RESEARCH DOSSIER

Heterogeneous Agents and Economic Aggregation

Abstract. Economic aggregates are functions of individual choices, resources, constraints, institutions, and the distribution linking them. This monograph examines the conditions under which heterogeneous economies admit exact or approximate aggregation, and the circumstances in which aggregate inference requires distributional information. It integrates the historical literature on demand representation, incomplete-market macroeconomics, precautionary saving, liquidity-dependent spending, monetary and fiscal transmission, firm productivity, production networks, and distributional measurement. Formal illustrations distinguish accounting identities from behavioral representations, show why unequal marginal responses and the assignment of shocks affect aggregate demand, and demonstrate the conditional productivity consequences of factor allocation. Evidence from published research and institutional statistics is evaluated in relation to its population, measurement definitions, model assumptions, and counterfactual scope. The analytical synthesis establishes that the sufficiency of a representative agent is specific to an economic object, intervention, institutional regime, and accuracy criterion. Distributional data have explanatory value when they preserve the dependence among decisions, constraints, and shock exposures that determines the aggregate under investigation. No original econometric estimate, numerical replication of a structural model, or newly named theoretical system is claimed.

Keywords: Heterogeneous agents; economic aggregation; representative agent; incomplete markets; wealth distribution; liquidity; marginal propensity to consume; policy incidence; general equilibrium; misallocation; production networks; identification

The object of aggregation

Economic aggregation transforms a collection of decisions, endowments, exposures, and interactions into a quantity or relationship defined at a higher level of organization. A country's consumption expenditure, for instance, is the sum of appropriately defined household expenditures. A country's consumption function is a different object: it describes how that sum changes when income, prices, wealth, expectations, or institutions change. The first operation is an accounting construction. The second is an explanatory claim about behavior. The difference is central to the scientific assessment of representative agents.

An agent is a decision-making unit within a specified model. A household, firm, financial intermediary, or government can be assigned a decision problem, but the correspondence between modeled units and observed statistical units must be stated. A household includes people who may pool resources imperfectly; a firm may include establishments with distinct production technologies. A person-count-weighted average and a household-count-weighted average answer different questions, even when both are computed accurately. Aggregation therefore starts with the identification of the population, unit, accounting boundary, time interval, and outcome.

Heterogeneity refers here to variation across units in states, preferences, information, constraints, technologies, institutional treatment, or connections. A common preference specification can coexist with very different decisions when initial assets or labor incomes differ. Different preferences can coexist with similar observed actions under binding restrictions. Observable dispersion alone cannot establish which mechanism is responsible. The scientific question is whether a dimension of variation changes the statistic or counterfactual that the researcher seeks to explain.

The dossier concerns the distributional foundations and limitations of aggregate inference. Exact demand aggregation and distributional computation are treated only to the extent necessary for this object; detailed characterization theorems and numerical solver design belong to adjacent specialized studies. The research proceeds by analytical reconstruction, explicit counterexamples, and comparative reading of published results. Numerical illustrations below are constructed examples in normalized units, not independently estimated national statistics.

From equilibrium existence to aggregate behavior

The modern aggregation problem grew alongside formal general equilibrium. Arrow and Debreu's (1954) equilibrium existence study integrates consumer choice, production, and exchange in a conditional mathematical existence result. The existence of a clearing price vector gives an answer to a feasibility-and-consistency question under specified assumptions. A claim of unique equilibrium, a stable price-adjustment process, or a single utility function reproducing all market demand requires additional arguments.

The historical literature on market excess demand includes Sonnenschein's (1972) market excess demand functions, Mantel's (1974) aggregate excess demand characterization, and Debreu's (1974) excess demand functions. The common lesson of this research tradition is the limited inheritance of individual restrictions by market-level functions. Its mathematical characterization has domain, consumer-number, regularity, and boundary qualifications; this dossier uses it to motivate a methodological distinction rather than asserting a new or unrestricted characterization theorem.

Kirman's (1992) representative individual critique directs attention to the behavior of groups whose members exchange, coordinate, or fail to coordinate with one another. A model in which all members face the same choices and own identical claims can eliminate some of the interactions that explain actual outcomes. A reduced agent can still have predictive utility for selected aggregates. That utility must be established for the question and regime at issue, particularly when distribution, unemployment, or trade is itself the outcome of interest.

Three distinct inquiries emerged: which market aggregates admit exact representation, which dynamic economies admit useful lower-dimensional approximations, and which mechanisms require distributional state variables. A failure of exact aggregation is compatible with an accurate approximation for a specific statistic. Conversely, accurate fit of a long historical average can coexist with large errors in response to policy. The three inquiries therefore require different standards of evidence.

Population measures and accounting identities

Let i denote an economic unit, w_i its nonnegative population weight, and x_i a scalar outcome measured on a common basis. Normalize weights so that their sum is one. The population mean is X = sum_i w_i x_i; the total equals N X when N is the population count represented by those weights. With unnormalized survey weights, the corresponding mean divides the weighted sum by the sum of weights. If the underlying unit changes from household to person, N and the weights change accordingly.

A distribution μ assigns probability mass to observable or modeled states s. An aggregate outcome can be written X = integral x(s) dμ(s). The notation separates the decision rule x from the population distribution μ. In continual time or discrete dynamics, either can change. Monetary valuation requires a price and nominal unit; real quantities require a deflator or a quantity index appropriate to the comparison. Adding nominal balances without reconciling currency and valuation date produces no meaningful cross-country wealth statistic.

