RESEARCH DOSSIER
Optimal Indemnity and Coinsurance Economics
Abstract. Optimal indemnity economics studies how an insurance contract should distribute a stochastic loss between the insured and the risk bearer when premiums, preferences, information, prevention, and enforceability are constrained. The object of choice is the entire state-contingent indemnity function, not merely the headline coverage percentage. This monograph develops the expected-utility benchmark, the deductible result under proportional premium loading, restricted coinsurance design, coverage limits, state-dependent utility, and risk-bearing costs. It then integrates unobservable prevention, verification, adverse selection, behavioral choice, and empirical health-insurance cost sharing into a unified but explicitly assumption-dependent appraisal of insurance benefit design. Under an exogenous loss distribution, risk-neutral insurer, indemnity bounded between zero and the realized loss, and a premium linear in expected indemnity, fair pricing supports full coverage and strictly positive proportional loading ordinarily supports a deductible-type allocation. That conclusion changes when the insured's actions affect loss frequency or severity, claims are costly to validate, preferences depend on the state, insurers bear capital risk, or contract options and market composition are endogenous. A worked numerical example demonstrates the observational equivalence of a deductible and proportional coinsurance in a binary-loss economy, while a three-state example distinguishes their distribution of residual risk. Peer-reviewed evidence establishes significant heterogeneity in deductible choice and cost-sharing responses without identifying one universally optimal schedule. The original contribution is a transparent analytical comparison, source-grounded qualification of model domains, and an explicit account of which data would identify alternative policy evaluations. No new named theory, econometric estimate, randomized experiment, or universal optimality doctrine is claimed.
Keywords: Optimal indemnity; insurance economics; coinsurance; deductible; stop-loss coverage; expected utility; premium loading; risk aversion; actuarially fair premium; state-contingent consumption; indemnity schedule; contract design; moral hazard; self-protection; adverse selection; insurance demand; asymmetric information; behavioral insurance; health care cost sharing; insurance welfare; coverage limits; insurer risk capital; structural identification.
Introduction
Insurance indemnity is a rule that assigns a transfer to a loss realization. An insurer may reimburse the entire covered loss, reimburse only the part above a deductible, reimburse a constant proportion, stop paying beyond a contractual limit, or apply several rules sequentially. These arrangements distribute risk, determine the premium base, alter the private marginal cost of insured events, and affect the feasibility of claim settlement. Optimal indemnity economics asks which rule is preferred under specified preferences, prices, information, and institutional constraints.
The classical question was sharpened by Mossin's rational insurance purchasing analysis, Arrow's generalized deductible study, and Raviv's optimal policy design article. Together, these works locate the economics of insurance in the choice of state-contingent transfers. They also demonstrate why a policy clause is not an economic conclusion: the optimal payout depends on the conditions assigned to wealth, loss, premiums, and the risk bearer.
This dossier evaluates indemnity as an economic allocation. Insurance Contract Law addresses the enforceable obligation and applicable legal meaning; Indemnity and Insurance Benefit Design addresses the taxonomy and operation of policy benefits. The present research asks what an insurer-insured allocation accomplishes when the outcome is valued under a stated welfare criterion. Its principal Science is Insurance Economics; the same question also bears on Risk Theory, Insurance Pricing and Reserving, Health Insurance, Casualty and Liability Insurance, and Insurance Contract Law.
The research method combines explicit constrained optimization, checks of limiting cases, comparative contract analysis, and critical reading of primary peer-reviewed studies and official institutional definitions. Mathematical statements derived here are identified as derivations from established frameworks. Empirical statements retain their original population, intervention, and identification limits. Publisher abstracts or archival bibliographic pages establish citation identity when the underlying subscription text was unavailable; no uninspected technical proof is attributed to a paywalled source.
The central distinction is between a first-best risk allocation when the stochastic loss is given and a feasible insurance contract when actions, expenditures, verification, or insurer risk are endogenous. A policy that smooths wealth across states can also alter those states. Consequently, insurance design requires an account of the distribution being insured and of the mechanisms through which the contract changes it.
