations adapt, workers learn, and bargaining splits the gains, and production networks carry the same insight across firms. Technology policy should therefore weigh competition, training, innovation, and ownership, and should count resilience as part of productivity rather than a cost set against it, since a narrow gaze on immediate output misses the institutions that decide whether technical progress becomes shared capacity or concentrated power.

Fragile Households and Unequal Shocks

Household finance brings macroeconomic policy into the bank account, for consumption follows income, but the response to an income change follows the balance sheet that receives it, so that two families can lose the same share of earnings and choose very differently because one holds liquid savings while the other has run out of credit. The distinction sounds obvious once stated, yet where I had often treated wealth as a single measure of security, I came to see that its form can matter as much as its amount.

Scott Baker’s linked account study made the point with unusual clarity, joining transactions, deposits, credit limits, and employer information at the household level, so that gross debt looks strongly tied to spending sensitivity until liquidity enters the picture and the independent role of debt flattens out once cash holdings and unused credit are added. Debt matters, that is, through the limits it places on access to funds, for a household with a large mortgage and ample liquidity can ride out a modest income drop, while a household with little debt can still hit an immediate crisis when cash is scarce and credit has closed (Baker 2018).

Figure 11 stays with me because it peels away a common intuition layer by layer (I confess a sentimental attachment to a small number of figures, and this is one of them), and the result carries a practical implication, since policies aimed at cutting leverage may do little for short-run consumption when monthly obligations hold steady, whereas policies that ease a current payment or hand over spendable funds act on the very margin households actually face. Debt still matters, even so, because heavy obligations drain liquidity over time and damage credit after a missed payment, the causal chain running through those intermediate states.

I was less sure about the scale of the income shocks used for identification, for Baker maps direct deposit descriptions to employers and uses firm events as the variation, and although the validation tests are clever, including checks for anticipatory responses before a job begins, the resulting income movements are often small, and a response to a minor earnings wobble may differ sharply from the response to dismissal in a recession, since fixed costs grow harder to cut when income falls steeply and credit limits can tighten at the same moment. The paper identifies a local response around the shocks it observes; reaching to economic collapse takes another step.

Work using JPMorgan Chase records reinforces the weight of liquid buffers and shows how unevenly they are spread, for ordinary income fluctuations move consumption more among households with little liquid wealth, and racial gaps in cash buffers help explain gaps in smoothing even between families with similar income (Ganong et al. 2020). The result ties household finance back to the earlier mobility material, since a history of unequal wealth accumulation shapes the response to a routine payroll change, and macroeconomic volatility reaches households through inherited financial capacity.

The study of exogenous wealth and unearned income looks at a different side of the same balance sheet, for Golosov and his coauthors exploit changes that arrive with no matching change in labor productivity, and find that households cut labor earnings substantially after receiving extra unearned resources, the average response large enough to matter for any permanent transfer proposal. Its spread is the more revealing part, since higher-income households tend to adjust through hours or effort, whereas lower-income households are likelier to move across the participation margin altogether (Golosov et al. 2024).

The pattern changed how I understand work incentives, for labor supply is an abstract quantity in a model, yet households meet its margins in very different ways: a professional with control over a schedule can dial back intensity and keep a career intact, while a lower-paid worker may hold a job built around fixed shifts, with little room for a partial cut, so that the real choice narrows to employment or exit (labor supply in the textbook is a smooth dial; on a warehouse floor it is a switch). A universal transfer therefore produces different behavioral forms even when every recipient gets the same check, and its fiscal effect depends on how those forms are distributed.

The business ownership findings added a more personal dimension, for extra wealth raised entry into lower-income business activity, which the authors read as evidence that some ownership carries consumption value, a household spending its financial freedom on autonomy or meaningful work even when profits stay modest. I wanted more detail about these firms, however, since a small online shop and a local landscaping company can share an income threshold and yet stand for very different investments and ambitions, and the administrative category, recording only entry, leaves most of the enterprise’s meaning outside the data.

Universal basic income sits squarely inside this ambiguity, for a stable transfer can supply insurance and strengthen bargaining power even as it lowers taxable labor income and shifts participation, and the study leaves the welfare value of those changes open: time away from an inflexible job may go to caregiving or education, while reduced employment may slow skill accumulation. A complete evaluation thus needs fiscal cost, household welfare, effects on skill, and the public value placed on nonmarket time, so that the evidence makes simple budget arithmetic untenable and leaves the social objective to public judgment.

The fiscal stimulus literature gives this liquidity logic a macroeconomic form, for Kaplan and Violante model households that hold substantial illiquid wealth and almost no cash, households that can look wealthy on a balance sheet and still spend a large share of a temporary payment (asset-rich and cash-poor: the family that owns a house, a car, and roughly nine dollars), so that the category of wealthy hand-to-mouth consumers explains why stimulus responses stay strong well beyond the conventionally poor (Kaplan and Violante 2014). The marginal propensity to consume, on this account, tracks liquid position closely.

The framework reconciled two facts that once seemed at odds, namely that many households own homes or retirement accounts and that many of the same households say they could not cover an unexpected expense (net worth being cold comfort at the emergency-room window, where they do not accept home equity); for their assets convert slowly and carry penalties or transaction costs, whereas a tax rebate lands in the checking account at once, so that the payment’s effect depends on that location, and a nationally financed transfer can prop up demand even in an economy with substantial measured household wealth.