The median is an order statistic rather than a sum. A median wealth household generally cannot reproduce the total of wealth, the mean spending response, or the marginal investor's valuation. Even replacing the median with the arithmetic mean does not resolve these problems. Aggregation of ratios requires a denominator: a country's aggregate debt-to-income ratio is the ratio of total debt to total income, which is an income-weighted average of individual debt-to-income ratios if denominators are positive. An unweighted mean of ratios answers a different question. Zero income requires separate treatment.

An accounting identity makes measured levels consistent with their components. A behavioral equality holds only if the assumptions generating it are satisfied. Confusing the two lets a mechanically correct total acquire an unwarranted interpretation. The analysis of aggregation is consequently a study of both measurement and structural response.

Exact aggregation of demand

Consider demand for one good at fixed prices p by agents with expenditure or income y_i. If individual demand has the form q_i(p,y_i)=a_i(p)+b(p)y_i with a common slope b(p), total demand is the sum of intercepts plus b(p) times total income. Redistribution of a given total income leaves this particular market demand unchanged as long as the demand formula remains valid and the population is fixed.

The intercepts permit heterogeneous demand levels; common slopes produce distributional invariance. If two agents have unequal slopes b_1 and b_2 and a transfer t moves income from agent 2 to agent 1, the change in combined demand is (b_1-b_2)t. This identity identifies why linearity alone is insufficient. Heterogeneous but parallel affine rules can aggregate exactly; affine rules with unequal slopes generally retain information about distribution.

Gorman's (1953) community preference fields is a foundational treatment of this relationship between individual Engel curves and community representation. A standard polar-form illustration takes individual indirect utility to be v_i(p,y_i)=A_i(p)+B(p)y_i, with common positive B and differentiable functions. Roy's identity yields q_ij=-(∂A_i/∂p_j + y_i∂B/∂p_j)/B. Each income coefficient is shared across individuals. This is an illustration under the required regularity and normalization of utility representations, rather than a proof that every possible representation is valid for all prices and income levels. Corners, nonconvexities, changing membership, or nonparallel slopes can narrow the applicable domain.

Exact aggregation of market quantities also does not automatically provide an ethical ordering of alternative distributions. Welfare representation requires a specification of whose utility counts, how it counts, and whether the proposed comparisons respect interpersonal preferences and the relevant normative criteria.

Nonlinearity and the omitted distribution

At fixed prices, take a twice differentiable scalar decision function f. If x has mean m, then a Taylor expansion near m yields E[f(x)] approximately equal to f(m) plus one half of f''(m) Var(x), provided a suitable remainder is small. The first-order linear term averages to zero. With substantial dispersion, asymmetry, or discontinuities, the neglected terms may be large. For convex f, Jensen's inequality gives E[f(x)] at least f(E[x]) when expectations exist; concavity reverses the inequality.

Consider two equally weighted populations. In A, both agents have resources 5. In B, their resources are 1 and 9. Suppose a stylized decision is f(x)=sqrt(x). Both populations have mean resources 5. Mean decision in A is sqrt(5), approximately 2.236, while mean decision in B is (1+3)/2=2. A representative decision at the mean therefore overstates the second population's average by about 0.236. The example uses arbitrary units and an illustrative concave function, not an empirical household consumption equation.

The implications grow stronger when nonlinearities are concentrated near constraints. A small difference in liquid resources can determine whether a household delays a necessary purchase, an entrepreneur hires a worker, or a lender breaches a capital threshold. A distribution with a substantial mass near the threshold may be sensitive to a shock even while mean wealth remains high. Quantile information, tail probabilities, or measures of proximity to a constraint can then explain more than an additional decimal place in the mean.

There is a useful limit to this argument. When the decision rule is affine on the entire support of interest, dispersion does not affect the aggregate through curvature. When the rule is only locally affine, accuracy depends on the support after the contemplated shock, not simply the baseline support. This explains why aggregation tests should include changes in policy magnitude as well as small perturbations.

Forms of representative-agent representation

A representative agent may be a statistical average, a structural agent whose choices reproduce aggregate demand, a reduced-form dynamic state, or a normative construction whose utility represents a social ordering. These four roles are often presented using similar notation, but their claims differ. A statistical mean is defined by arithmetic. A structural representation requires conditions on preferences, endowments, or market arrangements. A dynamic representation requires an approximation to transitions. A normative construction requires explicit ethical premises.

For a particular equilibrium path, a researcher may infer a residual aggregate preference or technology term that makes a single-agent model fit observed data. Such a residual need not correspond to an invariant household preference. Changing the distribution of consumption or financial exposure can alter the residual even if each household's preference remains unchanged. The residual may still be useful for descriptive accounting, provided it is not silently used as a causal parameter.

Representative-agent models remain useful as transparent benchmarks. Their parsimony helps isolate intertemporal substitution, resource feasibility, market clearing, and comparative statics. Their performance should be evaluated against the relevant counterpart: for aggregate capital in a particular calibrated growth economy, approximation can be strong; for targeted fiscal incidence, it may suppress the central mechanism. The correct comparator therefore changes with the question.

A representative model cannot recover individual welfare from aggregate consumption unless welfare is itself identified by the representation. Even an economy whose market demands aggregate exactly may contain individuals with very different experiences of price changes, unemployment, or wealth transfers. The analyst must preserve that distinction when moving from positive results to policy evaluation.

Risk sharing and incomplete markets

Uncertainty adds state-contingent requirements to aggregation. If households can trade a complete set of contingent claims, have appropriate common beliefs and preferences, and participate without relevant frictions, risk-sharing conditions can make aggregate consumption tractable under additional restrictions. Individual consumption shares can become stable in the familiar identical homothetic preference case with fixed Pareto weights and complete insurance. That conclusion is conditional on the risk-sharing environment.