Insurance Indemnity as a Payoff Function
The contractual state space
Let L be a nonnegative stochastic covered monetary loss and w initial wealth before the premium. An indemnity function I maps a covered loss realization l into a payment I(l). For the elementary indemnity model, the admissible set is
The realized retained loss is R(l)=l-I(l). If P(I) denotes the premium paid in advance, terminal monetary wealth is
The insured evaluates the random final wealth, not an indemnity in isolation. The premium is a certain cost in the baseline one-period formulation. The payout is contingent on the realized event. The permissible set may also contain claims exclusions, waiting periods, waiting-time costs, sublimits, co-payments, aggregate limits, and proof requirements, which must be specified before treating observed payments as a function of loss.
The elementary constraint I(l)≤l expresses a monetary indemnity formulation for a covered financial loss. It is an analytical modeling convention rather than a statement that every insurance product in every jurisdiction uses an identical legal indemnity rule. Fixed-benefit life and accident coverage, valued policies, replacement-cost provisions, and reinsurance transfers can require a different state description. An insurance payout is economically described by its feasible state-contingent transfer, with the contract's legal object determining which state is covered.
The distribution of L matters as much as its support. Loss frequency, severity, and dependence create different demand for tail protection. Two policies with identical expected payouts may allocate insurance to different loss states. Where an insurance portfolio is subject to correlated catastrophe losses, the capital cost of paying large indemnities can differ from the cost of paying many small independent claims.
Distinct benefit schedules and meanings of coinsurance
For a straight deductible d≥0 with no cap, the usual excess-loss payment is
A pure proportional indemnity with insurer share α∈[0,1] pays I_α(l)=αl. The insured retains the fraction 1-α. A deductible followed by proportional participation pays I_{d,α}(l)=α(l-d)+. A contract with a maximum insurer payment m pays I{d,m}(l)=min{(l-d)_+,m}. An aggregate stop-loss or annual out-of-pocket ceiling operates on accumulated losses and cannot always be represented by a function of one claim alone.
The phrase coinsurance has several distinct technical uses. In elementary insurance economics it denotes proportional sharing of a specified loss. In US health insurance it commonly denotes the patient's stated percentage of covered allowed costs after a deductible: the official HealthCare.gov coinsurance definition expressly describes the consumer's share rather than the insurer's α. In property insurance a coinsurance clause can penalize a policyholder for purchasing a limit below a stipulated fraction of insurable value; that formula is tied to limits and valuation, not simply an insurer's constant fraction of every loss. In life reinsurance, coinsurance can mean a transaction transferring a share of specified insurance obligations and associated economics to a reinsurer. The same word therefore requires a declared state space and contractual unit of analysis.
For the health-insurance case, if the patient pays share q after deductible d and no annual cap binds, the insurer share of eligible accumulated expenditure x is approximately (1-q)(x-d)_+, subject to the particular plan's service-specific rules. Once an applicable annual out-of-pocket maximum is reached, insured cost sharing may cease for covered in-network services. The official out-of-pocket maximum explanation also identifies expenditures outside this ceiling, including premiums and uncovered services. These details make a constant-α description unreliable across the entire benefit year.
The premium and the insurer's objective
The simplest premium is actuarially fair: P(I)=E[I(L)]. It transfers expected losses without a loading and abstracts from operating expenditure, taxes, capital, and insurer risk preferences. A conventional linear-loading representation is
The loading λ describes the proportional amount added to expected indemnity, not necessarily a competitive equilibrium or an insurer's actual profit rate. A fixed administrative charge K produces P(I)=K+(1+λ)E[I] for policies sold, and affects the decision to purchase rather than the marginal choice among indemnities if K is independent of the schedule. A per-claim adjustment expense, a convex tail-risk capital requirement, or a premium dependent on prevention cannot generally be compressed into a constant λ.
The insurer's constraint can be expressed in several ways: actuarial balance, expected zero profit, a capital or ruin limit, a risk-adjusted value criterion, or a regulatory and contractual obligation. If the insurer is itself risk averse or faces scarce risk-bearing capital, expected transfer alone is an incomplete valuation of its liability. Borch's reinsurance equilibrium analysis places risk sharing between risk-bearing parties at the center of an alternative allocation problem.
Optimization criteria and economically feasible indemnity
The standard private choice criterion is expected utility of terminal wealth, E[u(W_I(L))], with u'>0 and u''<0 on the relevant domain. These assumptions express positive marginal value of wealth and strict risk aversion. In other designs the insured values health, interruption costs, or nonmonetary state characteristics, requiring a utility u(W,s) that varies with the state s.
