Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Welfare Identification in Insurance Markets

Abstract. Welfare identification in insurance markets is the empirical and theoretical determination of how an observed insurance allocation compares with feasible counterfactual allocations once willingness to pay, risk-dependent costs, contract incentives, insurer behavior, policy transfers, and the distribution of welfare are explicitly represented. This monograph develops a unified critical reading of the demand-and-cost approach, its extensions to multiple contracts and structural choice models, and its application to selection, mandates, subsidies, market power, choice frictions, and public insurance. A central result is that variation in premiums can identify the demand schedule and the composition of insurer costs under carefully specified conditions, while the interpretation of the resulting area as social surplus requires additional economic assumptions. Adverse and advantageous selection alter the difference between average and marginal cost, but neither a positive correlation test nor the mere presence of a selection gradient establishes the magnitude of welfare loss. The analysis reconstructs marginal-surplus calculations, examines alternative equilibrium concepts, distinguishes positive estimates from normative judgments, and compares evidence from health insurance, annuities, automobile insurance, supplemental coverage, and agricultural index insurance. It concludes that credible welfare estimates are conditional statements about contract quality, consumers' valuation, cost incidence, causal identification, and the institutional environment. The contribution is an original analytical synthesis and source-by-source methodological audit, rather than a new theoretical system or an original econometric estimate.

Keywords: Insurance economics; welfare identification; adverse selection; advantageous selection; demand estimation; marginal cost; average cost; consumer surplus; causal identification; structural econometrics; moral hazard; insurance mandates; risk adjustment; market power; behavioral frictions; distributional welfare; social insurance; external validity.

Introduction

Insurance converts uncertain contingent losses into a schedule of premiums, indemnities, contractual obligations, and claims. This change affects the distribution of household consumption across states of the world, the allocation of risks among institutions, and the price at which individuals can secure protection. Its welfare consequence cannot be inferred from coverage growth alone: a greater number of policies may represent valuable risk sharing, an inefficient expansion caused by subsidized premiums, an improvement in access, or a change in the contract itself. A rigorous evaluation must say whose welfare is measured, which resources are consumed, which payments are transfers, and which alternative allocation is feasible.

The analysis begins with Arrow's welfare economics of medical uncertainty, in which uncertainty, trust, information, and the institutional organization of medical care matter for market performance (Arrow 1963). The economics of hidden information became sharper with the market for lemons (Akerlof 1970) and the competitive insurance equilibrium model (Rothschild and Stiglitz 1976). These works establish reasons why observed contracts can differ from familiar first-best allocations. They do not supply a numerical welfare loss for an arbitrary insurance market; empirical identification is a distinct task.

The modern reduced-form bridge between theory and measurable welfare was developed in Estimating Welfare in Insurance Markets (Einav, Finkelstein, and Cullen 2010). Exogenous variation in insurance prices, combined with enrollment and cost observations, can identify both demand and the marginal composition of insured costs. The broader Beyond Testing review (Einav, Finkelstein, and Levin 2010) explains why determining whether asymmetric information exists is only the first step in a policy evaluation. Welfare requires the entire relevant demand-and-cost structure and an appropriate market counterfactual.

The empirical examples in this Research Dossier span several dimensions. The principal analytical benchmark is voluntary enrollment in a specified insurance contract with a uniform premium. A second dimension is choice among differentiated plans, where a family's valuation depends on covered services, provider networks, cost sharing, and contingent benefits. A third is public provision or mandated participation, where fiscal cost, incidence, and distributional weights become explicit. A fourth dimension concerns behavioral and institutional frictions that interrupt the mapping from observed choice to informed welfare valuation.

The evidence is international in scope, but its identifying datasets are necessarily local: a particular employer's insurance menu, a national annuity submarket, one country's automobile-insurance sample, particular publicly subsidized exchanges, or households receiving specified agricultural-risk interventions. The task is to preserve the internal credibility of each study while describing the conditions under which its insights could travel. Numerical estimates below are attributed to their respective studies and retain their original denominators, baseline conditions, and population boundaries. No new empirical dataset is claimed.

Welfare Measures and Economic Surplus

Welfare objects and units of valuation

Economic welfare is a comparison among feasible allocations under an explicitly stated evaluative criterion. In a conventional private-insurance demand model, the relevant local measure is the money-metric net benefit of access to a specified contract: the insured person's willingness to pay for that contract, less its marginal real resource cost. If individual ii has maximum willingness to pay viv_i and social marginal cost cic_i, the incremental private-plus-resource surplus under quasilinear and otherwise suitable assumptions can be represented by vi−civ_i-c_i. This accounting is intentionally local to the defined contract and population.

Expected utility supplies a more general foundation. Let wealth before the potential loss be ww, loss LL occur with probability π\pi, premium be pp, and indemnity on loss be II. For a fixed contract with state-independent preferences, expected utility is

U(I,p)=(1−π)u(w−p)+πu(w−p−L+I).U(I,p)=(1-\pi)u(w-p)+\pi u(w-p-L+I).

An uninsured option has utility U(0,0)=(1−π)u(w)+πu(w−L)U(0,0)=(1-\pi)u(w)+\pi u(w-L). The willingness-to-pay premium vv for an otherwise specified indemnity solves U(I,v)=U(0,0)U(I,v)=U(0,0), whenever that equation has a unique solution. With increasing, concave uu, risk transfer can have value beyond its expected indemnity. Yet the benefit of the contract can also depend on prevention, use of services, claim settlement, and insurer performance. A certainty equivalent must use the correct underlying states and preferences.

Observed premiums are payments, not automatically resource costs. Premiums transferred from insured persons to insurers may fund claims, reserves, operating expense, taxes, capital remuneration, and profits. Claims paid to a victim can be a transfer while claims-related medical services or repair inputs consume resources. The welfare assessment has to identify the counterfactual: medical treatment would sometimes be funded elsewhere in the absence of the contract, and an indemnity could alter treatment or prevention choices. In social insurance, the taxes supporting coverage and the claims financed by public budgets enter different accounts; financing incidence and fiscal externalities may matter as much as the face value of the insurance benefit.

A useful operational partition therefore records at least four objects: the money-metric valuation of protection, the change in real resource use, the change in transfers and their recipients, and the policy's fiscal or administrative cost. In The Value of Medicaid, Finkelstein, Hendren, and Luttmer (2019) demonstrate why the resource cost of expanding public coverage, the government's total outlay, and transfers to providers of previously uncompensated care must be distinguished. Their estimated welfare valuation is sensitive to the framework for valuing protection; this is a genuine feature of the object being measured rather than a problem fixed by more precise premium data.