If insurance against individual employment, health, or business risk is incomplete, households typically finance shocks using current earnings, liquid buffers, borrowing, transfers, or asset sales. An aggregate resource constraint can hold while individual budget constraints differ greatly. Households facing a borrowing limit may place a high marginal value on an extra unit of immediately spendable resources, even when other households can smooth across time.

An illustrative one-asset problem chooses consumption c and next-period assets a' with c+a'=(1+r)a+y, a'≥a_min, and a discounted expected utility objective. For an interior solution with differentiable preferences, the Euler equation is u'(c)=β(1+r)E[u'(c')]. At a binding lower asset limit, the relevant inequality is u'(c)≥β(1+r)E[u'(c')]. The sign follows from the inability to reduce next-period assets further to finance current spending. A transitory windfall can therefore have a large effect on current consumption without invoking different tastes.

Huggett's (1993) incomplete-insurance interest-rate study investigates the safe return in economies with individual risk and imperfect insurance. Aiyagari (1994) subsequently relates uninsurable income risk to aggregate saving and wealth dispersion. The direction and size of a general equilibrium return change depend on the supply of assets, production technology, limits on debt, shock persistence, and other closures. A household incentive to accumulate buffers cannot by itself establish a quantitative national saving effect.

Precautionary saving and stationary distributions

Aiyagari's (1994) precautionary saving study reports a modest aggregate saving-rate increment attributable to uninsured idiosyncratic risk in its moderate-parameter experiments, less than three percentage points in the institutional summary. That figure concerns the paper's specified quantitative model and plausible parameter range. It is neither a measured historical national effect nor an upper bound for economies with different risks, credit markets, or institutional support.

The key advance is the explicit coexistence of individual policy rules, asset-market clearing, and an endogenous distribution of assets. Given a price vector, a policy function and the probability law of idiosyncratic income induce a transition law for household states. A stationary distribution is invariant under that transition. Asset prices in equilibrium must reconcile the desired aggregate holdings with the supply of claims or physical capital specified by the economy.

Stationarity is a distributional property. An individual can move between income and wealth positions even when population shares are unchanging. Consequently, the presence of a stable wealth distribution does not imply the absence of exposure to risk, and stability in aggregate saving says little about the intensity of household asset adjustment. The appropriate empirical counterparts include both cross-sectional distributions and longitudinal transitions.

It is important to separate the prudence mechanism from other saving motives. Under standard conditions, convex marginal utility supports precautionary demand for assets, but retirement, bequests, entrepreneurial investment, housing purchases, transaction costs, and public pensions can generate additional patterns. Cross-sectional wealth dispersion does not uniquely identify preference heterogeneity or the magnitude of uninsured risk. Identification requires evidence on income transitions, consumption, balance sheets, and institutional constraints.

Approximate aggregation in dynamic economies

Krusell and Smith's (1998) income and wealth heterogeneity study reports that the mean of wealth is highly informative about aggregate dynamics in the stationary stochastic equilibrium of its calibrated growth model. The same paper identifies a limitation of its baseline model in matching observed wealth dispersion, and obtains a better distributional match by introducing thrift heterogeneity. The contrast matters: a model can characterize aggregate dynamics effectively while missing cross-sectional features important to another outcome.

To understand the approximation, consider a forecast of next-period aggregate capital using present capital and an aggregate productivity indicator. If the induced individual saving functions are nearly parallel across the region where most households reside, higher moments add little information to the forecast for that economy. The resulting aggregate law can be computationally efficient even though the individual distribution remains part of the full model's state.

An approximation is a proposition about a loss function and a domain. One should examine errors in the statistic's level, one-step prediction, long-horizon equilibrium path, and response to counterfactual interventions. Mean squared prediction error in stationary simulations is insufficient when the relevant policy induces a large shift in liquid wealth, labor earnings, or the fraction of constrained households. Out-of-regime forecasting can expose the missing state variables.

The apparent conflict between approximate aggregation and distribution-sensitive fiscal or monetary responses dissolves once the dependent variable, horizon, policy, and equilibrium regime are specified. Accuracy of capital forecasts does not imply accuracy of household consumption elasticities. The scientific accomplishment of approximate aggregation is precision about where a reduction works, rather than a universal endorsement of a representative household.

Liquidity, portfolios, and the marginal propensity to consume

The marginal propensity to consume (MPC) measures an incremental change in expenditure or consumption per unit of incremental resources over a specified horizon. Its empirical definition depends on whether the shock is unanticipated, transitory, persistent, gross or net of taxes, cash or valuation-based, and on how consumption is observed. An MPC over a quarter is not interchangeable with an annual MPC. Responses to large transfers may differ from the derivative at a small perturbation.

Carroll, Slacalek, Tokuoka, and White (2017) develop a distribution-sensitive calibration in wealth and consumption responses. Their model generates an aggregate annual MPC around 0.2 under one wealth target and around 0.4 under an alternative liquid-wealth target, as described in the author-hosted introduction. These are calibrated model implications. The contrast demonstrates how the chosen balance-sheet target can affect the frequency of households in steep regions of consumption policy rules; it does not establish either number as a universal empirical MPC.

Kaplan, Violante, and Weidner's (2014) wealthy hand-to-mouth households distinguishes liquid resources from illiquid holdings, such as housing and retirement wealth. A household may own valuable assets while having little immediately available liquidity. A single net-worth measure can group together a liquid saver and an illiquid owner whose current spending opportunities differ. This makes the composition of wealth relevant to the aggregate effect of a cash injection.

Liquidity classification is a measurement decision. Credit lines, transaction costs, expected refinancing opportunities, housing collateral, and access to family transfers can alter effective liquidity. A survey measure of deposits alone is an imperfect proxy for that concept. The appropriate analysis reports asset categories, debt, and institutional access and treats alternative classifications as sensitivity checks. The distribution of consumption responses is as important as the level of average net wealth when cash-flow policy is under examination.