A contractual optimum must state whose welfare counts. Maximizing the buyer's utility at an exogenously specified premium schedule differs from maximizing a weighted sum of insured and insurer utilities, social surplus with external effects, or a regulator's distribution-sensitive criterion. The optimum can be private, Pareto efficient conditional on an information structure, or socially preferred among feasible regulations. The vocabulary should preserve these distinctions.
Budget constraints, participation requirements, deductibility and tax treatment, credit limits, exclusions, claim proof, and insurer solvency delimit actual feasible offers. A mathematical indemnity whose payments cannot be verified or financed is outside the implementable contract set. This observation is the link between optimal insurance design and contract economics.
Deductibles and Coinsurance
Fixed deductibles, proportional participation, and caps
A deductible places the first d units of an eligible loss on the insured and insures marginal losses above d at 100 percent, assuming the covered claim remains below any cap. Straight coinsurance places a constant fraction of every eligible marginal loss on the insured. A deductible followed by coinsurance places full marginal responsibility on the insured below d and partial responsibility above d. A maximum indemnity m returns full marginal responsibility to the insured once the payment limit is reached.
These marginal retention schedules are not interchangeable. For deductible coverage R_d(l)=min(l,d), the derivative of retained loss with respect to eligible l is one below d and zero above d (except at the kink). For proportional insurance R_α(l)=(1-α)l, the retained marginal loss is 1-α throughout the covered range. For a payment cap R_{d,m}(l)=l-min((l-d)_+,m), the insured's marginal retention is one below d, zero between d and d+m, and one again beyond d+m.
These slopes matter for decisions that influence the size of a potential loss. If a policyholder can reduce a loss by one dollar, the direct monetary benefit of prevention depends on the marginal retained loss in the affected range. If the choice instead changes the probability of any loss, the full utility gap between the no-loss and loss outcomes matters. Frequency and severity responses therefore require separate incentive analyses.
The restricted proportional-coverage first-order condition
Suppose available insurance is restricted to I_α(l)=αl, 0≤α≤1, while the premium is (1+λ)αE[L]. Expected utility becomes
For an interior optimum α* and differentiable u,
The condition balances the state-weighted gain from a greater payout against the cost of its sure premium. A second derivative, E[u''(W_α)(L-(1+λ)E[L])²], is nonpositive under concavity, so an interior zero of the first derivative is a global maximum within the proportional family. Boundary evaluation is required when the derivative does not cross zero.
With λ=0, α=1 yields constant final wealth and is optimal under the benchmark. With λ>0, positive expected losses, and feasible complete coverage, the derivative evaluated at α=1 is negative: V'(1)=-λE[L]u'(w-(1+λ)E[L]). Thus complete proportional insurance is not optimal within this family under those assumptions; the selected share is less than one, possibly zero. This local calculation does not imply a positive insured share in every real policy because actual policies may be subsidized, tax-affected, state-dependent, or subject to external objectives.
The proportional family restricts which loss states can receive protection. The optimal α within that family need not reproduce the best state-contingent indemnity. A comparison is meaningful only with identical premiums, distributions, eligibility definitions, and information assumptions.
Worked binary-loss calculation
Consider an illustrative single-period economy with initial wealth 200, a loss of 100 with probability 0.10 and no loss otherwise, log utility u(c)=ln(c), and proportional premium loading λ=0.25. The monetary amounts are dimensionless examples; these are author-calculated values, not observations from an insurer or empirical study.
For a deductible 0≤d≤100, the insurer pays 100-d after the loss and zero otherwise. The premium is P_d=0.125(100-d). Final wealth is 200-P_d in the no-loss state and 200-d-P_d in the loss state. The objective is
Differentiation yields an interior optimum d*=300/7≈42.857, insurance payout 400/7≈57.143 on the loss, and premium 50/7≈7.143. The final wealth levels are approximately 192.857 without a loss and 150.000 with a loss.
Under pure proportional insurance α, premium P_α=12.5α and loss indemnity 100α. Differentiating
gives α*=4/7≈0.5714. Its payout at the only positive loss is again 400/7 and its premium 50/7. The two distinct contractual descriptions have exactly the same state-contingent payments and expected utility in this two-point loss economy.
This example supports a substantive identification conclusion: observing one positive loss size cannot distinguish the relative merits of a deductible and a constant proportional share when the indemnities agree at that loss. A claim that one form is superior requires additional loss states, different restrictions, or a prevention and verification mechanism. The example also illustrates why one cannot infer a contract's general marginal insurance schedule from its payout on a single realized event.