Regional multiplier evidence raises another identification question, for Nakamura and Steinsson use variation in military procurement across American regions inside a single monetary system (the Pentagon, whatever else it is, being a generous natural experiment), finding that areas hit with larger spending shocks see stronger relative output growth, which yields a sizable local multiplier (Nakamura and Steinsson 2014). The comparison is powerful because national monetary policy and the exchange rate are held in common, yet carrying it to the aggregate economy still takes a model, since spending can pull resources from another region, and a national expansion can move interest rates through channels that a relative regional shock never touches.

The distinction matters because fiscal analysis often borrows a multiplier and leaves its setting behind, for a local estimate answers what happens when one region gets spending while the national stance holds roughly fixed, whereas a national estimate asks what happens when aggregate demand changes and monetary policy responds; both questions matter, but they are not the same question, and a data-driven approach requires the policy scale to match the empirical scale.

Temporary transfers and permanent basic income proposals run on different expectations, for a constrained household can spend a one-time payment fast without showing the long-run labor supply response that a permanent guarantee would draw out, whereas a recurring transfer resets the value of participation across all the years ahead, so that evidence from stimulus checks offers only thin guidance about a permanent income guarantee. How long a transfer lasts is part of what it does.

The strongest fiscal design would lean on administrative systems able to push out cash the moment aggregate conditions worsen, for automatic triggers cut legislative delay and make support predictable, and payment size can lean toward household liquidity while staying simple enough to administer; and although such targeting will always be imperfect, the alternative often delivers help only after consumption has already collapsed (relief that arrives once the lights are off being, technically, relief). Speed is a policy choice in its own right.

Inflation rewrites balance sheets without mailing a transfer notice, and Pallotti and his coauthors show how the recent euro area surge redistributed resources across age groups and national settings, with older households often holding nominal deposits whose real value slipped away while younger borrowers carried mortgage debt that inflation partly dissolved. Wage adjustment and energy exposure then opened large gaps between countries, and government relief reshaped the final incidence again, so that Germany and Italy saw severe losses among retirees while many younger French households came through better (Pallotti et al. 2023).

The age pattern was the paper’s most striking result, because it shows inflation as redistribution across nominal positions, for everyone meets higher prices, but nominal positions decide who gains as fixed claims erode, the balance-sheet logic that Doepke and Schneider had laid out in earlier work on nominal wealth redistribution (Doepke and Schneider 2006). The recent episode gave that logic immediate empirical form, since a household holding cash and a household owing a fixed mortgage can buy the same basket and feel opposite wealth effects (inflation quietly paying down the borrower’s debt and the saver’s savings at the same time).

Consumption composition complicates incidence further, for lower-income households pour a large share of spending into necessities whose prices rose the most, while housing costs stayed relatively sticky for many renters in the early period, which briefly softened their measured inflation until leases reset and price patterns shifted and the buffer wore thin. The timing matters, then, since a monthly index can understate future hardship when a large fixed expense has yet to adjust, just as it can overstate current loss when government relief is shielding a particular bill.

The cross-country comparison underscored how much institutions matter, for energy pricing and wage bargaining bent the path from a common external shock to household welfare, so that national averages carried policy architecture as much as exposure, and the result turned my attention to the institutions beneath the country ranking. Inflation arrives, after all, through markets that governments have already shaped, and contract structure, benefit indexation, tax relief, and regulated prices decide which households absorb the shock first; where an inflation shock lands is part of the shock itself.

Research on subjective inflation expectations helps explain why measured welfare losses can look modest beside the heat of consumer anger, for inflation breaks expectations again and again, confronting people with a price increase at the grocery store and the gas station, while wage gains come less often and may be credited to personal effort. Households also disagree about the inflation rate because they buy different baskets and remember the prices that stung (everyone being an expert on the price of the one thing they bought yesterday), and the surveys show persistent dispersion across consumers and firms (D’Acunto et al. 2022), so that the experience stays personal even when the index is national.

Job loss is among the sharpest income shocks a household can face, and once again the average hides the variation that matters, for when Athey and her coauthors break earnings losses down across worker characteristics and labor market conditions, education predicts vulnerability, yet the spread within education groups nearly matches the spread across them, and establishment quality and local demand can overpower any demographic guess, so that a younger graduate let go from a struggling market may suffer more than an older worker let go during a strong expansion (Athey et al. 2026).

The finding unsettled my confidence in demographic targeting, for age and education, though administratively simple, can miss the interaction between a worker’s task and the market that receives the worker, even as older employees in routine occupations surface as a useful high-risk group, especially where alternatives are thin. The rule has elegance because it captures a mechanism, yet its limit shows up after identification, since knowing who will lose the most says nothing yet about an intervention that could rebuild the lost match.

Where displacement happens matters, because workers move less than many models assume, and a rural manufacturing market may hold a single dominant employer, so that when that employer contracts, a worker loses both a job and the local demand for related experience at one stroke. Housing equity and family ties make migration costly, and the job search then unfolds inside a market whose opportunities have just thinned, the damage arising from severed firm-specific value and a poor outside option together.