Who receives the shock

Consider heterogeneous households with short-horizon spending derivatives m_i and a small individual cash-flow shock d_i. Holding prices and other channels fixed, the direct average consumption response is E[m_i d_i]. Algebra gives E[m_i]E[d_i]+Cov(m_i,d_i). The covariance is computed across the same weighted population. When the transfer goes preferentially to high-MPC households, the direct response exceeds that obtained by multiplying the population mean MPC by the mean transfer, provided the covariance is positive.

For a constructed economy of 100 equally weighted households, let 40 have marginal responses of 0.8 and 60 have responses of 0.1. The population mean MPC is 0.38. A total payment of 1,000 units spread equally implies 380 units of immediate spending in the linear approximation. If the same total goes entirely to the high-MPC group, the amount is 800 units; if it goes entirely to the low-MPC group, 100 units. In all three cases the transfer budget is identical. The example assumes constant slopes for the transfer sizes, no taxation response, no changes in relative prices, and no equilibrium feedback. It demonstrates incidence rather than estimating a multiplier.

For a zero-sum redistribution, the arithmetic mean shock is zero. The direct aggregate spending response is then the covariance term. Suppose the high-MPC group gains 300 units in total and the low-MPC group loses 300. Under the same assumptions, incremental spending is 0.8×300−0.1×300=210 units. The result holds because the spending derivatives of recipients and payers differ, not because the economy has acquired additional financial resources.

A causal estimate of the aggregate response needs the financing rule, timing, anticipated future tax effects, price dynamics, and production capacity. The covariance identity describes one direct channel, not a complete macroeconomic model. It also does not determine welfare: the identity reports expenditure changes, while welfare includes preferences, risk exposure, and the distribution of gains and losses.

Monetary policy and heterogeneous transmission

A policy rate change can affect spending through relative prices of present and future consumption, asset valuations, debt-service costs, labor income, and expectations. An agent whose borrowing rate adjusts immediately has a different cash-flow exposure from one holding a long-duration fixed-rate mortgage. A saver receiving higher interest income may increase spending even when substitution discourages it. These effects depend on contracts, inflation, and equilibrium changes in income.

Kaplan, Moll, and Violante's (2018) monetary policy according to HANK models heterogeneous households with uninsurable income risk and assets with different liquidity. The study's reported comparison finds that indirect labor-demand effects of an unexpected rate cut are larger than direct household substitution effects in its model. Fiscal interaction affects the overall response because financing changes have distributional consequences. The relative sizes are model-specific results rather than invariant facts about every monetary regime.

Auclert's (2019) monetary redistribution channels separates unequal earnings responses, unexpected inflation's revaluation of nominal positions, and exposure to real-interest-rate changes. The paper uses evidence from Italy and the United States to assess these channels. The analytical contribution of this distinction is that a policy shock redistributes resources across households whose propensities to spend differ. Aggregate exposure must therefore be weighted by the relationship between individual gains and behavioral responses.

Net financial positions can cancel across sectors in consolidated accounts even while cash-flow changes matter for behavior. A private liability is an asset to a creditor, but its revaluation may move resources from a high-MPC borrower to a low-MPC creditor or the reverse. Balance-sheet consolidation preserves accounting consistency while losing information about the identity of recipients. The transmission analysis must retain the exposure distribution until the relevant behavioral sums have been computed.

Fiscal instruments, financing, and general equilibrium

A transfer financed by taxes, debt issuance, or reduction in other expenditure is a distinct intervention in each case. Its short-run direct spending effect depends on who receives cash; its total effect depends on who expects future taxes, who holds new government liabilities, how producers adjust output, and whether monetary policy accommodates the change. The government's intertemporal budget constraint supplies a consistency condition rather than a behavioral estimate.

A liquidity-targeted payment can produce a bigger immediate consumption response than an untargeted payment of the same budget if targeted households have higher relevant spending derivatives. Implementation delays, eligibility errors, administrative costs, changes in labor supply, and future financing can change the ranking. The multiplier on output also includes spending on imports and inventories, price adjustments, substitution among goods, crowding in or out of investment, and capacity constraints.

A temporary income change and a permanent income change differ in informational content. Households may revise expected lifetime resources when a nominally temporary intervention signals a future policy regime. An announced tax rebate can influence decisions before disbursement. A program's date of enactment, payment date, observed expenditure window, and expected duration all need separation in empirical designs.

Policy evaluation should report the distributional outcome and aggregate outcome jointly. A program intended as insurance can be valuable even if its immediate aggregate expenditure effect is limited, whereas a high spending response alone is insufficient to establish a social gain. Heterogeneous-agent analysis clarifies these objectives by preserving the identity of beneficiaries and payers rather than collapsing their individual circumstances into a single balance sheet.

Causal evidence on consumption responses

Fagereng, Holm, and Natvik (2021), in MPCs and household balance sheets, exploit substantial lottery prizes and Norwegian administrative panel information to study spending and saving following transitory income shocks. Variation in prize size and household characteristics enables investigation of heterogeneous responses. The design's substantive merits include measurement at the household level and access to balance-sheet information. Its conclusions are conditional on the lottery population, observed expenditure construction, event timing, and identifying assumptions. The present dossier does not re-estimate the study.

Unayama, Komura, and Hattori's (2021) Japanese special cash payment study exploits differences in disbursement timing across municipalities during the COVID-19 period. Its English research summary reports an estimated MPC of approximately 10% and expenditure shifts toward categories less reliant on high-contact services. The value is specific to that design, payment, observation window, and exceptional conditions. It cannot be compared mechanically with annual calibrated MPCs from a different economy.