Multiple loss states and tail exposure
Introduce three potential losses, 0, 50, and 200. A straight deductible d=40 generates indemnities (0,10,160) and retained losses (0,40,40). Proportional indemnity α=0.8 generates (0,40,160) and retained losses (0,10,40). Both schedules pay 160 on the maximum loss, but they allocate a different amount to the intermediate loss. For a fixed loss distribution the premiums will also differ; an equal-premium comparison must adjust parameters or explicitly account for the difference.
The deductible protects high-loss states while deliberately leaving small losses uninsured; proportional coverage provides some protection even for modest losses and leaves the most severe losses partly uninsured. These characteristics follow directly from the transfer functions, without requiring a claim that either is universally preferred. For a concave utility, different expected payments, or heterogeneous losses, the preferred contract depends on wealth, state probabilities, loading, and any behavioral response.
A payment cap produces a third pattern. With I(l)=min((l-d)_+,m), retained loss grows again above d+m. That design may protect insurer capital or contain claim settlement costs but concentrates residual risk in the largest loss states. In catastrophic settings those states may be precisely where the insured's marginal utility of resources is highest. Risk capital and available reinsurance are therefore economically central to evaluating limits.
Health benefit design complicates the picture further. Service-level coinsurance, separate deductibles, network exclusions, and annual out-of-pocket ceilings make the insured's marginal price depend on past eligible expenditure and anticipated future care. HealthCare.gov's official deductible explanation clarifies that services may receive different treatment before a general deductible has been met. The schedule can be path-dependent even if each individual claim uses a simple-looking percentage.
Contract demand and behavioral identification
Observed deductible selection contains information about preferences only under an explicit model of beliefs, loss probabilities, selection, choice sets, and contract prices. Cohen and Einav's deductible-choice estimation allows heterogeneity in accident risk and risk aversion in automobile insurance. They find substantial heterogeneity in inferred risk preferences and discuss the pricing implications. The result depends on the structural identification framework; raw choices do not directly reveal an invariant coefficient of risk aversion.
Sydnor's homeowners deductible evidence shows consumers frequently buying low deductibles at prices well above the expected value of additional coverage, so a standard expected-utility fit would imply unusually high aversion to modest risks. Barseghyan and coauthors' insurance risk-preference estimates find that probability weighting helps explain deductible selections. These are evidence of behavioral and heterogeneity challenges to a literal calibration of the classical benchmark. They do not by themselves establish that policyholders make errors: valuation of convenience, liquidity, perceived probability, and other unobserved attributes must also be considered where relevant.
The economic implication is that a contract can be mathematically optimal under one calibrated utility and yet fail to maximize the same household's welfare under a richer preference or information specification. Stated certainty, comprehension, and an actual insurance choice are different empirical objects.
Optimal Coverage Conditions
State-dependent preferences and incomplete markets
The benchmark identifies the optimum by equating marginal utilities of terminal monetary wealth in covered states. With utility u(c,s), the insured's marginal utility of money may depend on the state s itself, such as illness severity, disability, or household disruption. Where one cannot purchase or deliver some needed good, a cash indemnity may not restore the relevant welfare component. Arrow (1974) emphasizes the importance of state-dependent marginal utility in the health-related insurance problem.
For a state s with priced indemnity i(s), an interior condition takes the form u_c(c(s),s)=common marginal premium cost under the same state-independent price assumption. Equal u_c need not imply equal c because the utility functions differ across states. This is a precise reason why an optimal schedule need not equalize final monetary wealth. The content of the insured event and the ability to convert a payout into needed resources are part of insurance welfare.
Incompleteness also arises from uninsurable outcomes. A fixed monetary payout may not indemnify emotional loss, lost time, business closure, or the absence of available medical capacity. A money-only model remains useful when the relevant monetary exposure is separately identified; extending its conclusion to all dimensions of harm requires additional data and a broader objective.