Risk aversion, money metrics, and incidence

Risk aversion makes equal expected-dollar contracts different in welfare. A risk-averse household can place positive value on exchanging volatile contingent consumption for a predictable premium even if the expected premium exceeds expected indemnity. The magnitude depends on wealth, liquidity, state-dependent needs, available informal insurance, and the shape of the utility function. Comparing dollar willingness to pay across income groups then requires care: a dollar can command different marginal utility at different levels of consumption, and a poor household's observed ability to pay can understate the social value assigned to protection under redistributive objectives.

The familiar area under a market demand curve represents an appropriate willingness-to-pay surplus when consumers choose rationally among clearly defined goods, the monetary numeraire has the required interpretation, prices are exogenous, and general-equilibrium income effects are sufficiently limited or handled explicitly. Outside those conditions, an area calculation is a descriptive money metric or a component of a broader welfare evaluation. The numerical precision of the area does not repair a mistaken definition of the welfare object.

A full accounting separates ex ante welfare from ex post outcomes. An insured household may experience no claim and still receive valuable risk reduction; a claimant may receive a large payment that is primarily a transfer rather than new social product. Ex ante valuations should also be distinguished from the ex post distribution of health, wealth, and access. The evaluation of insurance as social protection may additionally assign normative value to preventing catastrophic deprivation even where market revealed preferences do not fully reflect that value.

The informational content of observed choices

Purchasing a contract at a given price demonstrates a revealed preference only conditional on what the buyer knows, believes, can afford, and is offered. Liquidity constraints, hidden exclusions, lack of trust, mistaken premium comparisons, and insurer denials can each weaken an uncomplicated welfare interpretation. In the United States, Abaluck and Gruber's Medicare Part D choice study (2011) documents inconsistencies between older consumers' plan choices and the economic attributes of available plans. Handel's analysis of inertia and selection (2013) shows why reducing consumer inertia can improve individual choices while intensifying adverse selection and lowering equilibrium welfare in a particular employer setting.

There is also an elementary distinction between willingness to pay for a promise and willingness to pay for credible indemnification. A policy may offer a nominal payout, but its economic value depends on whether contractual triggers, claim processing, insurer solvency, and legal enforcement make the benefit credible. This issue is especially important for index insurance, where a loss can occur without the index triggering payment. Demand for the actual product is then a joint valuation of risk transfer and basis risk, not a measure of demand for hypothetical full indemnity.

These distinctions lead to a methodological instruction: specify the contract, the decision-maker's feasible choice set, the information environment, and the valuation convention before using observed enrollment to infer economic surplus. The researcher's estimate is an answer to that specified question. Without the specification, the number has no stable welfare meaning.

Demand and Cost Curves

Demand, average costs, and marginal selection costs

Consider a population of NN individuals, each offered one standardized insurance contract at a uniform premium pp. Let viv_i denote the willingness to pay for the contract, and let cic_i be the expected insurer cost associated with that individual's enrollment, conditional on the actual contract and induced behavior. At price pp, the enrollment set is S(p)={i:vi≥p}S(p)=\{i:v_i\geq p\}, and quantity is Q(p)=∣S(p)∣Q(p)=|S(p)|. A downward-sloping inverse demand function PD(q)P_D(q) describes the willingness to pay of the marginal enrollee when quantity rises to qq.

Total expected cost of enrolled persons is TC(q)=∫0qc(s) dsTC(q)=\int_0^q c(s)\,ds, where the integral orders consumers by their enrollment thresholds. Average cost is AC(q)=TC(q)/qAC(q)=TC(q)/q, and marginal cost is MC(q)=dTC(q)/dqMC(q)=dTC(q)/dq. The ordering is economic, not chronological: MC(q)MC(q) represents expected cost of the types who enroll at the margin as coverage expands. This can differ from each incumbent's expected cost and from a technological marginal cost of underwriting one more otherwise identical policy.

If enrollment is sorted by private risk and value, the composition changes with price. Higher prices may select people with higher expected claims, so that expanding quantity lowers average cost. This is the standard adverse-selection pattern. But if particularly cautious, financially sophisticated, or healthy individuals are more inclined to buy, expanding enrollment may increase average cost; this is advantageous selection. The mathematical sign of AC′(q)AC'(q) is informative only after the appropriate ranking and other costs have been defined.

Under differentiability, the accounting identity

MC(q)=AC(q)+q AC′(q)MC(q)=AC(q)+q\,AC'(q)

follows directly from TC(q)=qAC(q)TC(q)=qAC(q). Thus a downward-sloping average-cost curve implies MC(q)<AC(q)MC(q)<AC(q), while upward-sloping average cost implies MC(q)>AC(q)MC(q)>AC(q). This is a compositional implication, not proof that consumers know their own risk: insurer rating practices, behavioral selection, contract quality, and coverage-induced utilization can also affect the measured schedule.

The graphical welfare calculation

Suppose a particular contract is fixed, expected costs are properly measured, individuals' monetary valuations are socially appropriate at the chosen normative benchmark, and a stable ordering of enrollment decisions is meaningful. Total net surplus from coverage qq, omitting fixed terms common to all compared policies, is

W(q)=∫0q[PD(s)−MC(s)] ds.W(q)=\int_0^q [P_D(s)-MC(s)]\,ds.

A social planner able to choose which marginal individuals are covered, subject to the specified contract and cost structure, selects a quantity q∗q^* at an interior optimum satisfying PD(q∗)=MC(q∗)P_D(q^*)=MC(q^*), provided this is a maximum and all feasibility constraints are satisfied. In an idealized pooled-price competitive equilibrium with zero economic profits and a uniform premium, pe=AC(qe)=PD(qe)p^e=AC(q^e)=P_D(q^e). The competitive coverage condition involves average cost rather than the social marginal selection cost. Their difference is the source of the familiar selection distortion.

For adverse selection, if AC′(q)<0AC'(q)<0 in the relevant region, MC(qe)<AC(qe)=PD(qe)MC(q^e)<AC(q^e)=P_D(q^e). The welfare derivative at the equilibrium coverage level is positive: an additional marginal insured person contributes more valuation than expected incremental cost. If feasible, expansion raises aggregate surplus until marginal willingness to pay meets marginal cost. Under advantageous selection the inequality reverses locally; pooled-price equilibrium may induce coverage above the efficient quantity. These are local conclusions under the specified one-contract model; global uniqueness requires additional shape and equilibrium assumptions.