Both designs illustrate why a single average response may hide explanatory variation. An empirical study of a cash shock should distinguish pretreatment resources, preexisting constraints, the amount and timing of the shock, and the horizon over which behavior is measured. Selection into an event, anticipation, measurement error, and correlated local developments can each undermine a causal interpretation. In timing designs, pre-event dynamics and assumptions about comparison groups deserve explicit examination.

The difference between the observed response of treated units and a counterfactual national response remains important. Scaling a local estimate requires the target population, the contemplated treatment distribution, and general equilibrium adjustment. A lottery prize has different aggregate financing and price effects from a nationwide tax-financed payment. Microeconomic causal identification supplies an ingredient of aggregation, not an automatic substitute for aggregate policy identification.

International comparisons and economic institutions

The same amount of measured net wealth can have different significance across countries because of pension arrangements, mortgage contracts, tax rules, health insurance, household size, housing tenure, and access to unsecured credit. Institutional arrangements change both the observed distribution of states and the transition law after shocks. Cross-country differences in measured MPCs can therefore reflect distinct policies, shocks, measurement procedures, or genuine behavioral mechanisms.

The cross-country household portfolio evidence of Kaplan, Violante, and Weidner (2014) compares the United States, Canada, Australia, the United Kingdom, Germany, France, Italy, and Spain using household balance-sheet evidence. Their publisher summary describes a substantial group with low liquid resources despite illiquid assets, including estimates for the United States based on 2010 data. These historical results help distinguish total wealth from accessible financial resources. They are not current prevalence estimates for those countries.

The ECB's HFCS methodological guidance warns that household survey concepts and national accounts concepts differ and that cross-country comparisons require care concerning household size, housing ownership, survey coverage, and fieldwork. The 2021 wave was affected by pandemic-related disruptions. Comparing an observed median measured on a household basis to a per-person national aggregate would combine unlike concepts.

Institutional comparison is most informative when it specifies a mechanism. Fixed-rate mortgages delay the immediate cash-flow consequences of policy-rate changes for existing borrowers; adjustable-rate contracts can make those changes faster. The size of the resulting consumption effect also depends on refinancing, maturity, debt exposure, savings behavior, and equilibrium income. A measurement crosswalk should therefore accompany a model crosswalk. The present evidence spans North America, Europe, and Japan, with production evidence from China and India. This is a comparative selection rather than comprehensive global estimation.

From household surveys to aggregate accounts

Household financial surveys provide estimates of distributional positions that national totals cannot reveal. The Federal Reserve's 2019–2022 family finance report supplies distributions and participation measures for different assets and liabilities. Its historical 2022 results report that more than 45% of families held a credit-card balance after their last payment, and the median among those holding such debt was $2,700. The conditional median is not a sector-wide average balance, and the participation fraction is not the fraction of dollar debt held by a group.

A distributional accounting exercise should align population coverage, valuation dates, asset definitions, debt instruments, the treatment of pensions, and survey weights. Top wealth can be difficult to observe because of nonresponse, rare units, offshore structures, and complicated valuations. Administrative data can address selected gaps while introducing their own reporting scopes. Matching a survey mean to a national-accounts total by a mechanical proportional factor may distort distributions if discrepancies are concentrated in particular portfolios or groups.

One useful reporting set includes the weighted mean, median, relevant percentile shares, participation rates, conditional medians, and joint distributions of income, liquid assets, illiquid assets, debt service, and relevant shock exposure. Joint distributions matter because matching every marginal distribution separately does not identify their dependence. Two economies can have the same separate distributions of wealth and liability exposure while assigning those liabilities to different households.

The distinction applies equally to consumption. Different survey instruments measure actual expenditures, inferred expenditures, taxable transactions, card transactions, or broader consumption services. Durable purchases can generate services over years; current cash expenditure can therefore differ from contemporaneous consumption in the economic sense. Transparent statistical documentation is part of the scientific model rather than an accessory to it.

Heterogeneous firms and aggregate productivity

Economic aggregation concerns producers as well as consumers. When firm production functions or productivity differ, total output depends on the allocation of factors across firms and their technologies. A change in aggregate productivity may originate in improvements inside existing establishments, entry and exit, changes in market shares, or reallocation of labor and capital across them. The aggregation problem requires a definition of output and the relevant input quantities.

Take two plants with output y_j=A_j sqrt(k_j), a common input price and a total capital supply of 10. Let A_1=2, A_2=1. Under equal allocation k_1=k_2=5, total output is 3 sqrt(5), approximately 6.708. Maximizing output with a fixed resource requires marginal products A_j/(2 sqrt(k_j)) to agree at an interior optimum. It yields k_1/k_2=4, hence k_1=8 and k_2=2. Total output equals 2 sqrt(8)+sqrt(2)=5 sqrt(2), approximately 7.071. This represents an increase of roughly 5.4% over the equal allocation in the constructed example. There are no adjustment costs, uncertainty, or market-distortion measurements in this calculation.

The lesson is conditional: dispersion in physical marginal products signals an opportunity to raise the specified objective by moving a fungible factor when no other constraints or costs intervene. In observed economies one must distinguish marginal revenue products from physical marginal products, prices and markups, risk premia, factor quality, adjustment costs, and measurement error. Observed dispersion alone cannot identify the counterfactual output gain.

The measurement and welfare questions also differ. A reallocation that raises measured aggregate output may impose transition costs, alter employment risk, and change the distribution of producer and consumer surplus. The scientific result should report the objective being optimized, the constraints respected, and the interpretation of the output comparison.