Prevention, observability, and contractibility
Suppose the probability of a fixed covered loss D is p(e), where higher effort e lowers probability, p'(e)<0, at private effort cost k(e), k'(e)>0. If the insured's utility is
and c_0=w-P and c_1=w-P-D+I(D), the interior effort condition is
The equation says that greater preventive effort is privately worthwhile when the probability reduction times the utility value of avoiding the loss equals its marginal cost. When the indemnity eliminates the monetary loss difference, the purely monetary incentive in this deliberately stripped-down model vanishes. If effort has intrinsic benefits or prevents nonfinancial harms, this limiting result no longer applies without modification.
The insurer anticipates the policyholder's effort response. Indemnity and premium may change p(e), making the premium endogenous to contract design. Ehrlich and Becker's self-protection and market insurance framework places these jointly chosen responses into one economic analysis. Pauly's moral hazard commentary is an early contribution to the economic treatment of utilization responses. Shavell's insurance and moral hazard analysis distinguishes the design problem when care can and cannot be observed. In more general principal-agent settings, Holmström's observability analysis explains why informative performance signals can matter for incentive contracts.
There is a fundamental distinction between a loss-prevention action that reduces event probability and an expenditure choice that determines the amount of a bill after an event. A high deductible changes the marginal price faced by someone consuming health care and may also change preventive effort, but those channels have different welfare implications. A lower quantity of treatment is a spending outcome, not automatically a welfare gain.
The optimal insurance schedule under hidden action is a second-best contract. It can leave some risk with the insured to preserve effort, while reimbursing enough risk to provide meaningful protection. The balance depends on how behavior changes losses, the insured's utility, the cost of monitoring, and the value of the resulting reduced losses. A generic instruction to 'increase coinsurance to reduce moral hazard' has no welfare content until these parameters and their consequences are specified.
Claims verification, administrative cost, and capital
Indemnity requires a claim description and a mechanism to verify that it falls within the contract. Let a payment I(l) generate a claim adjustment expense a(l,I) and an expected dispute or fraud-control cost f(l,I). The insurer's required premium might be represented schematically by
where K(I) is the financing cost of the retained portfolio exposure. This expression is an accounting decomposition for analysis, not a calibrated universal pricing model. The decomposition demonstrates why low-severity claims may be uneconomic to process and why large payments may require expensive risk-bearing capacity.
If claim verification is expensive below a threshold, a deductible can reduce the number of submitted claims. That creates an administrative reason for deductible coverage separate from the Arrow wealth-smoothing argument. If a fixed benefit is easier to validate than an exact invoice amount, a standardized payment might lower monitoring cost at the price of imperfect correlation with the actual financial loss. When reimbursement depends on unverifiable states, the feasible set becomes smaller than the mathematical family of arbitrary I(l).
Verification also creates risk for the insured. Delays, denial, documentation costs, and disputes make realized reimbursement stochastic conditional on the underlying loss. An econometric evaluation must distinguish contractual payment entitlement from actual net receipt. The insured's expected terminal resources then incorporate the probability, timing, and cost of settlement.
Insurer capital creates additional constraints. The marginal funding cost of indemnifying a highly correlated disaster may exceed its expected value and differs from the cost of an independently diversified small claim. Coverage limits can consequently reflect financing and solvency pressures. Such limits may be rational for the insurer and still leave a large residual societal exposure. Welfare comparison requires the marginal costs on both sides of the contract.
Selection, pooling, and market equilibrium
When the insured knows more about expected loss than the insurer, the insurer faces adverse selection. Low-risk individuals may prefer different contracts from high-risk individuals, altering the pool purchasing each indemnity. Rothschild and Stiglitz's competitive insurance equilibrium model demonstrates the importance of information and self-selection for insurance market outcomes. A privately efficient indemnity for a known type need not be available at a pooling premium and need not remain an equilibrium offer when buyers select policies strategically.
Insurance menus can be designed to induce self-selection. A low-premium contract with a high deductible may attract lower expected-cost individuals, although observed selection also depends on risk preferences, income, beliefs, and prevention. Einav and Finkelstein's selection evidence review distinguishes adverse from advantageous selection across markets. Consequently a mechanically increasing cost curve by coverage is an empirical proposition, not a necessary property of every insurance population.
A market equilibrium optimum differs from the individual first-best. Changing one policy's coinsurance may alter which people enroll, how much they spend, and the average cost of the remaining pool. The feedback can change prices and the feasible contract set for all buyers. In an equilibrium analysis, the whole market response must be considered rather than holding all other contracts and premiums fixed.