Davis and von Wachter show that displacement during a recession inflicts especially severe long-run earnings losses, of which the initial unemployment spell is only one part, since reemployment can come at a lower wage and the gap can persist for years (Davis and von Wachter 2011), while graduating into a recession leaves a related scar through early job matches, with effects that fade slowly across a career (Oreopoulos et al. 2012). These studies changed how I see stabilization policy, for heading off a deep downturn protects future earnings paths that current output statistics never record.

Ganong and Noel’s study of unemployment insurance makes the immediate household response visible, for spending holds fairly steady while benefits arrive and drops sharply once eligibility runs out, and although recipients know the expiration date, consumption barely adjusts ahead of it. The cliff shows up in basic categories, food and medical spending among them, while insurance payments and debt service stay steadier, which suggests that households guard the obligations whose failure would be catastrophic (Ganong and Noel 2019).

My first reaction was disbelief (the reaction, I have since noticed, of a person who has never had to choose between the electric bill and dinner), since a forward-looking household should start saving before the last check, yet the data show how little room many recipients have for that, for benefit income may already stretch across a tight budget, and a double payment gets spent because delayed bills have piled up. Present bias may play a part, though liquidity explains most of the pattern more simply: money owed for this month’s rent and groceries leaves nothing for a cushion.

The spending categories also point to costs the account data never see, for less medical care can damage health later and cheaper food can carry nutritional harm, consequences that may hit public budgets after the observed window and may never show up as a consumption loss in the original dataset. The value of insurance therefore lies in visible expenditure smoothing and in protection against the choices that keep a household formally solvent by spending down its health.

Chetty’s distinction between liquidity and moral hazard supplies the theoretical frame, for unemployment benefits can stretch a job search both because recipients hold more cash and because the subsidy changes the payoff to accepting work, and where the first channel can raise match quality for constrained workers, the second is a conventional distortion (Chetty 2008). Design should therefore weigh their relative size, since a blanket fear of slackened search argues for benefits that are too small, while a flat disregard for incentives makes support less efficient.

The abrupt expiration in Ganong and Noel is especially hard to defend, for a gradual taper could blunt the consumption cliff, emergency reserves could meet health or housing shocks after formal eligibility ends, benefits indexed to the unemployment rate could lengthen support when jobs are scarce and pull back when openings return, and portable health coverage could close a separate cliff, designs that all answer to the state of the market and the household. Their administrative complexity is real, but the cliff that exists today has a large cost of its own.

Monetary policy reaches the same households through several balance-sheet channels, and McKay and Wolf combine changes in employment income, mortgage payments, asset values, and nominal positions, so that while each channel on its own looks sharply unequal, their aggregate estimate comes much closer to distributional neutrality, with consumption rising across much of the distribution after an easing (McKay and Wolf 2023). The conclusion surprised me, since public discussion tends to treat low rates as a straight gift to asset owners (a satisfying story, widely held, and only part of the picture), whereas employment gains among lower-income households can offset that channel.

I stayed hesitant, even so, about the apparent neutrality, for aggregation can net out large gains and losses inside broad groups, and a renter who finds work and a renter who stays unemployed live under different policies in practice, just as a homeowner who can refinance pockets a direct cash-flow benefit while a similar borrower blocked by credit standards pockets none. An average by percentile can therefore bury access margins that carry real political weight, so that near-neutral distribution in one summary can sit beside a sharply unequal lived experience.

The mortgage channel is central to the paper and to modern monetary transmission, for lower rates bite only when borrowers can swap an old fixed payment for a cheaper one, and the effect is path dependent, since refinancing today shrinks the pool available to refinance tomorrow, so that easing and tightening act asymmetrically: a rate cut prompts voluntary action among eligible borrowers, while a rate increase leaves completed refinancings untouched. Monetary history, in this way, gets written into household contracts.

Cross-country structure makes the asymmetry hard to generalize, for Norway runs heavily on floating-rate debt while the United States leans on long fixed-rate mortgages, and investor ownership and refinancing institutions differ too, so that although the Norwegian evidence pins down important distributional forces, its quantitative balance may tilt in the American setting. I wanted wider error bars around that translation, since a model calibrated to one contract regime can overstate the reach of its own transmission mechanism.

A second Ganong and Noel study isolates a related distinction between wealth and liquidity, for during the Great Recession some borrowers received principal reduction that moved long-run net worth without touching the monthly payment, while others received payment relief that raised cash flow at once, and spending responded strongly to the payment change and weakly to the principal change, default prevention likewise favoring the liquidity intervention within the range studied (Ganong and Noel 2020).

Once I translated the treatments into a household budget, the result felt obvious, for a deeply underwater borrower stays underwater after a moderate principal reduction, and current consumption can stay pinned because the mortgage bill has not moved, whereas payment relief lands in this month’s budget. Policy had embraced principal forgiveness after cross-sectional links between negative equity and default were read as causal, and the experiment identified liquidity instead as much of the channel that actually moves behavior.

The political implications are severe, for rules favoring principal modification may have crowded out private alternatives that offered more effective payment relief, and the estimated public cost of stopping a foreclosure through the weaker channel ran far above any plausible social benefit, so that a policy sold as help for borrowers could leave many of them with worse options. The case shows why mechanism testing should come before national coercion whenever it can: intuition deserves a pilot before it becomes a mandate.