A simple illustrative calculation makes the difference transparent. Assume, only for this invented mathematical example, inverse demand PD(q)=160−qP_D(q)=160-q, marginal cost MC(q)=40+q/2MC(q)=40+q/2, quantity measured in standardized policies, and total cost TC(q)=40q+q2/4TC(q)=40q+q^2/4. Then AC(q)=40+q/4AC(q)=40+q/4. Here average cost rises with coverage, indicating advantageous selection by the ordering defined above. The efficient condition gives 160−q=40+q/2160-q=40+q/2, or q∗=80q^*=80. The pooled break-even condition gives 160−q=40+q/4160-q=40+q/4, or qe=96q^e=96. The excess quantity of 16 has a total surplus loss ∫8096(MC(q)−PD(q))dq=192\int_{80}^{96}(MC(q)-P_D(q))dq=192 in invented monetary units. This example is an algebraic illustration, not an estimate for any country or insurer.

In adverse-selection applications the geometric configuration is reversed; its empirical area depends on the observed slopes and the relevant counterfactual quantity. Under suitable assumptions, Einav, Finkelstein, and Cullen (2010) estimate these ingredients in employer health insurance and report a relatively small study-specific welfare loss despite detecting adverse selection. The result demonstrates that the presence of asymmetric sorting and the size of the associated welfare distortion are separate empirical questions.

Identification from price variation

Prices can identify demand if their variation moves enrollment decisions without simultaneously altering unobserved determinants of demand, product value, or the population under study in uncontrolled ways. Let ZZ be an instrument or assignment causing premium variation and satisfying the needed exclusion and independence conditions. Observing p(Z)p(Z) and Q(Z)Q(Z) can trace a segment of demand. If claims and insurer expenses are observed for the corresponding enrolled sets, the researcher can estimate AC(Q(Z))AC(Q(Z)), and from TC(Q)=QAC(Q)TC(Q)=Q AC(Q) infer marginal selection costs where there is sufficient support and differentiability.

The identification argument is strongest when contract terms, choice sets, and the distribution of potential enrollees remain stable as prices change. A policy that both lowers premiums and expands covered services alters valuation and expenditure simultaneously. A subsidy that changes family income can produce additional responses beyond the price effect. A tax-financed intervention may change providers' incentives or insurer bargaining, violating the notion of a fixed cost schedule. Researchers must identify the policy-generated price movement and explain why it can be treated as exogenous to the outcomes being modeled.

An average-cost schedule is not an individual medical-risk regression. Observed mean cost at different enrollment rates captures the selected mix of enrolled persons, given coverage-induced utilization. To interpret it as a counterfactual cost curve, the researcher must justify that the potential costs of the marginal types would be the same under the proposed counterfactual contract and market environment. The derivative of a smooth fitted curve can magnify sampling error, so cost and demand slope uncertainty should flow through to the welfare calculation.

Observed equilibrium prices, by themselves, may be endogenous. Insurers set prices in response to expected claims, competitor behavior, administration, and regulation; buyers respond to both prices and plan attributes. Regressing enrollment on such prices without a valid strategy generally confounds demand with supply and omitted quality. Natural experiments, randomized premium variation, and credible cost-side shifts can address specific dimensions of this problem; none is a universal solution.

Extensions beyond a single binary contract

Real insurance choices often involve plans rather than insured/uninsured status. With multiple plans jj, household ii chooses a contract according to Uij(pj,xj,εij)U_{ij}(p_j,\mathbf{x}_j,\varepsilon_{ij}), while insurer cost is cijc_{ij}. Each price change can divert consumers among plans, change the risk composition of several pools, and change expenditure conditional on coverage. The welfare effect must account for the joint price vector, relative quality, and substitution across contracts. A single aggregate demand curve can obscure an economically material movement from high to low protection even when the total number of insured people remains fixed.

Bundorf, Levin, and Mahoney (2012), in Pricing and Welfare in Health Plan Choice, model small-employer plan selection and estimate welfare losses relative to a risk-rated feasible benchmark. Their analysis explicitly shows that a market's contribution or premium structure can generate inefficient sorting even where most consumers remain insured. Estimated welfare losses in their setting are 2–11 percent of coverage costs; the denominator, institutional setting, and benchmark matter for interpreting that range.

More generally, a choice of deductibles can have different welfare implications from a choice of insurers. If one contract causes greater use of medical services or produces a different probability of claim reporting, cost is endogenous to contract selection. The one-dimensional formula remains valuable as a transparent benchmark, but multi-contract welfare usually requires a model of jointly determined choices, incentives, and insurer equilibrium.

Adverse and Advantageous Selection

Separating selection, incentives, and preferences

Adverse selection, in its common insurance-market sense, means that buyers selecting greater insurance coverage have higher expected insurer costs than less-insured buyers, conditional on relevant prices and contracts. The category describes the relationship between insurance demand and expected costs; its interpretation as private information depends on the information set and insurer pricing rules. An insurer may know a characteristic correlated with loss risk but be prohibited from using it in premiums. Costly information acquisition and omitted risk characteristics can also reproduce the financial consequences of a private-information selection problem. A positive demand–risk correlation is therefore a clue about sorting, not an exhaustive account of why sorting occurs.

Advantageous selection means that people who choose insurance are, on the relevant risk-cost margin, less costly than those who decline it. This can arise when prudent people take preventive measures and purchase greater coverage, when cognitive skills influence both purchase and health care, or when differences in access and liquidity affect who enters the market. The adverse/advantageous classification is relative to the contract and the econometric conditioning set. A consumer can be adversely selected for one coverage option and advantageously selected for another.

The classic Rothschild–Stiglitz framework establishes how private risk information can induce screening by contract design, potentially yielding separating allocations, cross-subsidy problems, and nonexistence of some competitive equilibria; see their insurance-market equilibrium essay. That framework does not imply that all observed markets have adverse selection, nor that the empirical welfare loss is always large. It describes incentives under explicit assumptions about types, contracts, insurance provision, and competition. Applying its conclusions requires checking whether those assumptions are credible in the market at issue.

Preference heterogeneity creates a further identification difficulty. People more afraid of risk may buy more insurance while simultaneously taking precautions that reduce the probability of loss. Some individuals anticipate higher consumption of insured services; others place greater value on liquidity, convenience, or specific provider networks. The selection gradient is therefore the reduced-form outcome of multiple potentially countervailing channels. Even when cost selection is measured precisely, it does not identify the entire joint distribution of private risk, risk aversion, medical needs, and beliefs without further information or assumptions.