Misallocation evidence and production networks

Hsieh and Klenow's (2009) manufacturing misallocation study compares establishments in China and India with the United States. In their model-based counterfactual, reallocation toward the degree of marginal-product equalization observed in US manufacturing yields calculated TFP gains of 30–50% for China and 40–60% for India. These are hypothetical manufacturing productivity estimates from the study, not observed national GDP gains or guaranteed benefits of a particular reform. Their interpretation depends on the production and demand assumptions, measured plant characteristics, and benchmark dispersion.

The scientific issue is whether observed dispersion reflects distortions or economically meaningful differences in adjustment costs, technology, output quality, demand, residual risk, or capital measurement. A model that assigns all residual wedges to distortions can overstate the feasible scope of reallocation. A model that excludes all wedges can understate the significance of institutional barriers. Comparison across competing specifications and independent establishment evidence is required.

Network position further changes the aggregate importance of producers. An input supplier may have modest value added yet be difficult to replace. In an efficient production economy, first-order productivity accounting attaches weights reflecting economic importance in the production structure, including intermediate inputs. Higher-order effects depend on substitution and changing input shares. Baqaee and Farhi's (2017, revised 2019) nonlinear microeconomic shock aggregation develops this connection beyond a first-order productivity approximation.

Output accounting in networks must avoid counting intermediate transactions as final value added. At the same time, gross output and intermediate sales are informative about production links and shock propagation. The appropriate statistic depends on whether inquiry concerns final production, exposure, supply-chain fragility, or resource reallocation. Household and firm heterogeneity interact when production shocks alter labor incomes or commodity prices unevenly across households.

Distributional state dynamics

A heterogeneous-agent economy has an aggregate state z_t and a cross-sectional distribution μ_t of individual states s_t. Let g_t(s) denote a decision policy and Q_t(ds'|s,g_t(s),z_t) a conditional transition kernel. Then the next distribution is μ_{t+1}(B)=integral Q_t(B|s,g_t(s),z_t) dμ_t(s) for any measurable set B. The aggregate statistic at time t is an integral of the relevant decision or state over μ_t. This representation separates population dynamics from the current measurement.

A shock can influence today's decision policy without immediately changing the distribution of predetermined assets. Subsequent saving and income changes alter future distributions. In a simple discrete representation, a row vector of masses evolves by multiplication by a row-stochastic transition matrix. Because each row sums to one, nonnegative masses remain nonnegative and the total probability remains one. This provides a basic numerical conservation test.

Two distributions with equal means can evolve into distributions with different successor means if transitions depend nonlinearly on their states. For example, a rule that moves an agent toward a borrowing limit when income falls can have a larger aggregate effect in a population already near the limit. Stationary moments before the shock can conceal differences in the conditional distribution of vulnerability.

This is why a one-period calibration and a dynamic validation answer different questions. An approximation that fits current asset moments may fail to reproduce mobility, persistence, or the mass entering constraints during a recession. Longitudinal microdata and simulated transition diagnostics provide a way to test those dynamic implications separately from aggregate forecast errors.

Local response and computational reduction

Computational economics makes it possible to retain richer distributions while studying equilibrium responses. Auclert, Bardóczy, Rognlie, and Straub (2021), in sequence-space Jacobian methods, develop an approach that represents derivatives of aggregate equilibrium paths near a steady state and composes the response blocks. The cited author summary describes applications to a standard growth economy and New Keynesian economies with one or two assets.

For a generic equilibrium mapping F(X,Z)=0, with endogenous path X and exogenous path Z, a differentiable local solution satisfies F_X dX+F_Z dZ=0. If F_X is invertible on the chosen finite-dimensional approximation, dX=−F_X^{-1}F_Z dZ. The equation is an application of implicit differentiation and requires a well-defined equilibrium mapping, appropriate normalization, and a nonsingular derivative. It does not establish unique global equilibrium or accuracy for large nonlinear shocks.

Local derivatives should be checked against finite-difference calculations with multiple perturbation sizes. Distribution grids need tests for truncation, mass conservation, and the convergence of tail-sensitive statistics. Equilibrium residuals need reported tolerances; a solution that nearly clears a capital market may still have material budget violations for a subset of agents. A reproducible computational appendix would specify software version, grid construction, transition rules, parameter values, and perturbation procedures.

The present dossier does not implement or estimate a HANK or production-network model. Its mathematical examples illustrate underlying identities and mechanisms. Claims about computational performance refer to the cited methodological study, not a newly executed benchmark. Detailed algorithms, solver comparison, and replication of large-scale simulations are appropriate to the adjacent computational research dossier.

Identification, calibration, and policy invariance

A fitted distribution of wealth may result from heterogeneous discount rates, earnings processes, credit constraints, bequest motives, portfolio frictions, or other mechanisms. Observational equivalence arises when distinct model parameters yield similar distributions or aggregate moments. To distinguish them, one needs measurements or experiments that generate different predictions, such as consumption responses by liquidity, movements after income losses, and transitions across asset categories.

Calibration assigns parameters to match selected observed targets or externally measured quantities. Estimation describes a statistical procedure for learning parameters from a specified likelihood, moments, or other criterion. A calibrated model can be empirically informative, but calibration success is not equivalent to a unique causal explanation. Model-based quantitative conclusions should report which moments were targeted, which were untargeted validations, and which mechanisms were assumed.

Lucas's (1976) econometric policy evaluation critique supplies the historical motivation for caution when extrapolating behavioral relationships across policy regimes. In a heterogeneous economy, such changes can arise through both revised individual decisions and shifts in the population distribution. For instance, a persistent alteration of credit access can change the mass of households holding small liquid balances. An equation fit before the change can fail afterward even when microeconomic preferences remain constant.