Student debt relief raises a related question in different institutions, for canceling principal raises net worth and can change credit access, while cutting payments can free up current cash, and borrowers in default may respond differently from borrowers making regular payments; and evidence on student loan distress shows defaults concentrated among people with low balances who attended institutions with weak outcomes, which complicates any policy aimed at the largest nominal debts (Looney and Yannelis 2015). A useful design starts from the constraint it means to relieve.

Across these household mechanisms, liquidity becomes the common language of macroeconomic transmission, for fiscal transfers work through spendable resources, unemployment insurance holds off a cliff while it lasts, mortgage refinancing changes the monthly obligation, inflation redistributes through nominal contracts, and job loss does more damage when liquid buffers are thin, so that policies with similar effects on net worth can have very different effects on consumption. The balance sheet has a time dimension, and households live at its short end.

Expectations and Public Trust

Attention and expectations expose a force present in nearly every policy problem, for macroeconomic policy works partly through beliefs about the future: a household weighing whether to spend a transfer forms a view about income risk, a firm setting a price forms a view about costs and demand, a borrower mulling a refinance forms a view about future rates, and a worker planning a search forms a view about labor demand. These beliefs can be wrong and still be reasonable, since a household is not running a structural model so much as running late, and information is costly, attention is limited, time is scarce, and the relevant signal often arrives through an institution the recipient distrusts.

Research on subjective inflation expectations documents wide disagreement across households and firms, for people weight the prices they meet often, especially the salient purchases, and their expectations bend with personal experience and demographic position (D’Acunto et al. 2022), which helps explain why official inflation statistics can feel detached from everyday judgment. The index stands for an average basket, whereas a particular household carries a different basket and may meet its sharpest increase in the very category that dominates its monthly anxiety.

The gap between an index and an experience matters for communication, for telling a household that inflation is falling can sound absurd while the price level stays high, since disinflation means prices rising more slowly, yet many listeners hear instead a promise that old prices will come back (it is the news that the car is decelerating, delivered to a passenger who wanted it to reverse). The statement can thus be accurate and a communicative failure at the same time, and a central bank that ignores the difference can spend trust at exactly the moment credibility is most valuable.

Firm expectations carry similar frictions, for survey evidence finds wide dispersion among managers and shows that many firms update slowly when new macroeconomic information arrives (Coibion et al. 2018), a result that unsettled the representative-firm language of basic models. Businesses run on different information systems and planning horizons, and where a large company may keep economists on staff, a small restaurant may infer inflation from invoices and the wages down the street (the representative firm of the textbook turning out, on inspection, to be a great many firms that have never met), so that monetary policy enters these organizations through uneven channels of attention.

Information frictions can therefore move real outcomes, for a firm bracing for persistent cost growth may raise prices or hold back investment, a household bracing for recession may save even as current income improves, a bank bracing for losses may tighten credit, and a worker bracing for layoffs may put off a purchase, so that collective pessimism can turn partly self-confirming through the weaker demand it produces. Sentiment matters economically once it changes decisions, and the empirical problem is to separate justified concern from narratives that have built momentum past the underlying data.

Social media reshapes the environment in which those narratives travel, and when Allcott and his coauthors randomly encouraged participants to deactivate Facebook before the 2018 midterm election, deactivation raised offline activity and modestly lifted subjective well-being even as it cut exposure to political information and trimmed polarization a little (Allcott et al. 2020). The mixed result is useful precisely because the platform connects users, distributes information, supplies entertainment, and devours attention all at once (a single product that is a town square, a newspaper, a casino, and a clock), so that welfare hangs on effects that pull in opposite directions.

The mental health evidence raises a graver concern, for Braghieri and his coauthors, using the staggered arrival of Facebook across colleges, find student mental health worsening after access spread (Braghieri et al. 2022), and although the early service was not the current ecosystem, which bounds the design’s reach, the study still shows that social comparison and digital interaction can move well-being at population scale. An information technology can thus become a macroeconomic concern through health and labor supply long before it ever shows up in conventional productivity measures.

Platform welfare folds together design incentives, user value, attention effects, and market power, dimensions that make the problem harder than a standard consumer surplus calculation and that explain, too, why revealed preference falls short when the product is engineered to shape its own continued use.

The folk economics of housing belongs to the economics of attention as well, for weak mental models leave public opinion at the mercy of visible stories, so that a new building rises just as rents climb and construction takes the blame for the price level while the missing counterfactual stays abstract. An effective explanation, then, has to make that absent comparison concrete without dismissing what residents actually see, and an estimate turns politically usable only when interpretation connects it to lived experience.

Trust also depends on honesty about uncertainty, for central banks often fear that admitting uncertainty will weaken credibility, whereas false precision can cost more when the forecast fails (a forecast given to one decimal place and missed by two being its own kind of confession), and a range of plausible outcomes can be more credible when it comes paired with a clear reaction function. The public needs to know which evidence would change policy, so that communication becomes an extension of empirical method, stating the current estimate and the conditions under which that estimate will be revised.

Public misunderstanding grows from unstable beliefs, from policy language that withholds the mechanism, from lived experience at odds with official averages, and from distrust built by earlier failures, so that responsibility falls on the recipient and the institution alike, and better information needs a message able to survive contact with the household’s own evidence.