Evidence from different insurance lines

In their French automobile insurance test, Chiappori and Salanié (2000) do not find evidence of asymmetric information in their studied sample. The null result is local to the contracts, accident outcomes, and covariates available to the investigators. It neither proves informational symmetry in all French motor insurance nor negates the mechanisms described by asymmetric-information theory. Its substantive importance lies in showing that an empirical diagnosis must be supported by market-specific observation.

An opposing pattern appears in the U.S. supplemental health insurance literature. Fang, Keane, and Silverman (2008) find advantageous selection in Medigap and investigate sources including socioeconomic characteristics, financial planning, and cognitive ability. Their central result complicates the inference that individuals with stronger protection invariably represent higher expected claims. Moreover, controlling for health changes the descriptive relationship between enrollment and expenditure, demonstrating why the conditioning set is indispensable to substantive interpretation.

Annuities require a different orientation of risk. A life annuity pays for as long as the annuitant lives, so individuals with greater expected longevity may be costlier to the annuity provider and may place higher value on the payout stream. Contract riders and guaranteed periods add additional dimensions of choice. Einav, Finkelstein, and Schrimpf's U.K. annuity welfare analysis (2010) estimates a model of guarantee-period selection with private mortality information and heterogeneous preferences. Their estimate of approximately £127 million annually, or about 2 percent of annuitized wealth, is a structural estimate of distortion along the studied guarantee margin under their maintained model. It is not a percentage applicable to every life or retirement-insurance market.

These comparisons support a disciplined conclusion: the unit of welfare analysis is the insured margin under examination. It matters whether the margin is entering insurance at all, choosing a higher deductible, adding a contract guarantee, switching to a richer provider network, or selecting a more comprehensive indemnity. Each margin can reorder consumers by valuation and cost, and the same population may display different sorting along different margins.

Multiple contracts and endogenous utilization

Ex ante selection is not the only source of cost variation. Insurance changes the price of care, the incentives to prevent loss, the willingness to report claims, and the propensity to undertake risky activity. These responses are called moral hazard in the standard economic literature, without implying moral wrongdoing. In health insurance, a lower effective price at the point of service can increase utilization; part of this utilization may be valuable, and part may have a resource cost above consumers' marginal valuation under the policy objective being used.

Einav, Finkelstein, Ryan, Schrimpf, and Cullen (2013) document selection on moral hazard in employer health insurance: people differ in their utilization response to insurance and can select plans partly on the basis of that anticipated response. Their findings illustrate why holding a type's cost fixed across counterfactual contracts can be incorrect. Expanding insurance can change both who is covered and how much each covered person consumes. These are distinct margins that may interact.

Suppose ci(j)c_i(j) is the expected insurer expenditure for person ii under plan jj, and vi(j)v_i(j) is that person's money-metric value of plan jj. A change from j=0j=0 to j=1j=1 changes both valuation and expected expenditure. It is generally invalid to calculate the welfare of the richer plan using the average claim cost observed under the leaner plan. The needed counterfactual is ci(1)c_i(1), including the behavioral response to the richer coverage, not merely the cost of the same individuals under ci(0)c_i(0).

One useful empirical distinction is between pure selection, treatment effects of coverage, and their interaction. Price or plan-choice variation can identify a composition effect when it changes membership among otherwise comparable contracts; randomized plan assignment or plausible quasi-experimental changes may identify utilization responses; a structural model may be needed to combine these margins where assignment is endogenous. Any welfare calculation that combines estimates from separate designs must explain why the populations and states to which they refer are compatible.

The pricing consequence of selection is also modified by the institutional form of the insurer. Competitive mutual insurers, profit-seeking insurers, employment-based risk pools, public insurers, and compulsory annuity arrangements face different pricing incentives and constraints. A useful welfare model states whether premium revenue must break even at the policy or portfolio level, whether cross-subsidies are allowed, and who bears losses from unexpected claims. Without these commitments, an apparent average-cost price is an assumption rather than an observed equilibrium condition.

Policy Equilibrium Counterfactuals

Competitive benchmarks, subsidies, and mandates

Insurance policy can alter market outcomes through mandates, subsidies, risk classification, standardized benefits, risk adjustment, guaranteed issue, and insurer supervision. The welfare question is counterfactual: what equilibrium, including premiums, enrollment, insurer costs, and any relevant fiscal effects, would arise under a specified intervention? The pre-intervention allocation is one comparison point, but the efficient frontier may be unattainable because information, budgets, contract observability, and political constraints restrict the feasible set.

Under the single-contract adverse-selection benchmark, a subsidy that reduces the consumer-paid premium can expand enrollment toward the efficient quantity. The premium received by insurers and the premium paid by households differ by the subsidy. The public cost of this difference must be financed and its distribution accounted for. Where marginal social benefits exceed marginal real costs, expanding coverage is beneficial before financing costs; the final net welfare effect depends on the cost of public funds, any induced utilization, and whether the subsidy changes insurer markups.

A mandate can change the enrollment set directly and may change average claims cost by bringing lower-risk individuals into the pool. This mechanism is empirically relevant to Massachusetts health-insurance reform. Hackmann, Kolstad, and Kowalski (2015), in Adverse Selection and an Individual Mandate, estimate an annual welfare gain attributed to reducing adverse selection in the studied Massachusetts individual market, reporting approximately $51.1 million annually or 4.1 percent per person under their model and data. That figure describes a particular reform and market, alongside changed premiums, average costs, and markups; it is not a generic return on compulsory insurance.

A different caution emerges from the U.K. annuity application. The optimal-guarantee mandate analysis by Einav, Finkelstein, and Schrimpf (2010) finds that different required guarantee periods could yield very different welfare effects. Their reported counterfactuals include substantial gains from one mandate and losses from another. Eliminating contract choice is therefore not automatically equivalent to creating the first best. The policy designer needs an estimate of the joint distribution of risk and preferences and must describe how insurer prices would respond to the mandate.

Subsidies similarly require identification of who enrolls at the margin. Finkelstein, Hendren, and Shepard (2019) exploit Massachusetts subsidy discontinuities among lower-income adults to estimate the relation among premium contributions, take-up, and costs. In that setting take-up is highly price sensitive, and marginal consumers tend to be less costly than already insured individuals. Crucially, they find low willingness to pay relative to insurer costs in the portion of the distribution they identify. A regulator deciding on subsidies must distinguish the observed private valuation, access to care, risk protection, fiscal transfers, and normative weight assigned to low-income households.