Institutional and legal boundaries
Optimality claims must identify the covered economic unit and relevant institutional constraints. A statutory health insurer, a commercial property insurer, a liability carrier, and a reinsurer face different entitlements, conduct obligations, market structures, and information. Policy interpretation, insurable interest, minimum benefits, claims handling, and mandatory coverage rules are jurisdiction-specific. This dossier does not assert a single legal rule for all countries.
International comparison should therefore record the insured event, payment unit, limit, deductible, coinsurance definition, basis of claim valuation, insurer obligations, and premium financing before comparing percentages. A nominal 20 percent coinsurance rate has little comparative meaning without knowing whether it represents the consumer's share of an allowed health service, the insurer's share of a commercial loss, or a risk share ceded to a reinsurer.
Verification and Incentive Constraints
Ex ante welfare, observed choice, and actual payouts
The basic welfare object is an expectation over prospective outcomes, including premium payment and the full distribution of retained loss. A high realized indemnity is not itself proof that a policy was optimally purchased; it is one realization conditional on an insured event. Similarly, a reduction in claims expense can arise from lower use, different timing, a new exclusion, better prevention, a change in severity, or altered claim reporting. These effects cannot be assigned the same welfare sign.
The empirical design must observe or credibly identify the counterfactual: what would consumption, effort, use of services, and claim settlement have been under another feasible schedule? Without a counterfactual, a difference in payouts across insurance products cannot identify the welfare gain of a deductible or a coinsurance percentage.
A credible assessment should specify whether the objective is private expected utility, public net social benefit, producer surplus, financial protection, health outcomes, expenditure stabilization, or an integrated social objective. When multiple outcomes matter, their weighting must be explicit. An equilibrium price that balances insurer costs may still fail to maximize household welfare because of imperfect information or distributional concerns.
Empirical evidence on health cost sharing
The RAND Health Insurance Experiment supplies randomized evidence that medical use responds to cost sharing. Manning and colleagues' randomized medical-care study reports that catastrophic coverage reduced expenditures by about 31 percent relative to free care at the point of service and estimates a demand price elasticity near -0.2 in its experimental context. These findings identify expenditure responses to the studied insurance arrangements and populations; they are not a universal prescription for a health-plan deductible or a full social-welfare measurement.
Brot-Goldberg and colleagues' high-deductible transition study examines a large self-insured firm's move from free care to a nonlinear high-deductible plan. The authors report an approximately 11.8–13.8 percent decrease in total health spending and attribute the observed reduction to quantities rather than consumer price shopping. A reduction in service volume can include both low-value and high-value care; expenditure reduction alone therefore cannot establish net benefit.
The correct theoretical response to this evidence is to modify the loss and expenditure process rather than simply lower the insurer's payout under a fixed loss distribution. Health expenses are partly outcomes of utilization choices. Insured services also differ in quality, substitutability, and long-term effects. A static indemnity function may provide one contractual description but cannot by itself value the endogenous health consequences.
Einav and colleagues' selection on moral hazard study finds heterogeneous behavioral responses and evidence that consumers partly select plans according to their expected responses. Ignoring that heterogeneity can overstate predicted savings from offering a high-deductible plan in their empirical setting. The finding directly challenges an exercise that projects utilization from one average elasticity while holding enrollment fixed.
Empirical evidence on insurance preferences and selection
Structural estimates of deductible demand are not transferable coefficients without testing how risks, beliefs, contract terms, and wealth differ. Cohen and Einav (2007) model heterogeneous deductible choices; Sydnor (2010) identifies high-priced small-loss protection; and Barseghyan and colleagues (2013) document the importance of probability weighting in insurance. Taken together, they support the empirical relevance of preference heterogeneity and the need to test alternative behavioral specifications.
A person choosing a low deductible might value quick cash availability, be strongly averse to losses, misperceive probabilities, overweight a small chance of loss, or face a restricted choice menu. Claim records help identify expected cost but do not always identify beliefs. Observed selection is the joint product of risk, preferences, eligibility, pricing, marketing, and information.
Handel's health-plan inertia analysis illustrates how reducing choice frictions can interact with adverse selection and lower aggregate welfare in the specific setting studied, even as individuals make better-informed choices. This is a reminder that the market-wide effect of a policy intervention cannot be determined solely from an isolated household decision.