These findings reinforce a central principle, that a belief becomes an empirical object once it can be measured and tied to behavior, even as its content still calls for interpretation, for a survey answer may reflect confusion or inattention but may equally reflect a private economic reality the researcher has averaged away. Data-driven policy should examine that gap before correcting the person who reports it, since public trust begins with the possibility that disagreement carries information.

Adjustment Beyond the Aggregate

Trade offers a clean example of a national gain paired with a local loss, for although standard theory explains why exchange can raise total income when countries specialize, the China import shock literature asks how that gain actually entered labor markets, and Autor and his coauthors trace rising Chinese import competition across American commuting zones, using growth in Chinese exports to other wealthy countries as an instrument for the supply shock. The design separates China’s expanding manufacturing capacity from shifts in American demand, and a falsification exercise built on earlier local manufacturing conditions further supports an external supply-shock reading of the later exposure (Autor et al. 2013).

The scale of the estimate made me skeptical at first, since a single trading relationship seems to account for a remarkable share of the fall in American manufacturing employment over the period studied (an estimate large enough that one’s first instinct is to go looking for the mistake); but the instrument won my confidence because its logic matches the historical event, for Chinese productivity and market access grew across many destinations at once, and regions specialized in the exposed industries then absorbed the larger employment losses, so that the design gives causal structure to a transformation already visible in national data.

Local adjustment moved slowly next to the frictionless transition that basic trade models imply, for employment and labor force participation sagged in exposed manufacturing communities, wage losses lingered, and mobility brought little relief, so that a commuting zone built around furniture or apparel could stay depressed long after cheaper imports were already benefiting consumers elsewhere (the savings on the sofa accruing to the whole country, the cost of building it to a single town). That persistence makes transition a central part of welfare rather than a footnote to the gains.

Transfer payments exposed the weakness of the American response, for growth in Social Security Disability Insurance dwarfed growth in Trade Adjustment Assistance across the exposed regions, and a program built for workers harmed by trade stayed small beside a disability system with weak paths back to work (the country, lacking a working door marked “adjustment,” having quietly filed the displaced under “disabled”). The pattern revealed a political failure behind the employment estimate, since the state recognized displacement through a category of incapacity because its adjustment institutions could not reach far enough, trade policy and social insurance having drifted apart.

The later review of the China shock reinforces the persistence, for adjustment drags across local employment and earnings over a long horizon, which strains the expectation that workers will move quickly into expanding sectors (Autor et al. 2016), while Pierce and Schott connect the accelerated decline of manufacturing to the change in trade policy certainty around China’s accession to the World Trade Organization (Pierce and Schott 2016). Together these studies show that expectations and exposure both matter, since firms can reorganize before any tariff changes once the uncertainty about future access lifts.

Evidence from recent tariffs shows why adjustment policy cannot fall back on broad protection, for tariffs work through higher costs for consumers and import-using firms, and the washing machine episode brought substantial price increases after protection, with the increases spilling onto complementary dryers (the tariff fell on washers, and the dryers, which no one had touched, raised their prices out of solidarity), so that although domestic production rose, the consumer cost per job created was extremely high (Flaaen et al. 2020). A broad tariff also taxes the inputs American producers use, so that protection for one factory can weaken another farther down the supply chain.

Retaliation adds a geography of its own, for exporting regions can lose market access as foreign governments answer back, so that the policy can trade one set of local harms for another, and recent evidence on the Trump tariffs finds little aggregate employment benefit in protected areas alongside meaningful electoral effects, which suggests that symbolic recognition can matter even when the economic gains stay weak (Autor et al. 2024). A tariff signals that a government sees a community, and being seen turns out to be worth a surprising amount even when it is expensive and does not work; that political work helps explain its appeal even as it leaves the efficiency cost fully in place.

These results turned my attention from the objective to the tool, for rebuilding productive capacity can serve national security or regional stability, yet a uniform tariff reaches that objective only indirectly and at broad collateral cost, whereas industrial policy can aim at learning spillovers, coordination failures, strategic inputs, or neglected places more directly. Each rationale demands its own evidence, since a subsidy to a politically connected industry can survive in the language of resilience even when the market failure beneath it is vague (every subsidy, asked nicely, can produce a national-security rationale).

The new industrial policy literature gives the choice some structure, for Juhasz and her coauthors argue that successful interventions often address coordination problems and learning externalities inside a broader development strategy (Juhasz et al. 2024), and the state capacity that matters includes selection, monitoring, evaluation, and the nerve to cut off a failing project (the last being the rarest, since the firm always has one more quarter’s worth of reasons), since industrial policy loses credibility the moment every subsidized firm can recast failure as a request for more time.

The supply chain evidence strengthens a narrow case for intervention, for firms may underinvest in resilience when the social cost of a broken link runs past the private cost any one buyer can see, so that strategic inventories or diversified sourcing can lower systemic risk and public procurement can build demand for domestic capacity in critical goods. These policies should stay tied to a defined vulnerability, however, since the earthquake study shows that networks propagate shocks while leaving the relative safety of national production and diversified international sourcing to case-by-case analysis.