Competition, insurer pricing, and risk adjustment

Competitive pricing is not always literal actuarial break-even on each contract. Insurers may possess market power, differentiate products, incur switching costs, and pursue customers with different risk profiles. With imperfect competition, the observed premium may exceed marginal or average cost because of markups. Demand-and-cost estimates can then describe the allocation but need a model of insurer pricing to predict how premiums change under an intervention.

Mahoney and Weyl (2017), in Imperfect Competition in Selection Markets, study the interaction between selection and market power. Their analysis emphasizes that some conventional intuitions about competition and efficient allocation change when firms face endogenous cost composition. The relevant welfare gradient depends jointly on the demand slope, insurers' markup decisions, and how marginal costs vary with the selected population. Treating market concentration and adverse selection as two independent harms can double-count mechanisms or miss their interaction.

Starc (2014) investigates insurer pricing in Medigap, illustrating the importance of pricing behavior and differentiated supplemental contracts for consumer welfare. Demand elasticities alone do not reveal the social benefit of reducing premiums if price changes transfer surplus between consumers and insurers, change enrollment composition, and induce different equilibrium markups. A complete analysis must identify whether the proposed intervention affects productive efficiency, risk allocation, or the division of rents.

Risk adjustment transfers funds among insurers according to the risk characteristics of their enrollees. It can reduce incentives to avoid high-cost consumers, but the effect on welfare depends on the adequacy of the risk score, the compensated cost dimension, insurer quality choices, and the competition model. An adjustment that insulates insurers from genuine marginal cost differences may weaken incentives to control costs. An adjustment that poorly predicts costs can create new distortions in recruitment or service provision. It follows that a technically accurate predictive score is only an input into the welfare calculation; its effect on prices, coverage, and services remains to be identified.

Public coverage and the incidence of transfers

Public insurance extends the welfare question to government financing, implicit insurance, and redistribution. A government can finance coverage that markets would not supply at affordable premiums, while also shifting resources among taxpayers, beneficiaries, health-care providers, and other institutions. A program's outlays are therefore neither a complete measure of its real resource costs nor a complete measure of the social value it creates.

In the Oregon Medicaid analysis, Finkelstein, Hendren, and Luttmer (2019) estimate beneficiary value and transfer incidence using alternative frameworks applied to randomized coverage expansion. Their published analysis places willingness to pay for Medicaid between approximately $0.50 and $1.20 per dollar of the resource cost of providing coverage, depending on the approach, and identifies a substantial transfer to providers of previously uncompensated care. These are particularly useful distinctions because they separate the program's insurance value, expected resource expenditure, and the recipients of transfers, while acknowledging that welfare under a public program depends on more than market purchase decisions.

The social evaluation may attach distributional weights ωi\omega_i to monetary changes experienced by different individuals. A stylized welfare objective is ΔSW=∑iωiΔCVi−ΔR+ΔE\Delta SW=\sum_i\omega_i\Delta CV_i-\Delta R+\Delta E, where ΔCVi\Delta CV_i is an appropriate compensating-variation measure of household benefit, ΔR\Delta R denotes additional real resources and administrative costs not already included in the valuations, and ΔE\Delta E collects separately identified externalities. This is an accounting representation, not an empirical specification: if the household valuations already net out some costs or transfers, the corresponding component must not be subtracted again. Government revenues, financing distortions, and redistribution require consistent incidence treatment.

When the analyst adopts equal monetary weights, a dollar transferred between two individuals may cancel in aggregate surplus apart from administrative and incentive effects. Under unequal welfare weights the transfer itself changes social welfare. Neither convention is a neutral universal answer; each states a specific objective. This is why empirical findings about willingness to pay and fiscal cost should be presented separately from the recommendation to adopt a subsidy or public mandate.

International and institutional comparisons

The dominant empirical welfare literature provides strong methods but uneven geographic coverage. U.S. employer-sponsored health insurance offers detailed data on plan choice and claims, yet its bargaining arrangements and tax treatment are specific. The historical U.K. annuity guarantee market permits structural evaluation of mortality-related contract sorting, but guarantees are an unusual margin compared with purchasing an annuity in the first place. French automobile insurance permits a test of contract–accident correlation, rather than a complete welfare estimate for all motor risks. These studies are best read together as evidence about distinct identification problems.

Rainfall index insurance in India supplies a different lesson. Cole, Giné, Tobacman, Topalova, Townsend, and Vickery (2013), in Barriers to Household Risk Management, use randomized field experiments to examine take-up of rainfall insurance. They find that price sensitivity coexists with constraints involving trust, liquidity, and limited salience. Observing low take-up therefore cannot be read simply as evidence that risk reduction has low underlying social value. The relevant insurance product also differs materially from indemnity health coverage because payouts depend on an index.

Experiments in northern Ghana further demonstrate that insurance may affect productive choices before a loss occurs. Karlan, Osei, Osei-Akoto, and Udry (2014), in Agricultural Decisions after Relaxing Constraints, report increased investment and riskier agricultural production following insurance provision in the studied setting. A welfare accounting limited to premiums, payouts, and claim administration would overlook changes in production possibilities and investment. Yet the estimated production response cannot be transferred mechanically to other crops, countries, weather indices, or institutions.

Taken together, the international comparison shows that the welfare consequences of coverage are mediated by market design and by the activities coverage makes possible. A fixed price–cost diagram remains useful when it captures the principal margin. A larger structural or experimental evaluation is required when the intervention changes productivity, contract credibility, access to medical care, or the opportunity set itself. The scale of the modeling exercise should follow the causal channel rather than the desire for a single universal welfare statistic.

Identification and Endogeneity

Causal designs and exclusion assumptions

Welfare evaluation in an observational insurance market faces a two-stage inferential problem. First, the researcher must identify responses of demand, costs, and sometimes service use to a defined variation in insurance terms. Second, those responses must be combined into a counterfactual allocation using explicit market and welfare assumptions. The first stage is econometric; the second is both economic and normative. An internally credible causal effect on enrollment is necessary but insufficient for most broad social-surplus conclusions.

Price instruments require an exclusion argument. A variable that changes the employee contribution for a plan but also changes benefits, household wealth, employer communications, or competing options cannot generally be interpreted as a pure price shifter. Likewise, cross-regional price variation can correlate with disease burden, insurer bargaining, regulation, or supply networks. Relevant instruments must shift the margin of interest, be independent of the relevant unobservables under a defensible design, and have sufficient first-stage strength. Monotonicity or a specific behavioral model may be needed to interpret local treatment effects among induced switchers.