A rigorous evaluation separates estimation uncertainty, misspecification uncertainty, measurement uncertainty, and extrapolation uncertainty. The relevant test is not whether the researcher can fit past aggregate consumption using a sufficiently flexible function, but whether the fitted mechanism correctly predicts responses to informative variation when distributional and institutional changes occur. Where the data do not identify the relevant counterfactual, the result should be reported as scenario-conditional.

Welfare and the distribution of gains

The sum of household consumption is a positive economic quantity. Social welfare requires additional structure. Suppose a social objective is W=E[ω_i u_i(c_i)] with explicitly specified positive weights ω_i and utility functions. Under differentiability, the consumption component of a small change is E[ω_i u_i'(c_i) Δc_i]. Average consumption growth can therefore coexist with losses for a group and with different social welfare rankings under different weights.

Consider two people with consumption 2 and 8, total 10. Under a common logarithmic objective and equal weights, the sum of utilities is log(16). Moving both to 5 yields log(25), an increase even though total consumption is unchanged. Moving from the first distribution to consumption 1 and 10 increases total consumption to 11 while reducing the logarithmic sum to log(10). These constructed examples assume identical logarithmic utility and interpersonal equal weights, with no transition costs. They demonstrate a distinction between accounting and normative evaluation rather than a universal ethical ranking.

Risk, leisure, public services, health, expectations, and intergenerational consequences may also matter. Some policy comparisons can use Pareto dominance if every person's welfare change is nonnegative and at least one is positive. Many real policy changes create winners and losers, requiring a specified social criterion or separate incidence reporting.

A representative utility function that exactly reproduces market demand does not generally recover these distributional comparisons on its own. The science of aggregation can show which information is lost; the choice of normative criterion is a separate explicit commitment. Where no common criterion has been authorized, reporting aggregate effects and distributional consequences separately preserves the distinction between evidence and evaluation.

An integrated analytical comparison

The evidence reviewed here concerns different economic objects. Gorman's demand representation addresses distributional invariance under specified functional conditions. Aiyagari's incomplete-markets analysis studies saving and an endogenous wealth distribution. Krusell and Smith examine the accuracy of a reduced aggregate state. Carroll and coauthors analyze the relationship between wealth targets and spending responses. The monetary HANK and redistribution literature studies policy transmission; production-allocation literature studies the role of producer differences and linkages. These are complementary applications of aggregation rather than different estimates of a common parameter.

A coherent interpretation begins with the aggregate object and the intervention. Let the observed statistic be the expectation of a decision rule under a population distribution. The researcher should establish whether the contemplated intervention changes the rule, weights, distribution, relative prices, or their interactions. Exact aggregation applies if the relevant variation cancels under the assumed structure. Approximation applies when omitted variation contributes little under a stated error criterion. Explicit distributional modeling becomes necessary when omission materially changes the counterfactual or the welfare question.

The author's synthesis emphasizes that the distribution is economically consequential through its relation to the contemplated mechanism. Wealth inequality by itself does not prove a large fiscal multiplier; the relationship between wealth composition, transfer assignment, and spending derivatives is decisive. A representative model that predicts aggregate investment may still fail for redistribution or unemployment; the difference arises from what information each counterfactual requires. Production dispersion may indicate feasible gains only after the analyst establishes the constraints and economic meaning of measured marginal products.

This interpretation has methodological consequences for research design. Evaluate the low-dimensional approximation using a counterfactual-relevant loss function before deciding whether to retain a richer distribution. Validate joint distributions and transitions that affect the mechanism, rather than collecting every observable dimension indiscriminately. Treat normative judgments and accounting identities separately from behavioral approximation. These recommendations organize established analytical results; they are not offered as a new formal theory, named model, or verified claim of scientific novelty.

Testable implications and open research problems

Several research comparisons follow. First, compare a mean-resource prediction with a prediction that incorporates the joint distribution of liquid assets and net transfer receipts. If the latter improves out-of-sample prediction of spending responses to a cash intervention, the evidence supports retention of those distributional features for that task. The comparison requires a common spending horizon, comparable treatment assignment, and controlled financing assumptions.

Second, compare aggregate predictions and group-specific predictions after a monetary innovation using measured contract repricing schedules and labor-income exposures. An appropriate design must distinguish unexpected policy changes from information about future economic conditions. Third, estimate the sensitivity of production reallocation results to alternative markups, adjustment costs, and measures of factor quality. Predictions that depend entirely on one residual-wedge interpretation should receive correspondingly limited causal weight.

Fourth, validate a distributional transition law against observed mobility and changes in constraint exposure. A model can match stationary percentiles while failing to predict which agents deplete buffers after an adverse shock. Fifth, compare household survey statistics and national accounts under explicit concept alignment instead of treating the gaps as errors attributable to one source. Replication should document sampling uncertainty, imputation, weights, and alternative definitions.

The dossier establishes these as testable research questions, not empirical results newly obtained in this study. Further work should include access to complete original texts, reproduction of numerical results from code and data where permitted, and a harmonized cross-country measurement protocol. The selected evidence informs the comparisons but does not identify all global parameters or exhaust the research literature. The most important open issue is to determine which dimensions of state and dependence are sufficient for particular classes of counterfactuals across different institutional environments.

Conclusion

Economic aggregation is an operation whose validity depends on the object represented. A sum of observed economic flows is established by measurement and accounting. An equation that predicts the sum after incentives change requires structural evidence or a validated behavioral approximation. A social evaluation of that change requires a normative specification. A single label of representativeness cannot substitute for these three distinct requirements.