Worker adjustment deserves the same specificity, for Hyman’s quasi-random evidence on Trade Adjustment Assistance suggests that retraining can improve later outcomes for some displaced workers, especially when the education connects to credible labor demand (Hyman 2018). Active adjustment, in other words, can work; but access and effectiveness are separate questions, and the low take-up in China shock regions is an institutional failure in its own right, since a useful intervention can stay socially irrelevant when eligibility is obscure or enrollment arrives only after financial collapse (a program no one can find being, for policy purposes, a program that does not exist).

Wage insurance offers another margin, for a displaced worker who takes a lower-paying job can receive temporary compensation for part of the earnings gap, a design that rewards reemployment while admitting that old firm-specific value may be impossible to recover at once, and that can preserve labor force attachment for workers poorly served by a blanket training requirement. Older employees in routine occupations may gain the most, since a full occupational reset is often unrealistic, and adjustment policy should treat partial recovery as a legitimate outcome.

The Brexit evidence broadens the warning to barriers raised in the name of sovereignty, for survey-based work links the post-referendum years to weaker investment and productivity among British firms, with uncertainty arriving well ahead of any full change in trading rules (Bloom et al. 2025). The episode is not Chinese import competition, yet the comparison pays off, since trade integration can concentrate losses while trade separation can diffuse them, letting the cost accumulate through forgone investment, so that a policy can feel protective long before its macroeconomic cost comes into view (the bill for a border arriving slowly, in the form of investments quietly never made).

The tariff evidence changed my understanding of efficiency, for a policy with the largest estimated surplus can turn out inferior once administrative failure and political reversal enter the model, whereas a second-best instrument can become defensible when it is enforceable and durable. The claim needs restraint, even so, because “practicality” can excuse almost any favored intervention (the word “practical” having rescued a great many impractical ideas), and evidence should establish both the failure of the cleaner alternative and the mechanism through which the feasible policy actually improves welfare.

The comparisons that matter set reform beside a damaged status quo or another imperfect instrument, for a minimum wage meets labor market power, a housing subsidy enters a constrained neighborhood, unemployment insurance answers scarce jobs and thin liquidity, and a tariff enters a country with weak adjustment policy, so that the counterfactual decides the result, and the damaged status quo is the counterfactual that policy actually faces (reform being measured not against perfection, whatever its critics pretend, but against the mess already in place).

Data can sharpen that comparison by revealing incidence, by asking who receives the transfer, who changes behavior, which price moves, and which effect lasts, questions that return analysis to the mechanism and that mark, too, where an average has become ethically inadequate. A neutral mean can hide a severe loss for one group behind a moderate gain for another, a positive national effect can coexist with regional decline, a successful intervention can improve an outcome and still exclude the population that needed it most, and a program can raise total welfare and yet deepen distrust among the very people facing access barriers.

I no longer treat an empirical result as a verdict, for its authority rests on the institutional setting that produced it, the alternatives it survives, the population it covers, and the horizon over which its effect holds, so that the strongest studies are those which make their boundaries visible and keep the open questions open, earning their credibility through restraint.

I also grew more demanding of policy ambition, having come to see that empirical humility can live alongside bold action: the allocation of talent study suggests that exclusion has imposed enormous productive costs, the Opportunity Atlas shows that childhood place shapes adult possibility, the job loss evidence reveals scars that outlast the recession that caused them, and the unemployment insurance evidence reveals the cost of an abrupt benefit cliff, so that uncertainty can itself argue for action when the downside of waiting is large.

The governing question is one of fit, for a policy should reach the margin the evidence identifies and the population that bears the constraint, whereas principal reduction failed to move current cash flow for many underwater borrowers, retirement subsidies reward households already able to save, college mobility hinges on admission, and trade assistance stays weak when displaced workers face barriers to entry, each case a mismatch between a social objective and the mechanism chosen to chase it (good intentions delivered through the wrong instrument arriving, reliably, at the wrong address).

I now understand data-driven macroeconomic policy as a practice of moving between scales, in which national accounts name the event, administrative records reveal its distribution, causal designs isolate a mechanism, and institutional analysis decides whether the result can guide action, movements that guard policy against the seduction of a single number while keeping empirical work from drifting away from the lives stored in its rows.

My surprise often exposed an assumption I had carried into the evidence, for sometimes skepticism caught a real limit, while at other times it caught only my own resistance to a result that unsettled a familiar story (it is humbling how often “this cannot be right” turns out to mean “I did not want this to be right”), and I learned to ask whether the discomfort came from the research design or from my expectation, a self-scrutiny that belongs next to the formal econometrics.

Several of my positions moved in the process, for though I still favored steeper progressive taxation, I came to put more weight on administrative durability; I raised the priority I give unemployment insurance and narrowed my support for debt relief toward cash-flow mechanisms; I came to see monetary easing as distributionally broader once employment and refinancing sat in the same analysis; and I grew more supportive of housing construction where supply reform comes paired with protection during the transition. Each revision, in the end, tightened a single demand: that the tool match the mechanism.

Other views, by contrast, hardened: initial conditions matter because wealth and place shape the choices that come later, liquidity matters because households decide inside monthly budgets, adjustment matters because a gain measured across the nation can coexist with a loss that lasts years in one community, and political institutions matter because a policy lives mostly on paper when it lacks durability or access, lessons that keep recurring across topics which had first seemed unrelated.