Randomization can strengthen causal interpretation, but its scope remains specific. A randomized premium discount identifies a demand response to that discount under the intervention's information and payment environment. It may not identify demand for a permanent universal subsidy if consumers infer different product quality from a temporary discount or if insurers would reprice when the subsidy becomes universal. Randomized availability or assignment to insurance can identify the effect of having coverage on outcomes for eligible participants or compliers, but a structural welfare calculation still needs a valuation of those outcomes, the cost of the intervention, and the equilibrium responses outside the experiment.

Regression discontinuity studies often exploit eligibility thresholds or subsidy schedules. Their comparative advantage is credible local variation around the threshold; their immediate limitation is local identification. The Massachusetts subsidy design used by Finkelstein, Hendren, and Shepard (2019) provides a rigorous illustration of identifying enrollment and cost relationships from discontinuities while interpreting those relationships within the income and institutional range actually observed. Extrapolation to higher-income families or different insurance products requires an additional argument.

Difference-in-differences designs need appropriate assumptions about untreated trends and the policy's timing, composition, and spillovers. Mandates can alter not only take-up but plan prices, insurer strategies, and the medical care environment. An identification strategy designed to capture changes in premiums and participation must be accompanied by a counterfactual model translating those changes into selection-related welfare. Hackmann, Kolstad, and Kowalski (2015) illustrate such a combination of reform evidence and selection-based welfare modeling.

Structural identification and equilibrium feedback

A structural model specifies agents' preferences, informational environment, cost formation, and market rules so that observed behavior can identify economically interpretable parameters. In insurance, a typical model includes heterogeneous expected losses, risk preferences, choice shocks, plan benefits, and the supply-side pricing conditions. A structural approach permits counterfactuals beyond the direct support of observed price variation, but those extensions inherit the maintained assumptions about functional forms, risk distributions, decision rules, and equilibrium selection.

Identification of the joint distribution of risk and preferences is particularly demanding. Demand can be highly responsive to premiums without revealing whether marginal nonbuyers are healthier, more liquidity constrained, less risk averse, or simply inattentive. Claims data from enrolled participants reveal conditional expenditure among insured people; they may not reveal counterfactual spending or risk preferences for the uninsured. Researchers use panel information, health-risk predictors, contract menus, randomized premiums, and restrictions on utility or risk types to supply missing variation. The defensibility of the restrictions should be evaluated separately from model fit.

The U.K. annuity structural study by Einav, Finkelstein, and Schrimpf (2010) provides a strong example: the investigators specify how annuity guarantee preferences and mortality risk jointly govern selection, then simulate alternative guarantee mandates. Counterfactual welfare is meaningful within the estimated model; the study also reports sensitivity of welfare magnitudes to assumptions. This is stronger scholarship than projecting a simple correlation onto an unmodeled government policy, precisely because it distinguishes observed moments from the counterfactual structure that interprets them.

A market-level equilibrium model must account for feedback. If a subsidy attracts new low-cost enrollees, average claims cost can decrease and competitive premiums can decline further; if it induces greater use, marginal expenditure can rise; if it changes insurer market power, markups may fall or rise. Any of these changes can feed back into enrollment. The relevant equilibrium maps a policy parameter τ\tau into a vector (p(τ),Q(τ),C(τ),x(τ))(\mathbf p(\tau),\mathbf Q(\tau),\mathbf C(\tau),\mathbf x(\tau)) of prices, coverage, costs, and contract qualities. Holding all but the first quantity fixed implicitly imposes a partial-equilibrium assumption.

Market definition is itself a substantive issue. A market could mean an employer's employee-benefit menu, a national supplemental-insurance segment, a regional regulated exchange, or a set of substitute public and private coverage arrangements. Consumers may also substitute toward savings, family transfers, informal insurance, credit, or self-protection. Excluding important substitutes can overstate the value of a contract relative to the correct no-purchase alternative, while ignoring spillovers can understate social benefits.

Measurement, uncertainty, and robustness

The data needed for strong welfare identification include premiums actually paid by consumers, full contractual benefit rules, enrollment and switching outcomes, ex ante characteristics, claims and administrative costs, contract-induced utilization, relevant insurer revenues, and the institutional rules that govern pricing. Missing values, delayed claims development, changes in provider coding, risk-score manipulation, lapse behavior, and disputes over payment can each affect the estimated marginal cost curve. A credible analysis should state whether the cost measure is billed charges, paid claims, allowed amounts, insurer accounting expenditure, or social real-resource cost.

Cost measurement also differs by insurance line. For life annuities the relevant expected cost is a stream of contingent payments with discounting and mortality assumptions. For property insurance it includes event frequencies, repair costs, catastrophe exposure, reinsurance and capital charges. For medical insurance it can include treatment quantities and prices induced by insured status. These costs cannot be pooled into a generic risk score without an explicit economic aggregation rule.

Demand estimation brings its own uncertainty. Discrete contracts generate step functions in small samples; a smooth demand curve is an approximation whose derivatives depend on specification. Local price variation can leave the tails of demand poorly identified. Variation in actuarial costs may only reveal average claims for those who purchase at observed prices, leaving counterfactual costs of never-enrolled groups unobserved. Sharp counterfactual welfare estimates in such settings require explicit restrictions; otherwise bounds may be the appropriate scientific product.

Uncertainty should be transmitted to welfare estimates rather than confined to coefficient tables. The object W(q)W(q) is an integral of two estimated functions, and equilibrium qeq^e is itself estimated. Confidence intervals or sensitivity regions must reflect uncertainty in slopes, intersection points, predicted utilization, and counterfactual premiums. A study can establish with high confidence that selection exists while leaving the magnitude of welfare loss uncertain. Another can bound the sign of a particular mandate while retaining a wide interval for its magnitude. These are scientifically meaningful outcomes.

One further distinction concerns prediction and identification. A sophisticated machine-learning model may predict claims accurately while failing to identify the causal effect of a subsidy or changed deductible. A predictive risk score answers who is expected to cost more under the observed environment. A welfare counterfactual asks what the same individuals and institutions would do under a changed environment. Separating those questions is especially important when predictive variables would themselves change in response to incentives, regulation, or coverage status.

Robustness and Distributional Effects

Partial identification and sensitivity analysis

A welfare estimate is robust when the substantive conclusion survives defensible changes in data treatment, model structure, and evaluative assumptions. The first robustness question concerns the price support: does the observed variation identify the margin on which the counterfactual policy acts? If a large mandatory expansion moves participation beyond the observed range, the additional cost and willingness-to-pay curves are extrapolated. Results should distinguish data-supported intervals from model-extended regions.