Exact demand aggregation demonstrates the importance of shared slopes and the domain of individual decision rules. In incomplete markets, risk and liquidity alter the distribution of household resources and marginal valuations. The aggregate effects of redistributive shocks depend on the covariance between exposure and responsiveness. Firm heterogeneity and network interactions change the importance of resource allocation, substitutability, and shock location. In dynamic settings, even an informative current mean can become insufficient when heterogeneous transitions change the next distribution.

The proper role of a representative agent is accordingly conditional. It provides a transparent benchmark and can accurately approximate selected aggregates, as illustrated by the stochastic growth literature. A distributional representation is warranted for questions where outcomes depend on resources, constraints, gains and losses assigned to different units, or the interaction of heterogeneous production capacities. Empirical evidence should evaluate that dependence at the horizon and institutional setting relevant to the policy or scientific question.

The research contribution is the integrated interpretation of established results, explicit arithmetic examples, and separation of accounting, behavioral, computational and welfare claims. The comparative literature supplies important support for this synthesis; it does not establish a universal numerical elasticity or a new general theory. The practical scientific conclusion is that aggregation should preserve exactly the economically relevant dependence needed for the intended explanation and make the approximation error and counterfactual domain visible.

References

Aiyagari, S. Rao. 1994. Uninsured Idiosyncratic Risk and Aggregate Saving. Quarterly Journal of Economics 109 (3): 659–684. DOI: 10.2307/2118417. Federal Reserve Bank of Minneapolis Working Paper 502. Research record.

Arrow, Kenneth J., and Gérard Debreu. 1954. Existence of an Equilibrium for a Competitive Economy. Econometrica 22 (3): 265–290. DOI: 10.2307/1907353. Reprinted in Mathematical Economics (1983), 68–97. Research record.

Auclert, Adrien, Bence Bardóczy, Matthew Rognlie, and Ludwig Straub. 2021. Using the Sequence-Space Jacobian to Solve and Estimate Heterogeneous-Agent Models. Econometrica 89 (5): 2375–2408. DOI: 10.3982/ECTA17434. Research record.

Auclert, Adrien. 2019. Monetary Policy and the Redistribution Channel. American Economic Review 109 (6): 2333–2367. DOI: 10.1257/aer.20160137. Research record.

Baqaee, David Rezza, and Emmanuel Farhi. 2017. The Macroeconomic Impact of Microeconomic Shocks: Beyond Hulten's Theorem. NBER Working Paper 23145, revised 2019. Published 2019, Econometrica 87 (4): 1155–1203. DOI: 10.3982/ECTA15202. Research record.

Board of Governors of the Federal Reserve System. 2023. Changes in U.S. Family Finances from 2019 to 2022: Evidence from the Survey of Consumer Finances. October 2023. Research record.

Carroll, Christopher D., Jiri Slacalek, Kiichi Tokuoka, and Matthew N. White. 2017. The Distribution of Wealth and the Marginal Propensity to Consume. Quantitative Economics 8 (3): 977–1020. DOI: 10.3982/QE694. Research record.

Debreu, Gérard. 1974. Excess Demand Functions. Journal of Mathematical Economics 1 (1): 15–21. Research record.

European Central Bank. Household Finance and Consumption Survey – 2021 Wave – Frequently Asked Questions. 2021 survey wave; accessed November 17, 2026. Research record.

Fagereng, Andreas, Martin B. Holm, and Gisle J. Natvik. 2021. MPC Heterogeneity and Household Balance Sheets. American Economic Journal: Macroeconomics 13 (4): 1–54. DOI: 10.1257/mac.20190211. Research record.

Gorman, W. M. 1953. Community Preference Fields. Econometrica 21 (1): 63–80. DOI: 10.2307/1906943. Research record.

Hsieh, Chang-Tai, and Peter J. Klenow. 2009. Misallocation and Manufacturing TFP in China and India. Quarterly Journal of Economics 124 (4): 1403–1448. DOI: 10.1162/qjec.2009.124.4.1403. Research record.

Huggett, Mark. 1993. The Risk-Free Rate in Heterogeneous-Agent Incomplete-Insurance Economies. Journal of Economic Dynamics and Control 17 (5–6): 953–969. Research record.

Kaplan, Greg, Benjamin Moll, and Giovanni L. Violante. 2018. Monetary Policy According to HANK. American Economic Review 108 (3): 697–743. DOI: 10.1257/aer.20160042. Research record.

Kaplan, Greg, Giovanni L. Violante, and Justin Weidner. 2014. The Wealthy Hand-to-Mouth. Brookings Papers on Economic Activity 2014 (1): 77–138. Research record.

Kirman, Alan P. 1992. Whom or What Does the Representative Individual Represent? Journal of Economic Perspectives 6 (2): 117–136. DOI: 10.1257/jep.6.2.117. Research record.

Krusell, Per, and Anthony A. Smith, Jr. 1998. Income and Wealth Heterogeneity in the Macroeconomy. Journal of Political Economy 106 (5): 867–896. DOI: 10.1086/250034. Research record.

Lucas, Robert E., Jr. 1976. Econometric Policy Evaluation: A Critique. Carnegie-Rochester Conference Series on Public Policy 1: 19–46. DOI: 10.1016/S0167-2231(76)98003-6. Research record.

Mantel, Rolf R. 1974. On the Characterization of Aggregate Excess Demand. Journal of Economic Theory 7 (3): 348–353. DOI: 10.1016/0022-0531(74)90100-8. Research record.

Sonnenschein, Hugo. 1972. Market Excess Demand Functions. Econometrica 40 (3): 549–563. Research record.

Unayama, Takashi, Norihiro Komura, and Takahiro Hattori. 2021. Impacts of Cash Transfers on Consumption during the COVID-19 Pandemic: Evidence from Japanese Special Cash Payment. RIETI Discussion Paper 21-J-022, April. Research record.