I also grew more aware of the gap between identifying a harmed group and knowing how to help it, for the job loss paper can rank vulnerability with remarkable precision, the housing literature can pick out the neighborhoods where opportunity is greater, wealth data can locate concentration at the extreme top, and inflation data can show which households carry a price shock, so that diagnosis creates responsibility while leaving the remedy uncertain, and policy then has to face mobility costs and administrative limits along with behavior and political resistance. Evidence narrows the search; institutional invention completes the task.

Identification has become inseparable from the form of an economic argument, for an instrument points attention at one source of variation, an event study controls the order in which evidence appears, a robustness test answers an objection before it is raised, and a sampling rule fixes the population the claim covers, so that the strongest empirical arguments are those which arrange the evidence while preserving the uncertainty around it.

I distrust models offered as complete worlds, for every model chooses a set of forces and leaves the rest outside (a map that showed everything would be the size of the territory and just as hard to read), and although the choice can be productive when the boundary is clear, trouble starts the moment a local mechanism is promoted into a universal policy claim. I now ask what a framework makes visible and what it renders silent, a question that belongs to economic judgment as much as to literary criticism.

Disagreement need not harden into dismissal, for I can accept an estimate while resisting its extrapolation, back an objective while faulting the chosen instrument, grant a behavioral response while leaving its welfare value open, and revise a prior without treating the new result as final, distinctions that let agreement coexist with uncertainty.

The most durable lesson concerns the average. Macroeconomic policy needs aggregates, because governments must act on economies too large for any dataset to narrate household by household; yet the average turns dangerous the moment it is allowed to close the inquiry, for after the average comes incidence, after incidence the mechanism, after the mechanism the test of feasibility, after feasibility the question of whose welfare counts and when, and after that, whenever the evidence shifts, the obligation to begin the whole reckoning again. Data cannot take that last step for anyone. What they can do is put the foreclosed county, the older worker in the emptying town, and the family priced out by the building that was never approved back into the sentence, and then refuse to let the average finish it on their behalf. The average is where governing begins; it was never meant to be where judgment ends.