The second concerns the ranking of marginal consumers. Average cost can fall with increased enrollment because healthier people enter, but it can also fall because coverage terms change, providers alter prices, or claim measurements differ. An appropriate audit tests whether estimated selection patterns persist with consistent benefits, expenditure measures, demographics, time periods, and counterfactual treatments. The unused-observables approach of Finkelstein and Poterba (2014), in Testing for Asymmetric Information Using Unused Observables, illustrates an alternative way to probe insurer pricing information in the historical U.K. annuity setting, distinct from simply correlating coverage with outcomes.

The third concerns behavioral welfare. If consumers misunderstand contract attributes, a demand curve based on actual choice is not necessarily equivalent to a normatively appropriate valuation curve. One approach estimates corrected or informed choices under a model, but that correction must itself be identified and its uncertainty acknowledged. The Medicare Part D evidence (Abaluck and Gruber 2011) demonstrates a concrete problem for unqualified revealed-preference calculations. Moreover, an intervention improving choice quality can change market composition in ways that offset direct consumer gains, as Handel's inertia study (2013) emphasizes.

The fourth concerns market structure. The assumed zero-profit competitive pooled equilibrium should be replaced or compared with a differentiated-oligopoly pricing model when insurers possess market power. The selection-market competition framework (Mahoney and Weyl 2017) provides a reason to evaluate selection and pricing jointly. A subsidy calculated under price-taking competition may have a different incidence and coverage effect when insurers absorb part of it as higher margins.

The fifth concerns the monetary object. An investigator may alternately report private compensating variation, total insurer expenditure, real resource costs, public budget costs, and a distributionally weighted social objective. These outputs should appear separately before any aggregation. For public coverage, one may find low private willingness to pay relative to total government spending and still conclude that provision yields important social gains when the analysis assigns value to redistribution, financial protection, or previously uncompensated care. The methodological obligation is to make that valuation explicit.

For each central claim, this dossier distinguishes three epistemic levels. A verified study result is a statement about the published design, data, and estimate. A model implication is a statement that follows from expressly listed assumptions. An original synthesis is an interpretive conclusion comparing those two levels across cases. None should be presented as though it were a universally estimated fact. This separation is central to the research's reliability.

Distributional weights and consumer protection

Aggregate surplus may obscure who wins and loses. An intervention that increases total economic surplus can raise premiums for some high-risk households, lower protection for vulnerable groups, or transfer rents to insurers. Conversely, a cross-subsidy can lower aggregate money-metric efficiency while serving an explicit redistributive purpose. A complete distributional evaluation therefore reports, where the data permit, changes by risk type, income, age, health status, exposure, and access to alternative protection.

Risk classification illustrates the tension. Premiums based on accurate risk estimates can improve marginal price signals and reduce inefficient sorting across plans, but they may expose individuals to reclassification risk when health or exposure changes. Bundorf, Levin, and Mahoney's health-plan pricing evaluation (2012) considers this issue while comparing uniform contributions and more differentiated pricing. Efficiency gains from risk rating must therefore be assessed alongside the value of protecting individuals from unpredictable future changes in their risk classification.

Consumer welfare also depends on the credibility and accessibility of contractual promises. High actuarial payout ratios may be insufficient for inclusion when policyholders cannot verify the index, trust the claims-paying institution, pay the premium at the available time, or understand the exclusions. The Indian rainfall insurance experiments and the Ghanaian insurance and agricultural investment experiments demonstrate that trust, liquidity, basis risk, and production responses can materially shape observed demand and outcomes. For such products, measuring insurance welfare solely as expected claims minus premiums would omit central elements of the choice.

Where empirical data support subgroup effects, an aggregate weighted welfare expression can clarify incidence. It is often useful to calculate an unweighted monetary-surplus benchmark, subgroup money metrics, and one or more transparent distributional-weight scenarios. No set of weights should be silently substituted for an empirical result. The social choice belongs to the stated criterion; econometrics determines the consequences conditional on that criterion and its assumptions.

Integrated evidence and research priorities

The reviewed literature supports a set of connected findings. Price variation can reveal welfare-relevant demand and selection-cost curves, but the credibility of those curves depends on the source of price variation and on the stability of contract and cost conditions. Selection can be adverse or advantageous; it can differ across contracts and coexist with heterogeneous preferences or moral-hazard responses. Competitive pooled pricing may diverge from marginal-surplus efficiency, but the magnitude of that divergence must be estimated rather than inferred from the existence of asymmetric information.

The policy evidence further shows that mandates and subsidies cannot be evaluated by enrollment changes alone. In certain settings they improve the insured mix and reduce selection-related welfare loss; in others the choice of mandated benefit or degree of subsidy can worsen economic welfare. Insurer competition, fiscal transfers, and imperfect consumer decisions introduce additional margins. Structural estimates allow a richer counterfactual, while requiring stronger assumptions; randomized and quasi-experimental designs strengthen causal credibility on observed margins, while sometimes providing less complete welfare information outside those margins.

The principal research priorities follow directly from these limitations. More transparent identification of contract-specific marginal costs would sharpen welfare-area estimates, especially in markets with multiple plan characteristics. Better linkage among price experiments, administrative claims, and credible information interventions would distinguish private valuations from choice errors and liquidity constraints. Joint models of selection, moral hazard, and insurer competition would reduce the risk of analyzing interacting mechanisms in isolation. Research on emerging and inclusive insurance should measure trust, basis risk, payment reliability, productive investment, and risk protection, while respecting the institutional scope of each field setting.

There is also a substantive need for policy evaluations that report real resource cost, budget expenditure, provider and insurer transfers, and household money-metric welfare as different objects. Such reporting would permit informed public comparison across regulated private markets and social insurance arrangements without conflating a higher fiscal outlay with a higher social resource cost. Wider use of subgroup welfare and sensitivity bounds would make the distributive implications and estimation uncertainty visible rather than relegating them to incidental qualifications.

Finally, reproducibility requires evidence provenance. Every substantive empirical claim must remain connected to a verifiable study, edition, population, and identification strategy. Every equation must state the economic assumptions under which its conclusion follows. Every counterfactual must specify which elements of market behavior are permitted to change. These are standards of analytical clarity, not the creation of a new theory.

Conclusion

Welfare identification in insurance markets is the disciplined reconstruction of the value and cost of insurance under observed and counterfactual arrangements. It begins with a clearly defined protection contract and a welfare criterion, estimates the relevant enrollment and cost relationships, and asks how equilibrium and policy alter those relationships. The demand–marginal-cost integral provides a transparent central calculation where its assumptions hold. The gap between average and marginal costs explains the direction of local selection distortions in the one-contract pooled-price benchmark. Neither the presence of asymmetric information nor the sign of a selection gradient determines the magnitude of welfare loss without identifying the curves and the relevant counterfactual equilibrium.

The empirical record establishes that insurance markets display heterogeneous patterns. Adverse selection can be present with limited estimated welfare loss; advantageous selection can arise from behavioral and socioeconomic sorting; a test can fail to detect asymmetric information in a particular market; mandates can improve or worsen welfare depending on their design; and subsidies can change both participation and public expenditure incidence. Imperfect competition, moral hazard, informed choice, trust, and contract performance alter the mapping between enrollment and social value. International applications demonstrate the importance of comparing institutions on analytically equivalent margins rather than importing local numerical estimates as global parameters.

The general scholarly conclusion is precise: a credible insurance welfare estimate is a conditional, evidence-supported comparison of feasible allocations, whose strength depends on the identification strategy, the economic model, the normative criterion, and the quality of its source record. This dossier provides an original comparative synthesis of those requirements and their consequences. It does not claim an original empirical dataset, independent replication of the reviewed estimates, a new named economic theory, or external peer review. The work is a research manuscript for archival editorial intake; any eventual first-publication date and production status are separate release facts.

References

  • Abaluck, Jason, and Jonathan Gruber. 2011. “Choice Inconsistencies among the Elderly: Evidence from Plan Choice in the Medicare Part D Program.” American Economic Review 101 (4): 1180–1210. https://doi.org/10.1257/aer.101.4.1180.

  • Akerlof, George A. 1970. “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism.” The Quarterly Journal of Economics 84 (3): 488–500. https://doi.org/10.2307/1879431.

  • Arrow, Kenneth J. 1963. “Uncertainty and the Welfare Economics of Medical Care.” American Economic Review 53 (5): 941–973. https://assets.aeaweb.org/asset-server/files/9442.pdf.

  • Bundorf, M. Kate, Jonathan Levin, and Neale Mahoney. 2012. “Pricing and Welfare in Health Plan Choice.” American Economic Review 102 (7): 3214–3248. https://doi.org/10.1257/aer.102.7.3214.

  • Chiappori, Pierre-André, and Bernard Salanié. 2000. “Testing for Asymmetric Information in Insurance Markets.” Journal of Political Economy 108 (1): 56–78. https://doi.org/10.1086/262111.

  • Cole, Shawn, Xavier Giné, Jeremy Tobacman, Petia Topalova, Robert Townsend, and James Vickery. 2013. “Barriers to Household Risk Management: Evidence from India.” American Economic Journal: Applied Economics 5 (1): 104–135. https://doi.org/10.1257/app.5.1.104.

  • Einav, Liran, Amy Finkelstein, and Mark R. Cullen. 2010. “Estimating Welfare in Insurance Markets Using Variation in Prices.” The Quarterly Journal of Economics 125 (3): 877–921. https://doi.org/10.1162/qjec.2010.125.3.877.

  • Einav, Liran, Amy Finkelstein, and Jonathan Levin. 2010. “Beyond Testing: Empirical Models of Insurance Markets.” Annual Review of Economics 2: 311–336. https://doi.org/10.1146/annurev.economics.050708.143254.

  • Einav, Liran, Amy Finkelstein, Stephen P. Ryan, Paul Schrimpf, and Mark R. Cullen. 2013. “Selection on Moral Hazard in Health Insurance.” American Economic Review 103 (1): 178–219. https://doi.org/10.1257/aer.103.1.178.

  • Einav, Liran, Amy Finkelstein, and Paul Schrimpf. 2010. “Optimal Mandates and the Welfare Cost of Asymmetric Information: Evidence from the U.K. Annuity Market.” Econometrica 78 (3): 1031–1092. https://doi.org/10.3982/ECTA7245.

  • Fang, Hanming, Michael P. Keane, and Dan Silverman. 2008. “Sources of Advantageous Selection: Evidence from the Medigap Insurance Market.” Journal of Political Economy 116 (2): 303–350. https://doi.org/10.1086/587623.

  • Finkelstein, Amy, Nathaniel Hendren, and Erzo F. P. Luttmer. 2019. “The Value of Medicaid: Interpreting Results from the Oregon Health Insurance Experiment.” Journal of Political Economy 127 (6): 2836–2874. https://doi.org/10.1086/702238.

  • Finkelstein, Amy, Nathaniel Hendren, and Mark Shepard. 2019. “Subsidizing Health Insurance for Low-Income Adults: Evidence from Massachusetts.” American Economic Review 109 (4): 1530–1567. https://doi.org/10.1257/aer.20171455.

  • Finkelstein, Amy, and James Poterba. 2014. “Testing for Asymmetric Information Using ‘Unused Observables’ in Insurance Markets: Evidence from the U.K. Annuity Market.” Journal of Risk and Insurance 81 (4): 709–734. https://doi.org/10.1111/j.1539-6975.2013.12030.x.

  • Hackmann, Martin B., Jonathan T. Kolstad, and Amanda E. Kowalski. 2015. “Adverse Selection and an Individual Mandate: When Theory Meets Practice.” American Economic Review 105 (3): 1030–1066. https://doi.org/10.1257/aer.20130758.

  • Handel, Benjamin R. 2013. “Adverse Selection and Inertia in Health Insurance Markets: When Nudging Hurts.” American Economic Review 103 (7): 2643–2682. https://doi.org/10.1257/aer.103.7.2643.

  • Karlan, Dean, Robert Osei, Isaac Osei-Akoto, and Christopher Udry. 2014. “Agricultural Decisions after Relaxing Credit and Risk Constraints.” The Quarterly Journal of Economics 129 (2): 597–652. https://doi.org/10.1093/qje/qju002.

  • Mahoney, Neale, and E. Glen Weyl. 2017. “Imperfect Competition in Selection Markets.” The Review of Economics and Statistics 99 (4): 637–651. Preprint and publication record: https://www.nber.org/papers/w20411.

  • Rothschild, Michael, and Joseph Stiglitz. 1976. “Equilibrium in Competitive Insurance Markets: An Essay on the Economics of Imperfect Information.” The Quarterly Journal of Economics 90 (4): 629–649. https://doi.org/10.2307/1885326.

  • Starc, Amanda. 2014. “Insurer Pricing and Consumer Welfare: Evidence from Medigap.” The RAND Journal of Economics 45 (1): 198–220. https://doi.org/10.1111/1756-2171.12048.