Back to top ↑

Works Cited

• Acemoglu, Daron. 2024. “The Simple Macroeconomics of AI.” NBER Working Paper 32487.
• Acemoglu, Daron, and Pascual Restrepo. 2019. “Automation and New Tasks: How Technology Displaces and Reinstates Labor.” Journal of Economic Perspectives 33(2).
• Akcigit et al. 2016. “Taxation and the International Mobility of Inventors.” American Economic Review 106(10).
• Allcott et al. 2020. “The Welfare Effects of Social Media.” American Economic Review 110(3).
• Athey et al. 2026. “The Heterogeneous Earnings Impact of Job Loss.” NBER Working Paper 34946.
• Autor, David. 2015. “Why Are There Still So Many Jobs? The History and Future of Workplace Automation.” Journal of Economic Perspectives 29(3).
• Autor et al. 2013. “The China Syndrome: Local Labor Market Effects of Import Competition in the United States.” American Economic Review 103(6).
• Autor et al. 2016. “The China Shock: Learning from Labor Market Adjustment to Large Changes in Trade.” Annual Review of Economics 8.
• Autor et al. 2024. “Help for the Heartland? The Employment and Electoral Effects of the Trump Tariffs.” NBER Working Paper 32082.
• Baker, Scott R. 2018. “Debt and the Response to Household Income Shocks.” Journal of Political Economy 126(4).
• Bell et al. 2019. “Who Becomes an Inventor in America? The Importance of Exposure to Innovation.” Quarterly Journal of Economics 134(2).
• Bloom et al. 2025. “The Economic Impact of Brexit.” NBER Working Paper 34459.
• Boerma, Job, and Loukas Karabarbounis. 2021. “Reparations and Persistent Racial Wealth Gaps.” NBER Working Paper 28468.
• Braghieri et al. 2022. “Social Media and Mental Health.” American Economic Review 112(11).
• Brynjolfsson et al. 2023. “Generative AI at Work.” NBER Working Paper 31161.
• Busso et al. 2013. “Assessing the Incidence and Efficiency of a Prominent Place-Based Policy.” American Economic Review 103(2).
• Carvalho et al. 2021. “Supply Chain Disruptions: Evidence from the Great East Japan Earthquake.” Quarterly Journal of Economics 136(2).
• Chetty, Raj. 2008. “Moral Hazard versus Liquidity and Optimal Unemployment Insurance.” Journal of Political Economy 116(2).
• Chetty et al. 2016. “The Effects of Exposure to Better Neighborhoods on Children.” American Economic Review 106(4).
• Chetty et al. 2018. “The Opportunity Atlas: Mapping the Childhood Roots of Social Mobility.” NBER Working Paper 25147.
• Chetty et al. 2020a. “Income Segregation and Intergenerational Mobility Across Colleges in the United States.” Quarterly Journal of Economics 135(3).
• Chetty et al. 2020b. “Race and Economic Opportunity in the United States.” Quarterly Journal of Economics 135(2).
• Choukhmane et al. 2024. “Who Benefits from Retirement Saving Incentives in the United States?” NBER Working Paper 32843.
• Clemens, Jeffrey. 2021. “How Do Firms Respond to Minimum Wage Increases?” Journal of Economic Perspectives 35(1).
• Coibion et al. 2018. “How Do Firms Form Their Expectations? New Survey Evidence.” American Economic Review 108(9).
• Cunningham, Scott. 2021. Causal Inference: The Mixtape. Yale University Press.
• D’Acunto et al. 2022. “Subjective Inflation Expectations of Households and Firms.” Journal of Economic Perspectives 36(3).
• Davis, Steven J., and Till von Wachter. 2011. “Recessions and the Costs of Job Loss.” Brookings Papers on Economic Activity.
• Derenoncourt, Ellora. 2022. “Can You Move to Opportunity? Evidence from the Great Migration.” American Economic Review 112(2).
• Diamond, Peter, and Emmanuel Saez. 2011. “The Case for a Progressive Tax.” Journal of Economic Perspectives 25(4).
• Diamond, Rebecca, and Tim McQuade. 2019. “Who Wants Affordable Housing in Their Backyard?” Journal of Political Economy 127(3).
• Diamond et al. 2019. “The Effects of Rent Control Expansion.” American Economic Review 109(9).
• Doepke, Matthias, and Martin Schneider. 2006. “Inflation and the Redistribution of Nominal Wealth.” Journal of Political Economy 114(6).
• Dube, Arindrajit, and Attila Lindner. 2021. “City Limits: What Do Local-Area Minimum Wages Do?” Journal of Economic Perspectives 35(1).
• Dynarski et al. 2018. “Closing the Gap: The Effect of a Targeted Tuition-Free Promise.” NBER Working Paper 25349.
• Elmendorf et al. 2025. “The Folk Economics of Housing.” Journal of Economic Perspectives 39(3).
• Flaaen et al. 2020. “The Production Relocation and Price Effects of United States Trade Policy.” American Economic Review 110(7).
• Ganong et al. 2020. “Consumption Smoothing of Typical Income Shocks.” NBER Working Paper 27552.
• Ganong, Peter, and Pascal Noel. 2019. “Consumer Spending During Unemployment: Positive and Normative Implications.” American Economic Review 109(7).
• Ganong, Peter, and Pascal Noel. 2020. “Liquidity versus Wealth in Household Debt Obligations.” American Economic Review 110(10).
• Golosov et al. 2024. “How Americans Respond to Idiosyncratic and Exogenous Changes in Household Wealth and Unearned Income.” Quarterly Journal of Economics 139(2).
• Hendren, Nathaniel, and Ben Sprung-Keyser. 2020. “A Unified Welfare Analysis of Government Policies.” Quarterly Journal of Economics 135(3).
• Hsieh et al. 2019. “The Allocation of Talent and United States Economic Growth.” Econometrica 87(5).
• Humlum, Anders, and Emilie Vestergaard. 2025. “Large Language Models, Small Labor Market Effects.” NBER Working Paper 33777.
• Hyman, Benjamin. 2018. “Can Displaced Labor Be Retrained?” Working paper.
• Jones, Charles I. 2015. “Pareto and Piketty: The Macroeconomics of Top Income and Wealth Inequality.” Journal of Economic Perspectives 29(1).
• Juhasz et al. 2024. “The New Economics of Industrial Policy.” Annual Review of Economics 16.
• Kaplan, Greg, and Giovanni L. Violante. 2014. “A Model of the Consumption Response to Fiscal Stimulus Payments.” Econometrica 82(4).
• Kermani, Amir, and Francis Wong. 2026. “Racial Disparities in Housing Returns.” American Economic Review 116(1).
• Kline, Patrick, and Enrico Moretti. 2014. “Some Simple Welfare Economics of Local Economic Development Programs.” Annual Review of Economics 6.
• Kopczuk, Wojciech. 2019. “Comment on Progressive Wealth Taxation.” Brookings Papers on Economic Activity.
• Looney, Adam, and Constantine Yannelis. 2015. “A Crisis in Student Loans?” Brookings Papers on Economic Activity.
• McKay, Alisdair, and Christian K. Wolf. 2023. “Monetary Policy and Inequality.” Journal of Economic Perspectives 37(1).
• Nakamura, Emi, and Jon Steinsson. 2014. “Fiscal Stimulus in a Monetary Union.” American Economic Review 104(3).
• Oreopoulos et al. 2012. “The Short- and Long-Term Career Effects of Graduating in a Recession.” American Economic Journal: Applied Economics 4(1).
• Pallotti et al. 2023. “Who Bears the Costs of Inflation? Euro Area Households and the 2021 to 2022 Shock.” European Central Bank Working Paper 2877.
• Pierce, Justin R., and Peter K. Schott. 2016. “The Surprisingly Swift Decline of United States Manufacturing Employment.” American Economic Review 106(7).
• Sacerdote, Bruce. 2007. “How Large Are the Effects from Changes in Family Environment?” Quarterly Journal of Economics 122(1).
• Saez, Emmanuel, and Gabriel Zucman. 2019. “Progressive Wealth Taxation.” Brookings Papers on Economic Activity.
• Smith et al. 2019. “Capitalists in the Twenty-First Century.” Quarterly Journal of Economics 134(4).
• Zimmerman, Seth D. 2014. “The Returns to College Admission for Academically Marginal Students.” Journal of Labor Economics 32(4).