Einav, Finkelstein, and Cullen's price-variation welfare method uses changes in insurance prices to estimate demand and how insurer costs vary with enrollment, permitting welfare analysis of selection. In their employer-based application, the estimated welfare cost of detected adverse selection is relatively small. The method demonstrates a tractable empirical bridge between contract demand and social welfare, subject to identified variation, measurement, and external-validity limits.
Causal and structural identification
For a causal comparison of two indemnity designs, a treatment must be specified with enough precision to determine which benefit schedule changed. 'High deductible' and 'more coinsurance' are incomplete descriptions if networks, premiums, out-of-pocket ceilings, or utilization management also change. An experimental assignment can identify an intention-to-treat effect, while effects of actual coverage use may require additional compliance assumptions.
For structural welfare identification, researchers require a model of preferences and beliefs, a model for losses and behavioral response, available contract menus, and reliable price and payment data. If one observes only insured claims, the distribution of uninsured losses can be censored or unreported. A policy with a large deductible may yield fewer submitted small claims even when the underlying loss frequency is unchanged. Administrative datasets that equate reported claims with all events risk attributing changed reporting to prevention.
Credible identification should address time selection, adverse selection, plan switching, dynamic incentives around an annual deductible, provider pricing responses, liquidity constraints, and spillovers to uninsured parties. The results of one workforce's health-plan transition need not identify the benefits of the same schedule for a national population or an unrelated line of insurance.
Cross-jurisdiction comparisons require special care: premium tax treatment, public subsidies, provider networks, claim accounting, legal indemnity rules, and data coverage can vary jointly with coinsurance rates. A regression of national insurance penetration or average deductibles on outcomes cannot isolate the marginal effect of an indemnity schedule without a defensible design.
Model validation and evidence limitations
A rigorous design evaluation should pass five independent checks. First, verify that contractual indemnities satisfy the coverage and payment bounds actually imposed. Second, verify premium arithmetic and insurer cost treatment. Third, compute statewise final wealth and expected utility under the stated preference assumptions. Fourth, distinguish parameters fixed by the model from choices changed by the contract. Fifth, inspect empirical evidence at the level of its original intervention, sample, and identification procedure.
Useful stress cases include λ=0, deterministic loss, zero demand, maximal retention, a very large loss beyond a limit, complete prevention, and an insurer unable to pay at the moment of a disaster. These cases often reveal whether a purported optimality statement is an artifact of hidden assumptions.
The present evidence record includes direct official definitions, original peer-reviewed publisher metadata and abstracts, and accessible full scholarly article versions for selected modern studies. Some foundational full texts are subscription restricted. This monograph does not claim a page-by-page reproduction of those unavailable texts. Its mathematical benchmark is independently derived here and compared with the verified scope of canonical sources. It offers no original dataset, new causal estimates, or formal systematic review with exhaustive publication screening.
Comparative Optimality Results
A comparative results matrix
The principal comparisons can be summarized without treating any result as a universal policy rule.
| Economic environment | Candidate allocation | Governing consideration | Scope of conclusion |
|---|---|---|---|
| Fixed loss distribution, strictly concave monetary utility, actuarially fair linear premium | Full eligible-loss indemnity | Jensen's inequality and constant terminal wealth | First-best private expected-utility benchmark |
| Fixed loss distribution and positive uniform proportional loading, flexible indemnity choice | Straight deductible-type payment | Equal marginal utility in states receiving marginal insurance | Classical constrained private optimum under the specified assumptions |
| Same environment, but insurer permits only proportional coverage | Interior or boundary insurer share α | Restricted first-order condition and concavity | Optimal only within the proportional menu |
| State-dependent utility | State-responsive indemnity | Marginal utility varies with insured event | Identical terminal monetary wealth need not be optimal |
| Unobservable prevention or utilization | Partial coverage or contract with other incentives | Behavioral response changes the distribution of losses or expenses | Incentive-constrained second-best, requiring behavioral parameters |
| State-specific verification or risk capital cost | Nonlinear layers, caps, or other admissible terms | Marginal cost of indemnity varies by state | Depends on claim costs, capital, and risk-bearing objective |
| Heterogeneous types with private information | Self-selection menu or altered pooling allocation | Enrollment changes premiums and average costs | Equilibrium welfare rather than isolated buyer's optimum |
| Behavioral preferences and restricted understanding | Menu and disclosure jointly evaluated | Beliefs, choice frictions, probability weighting | Welfare depends on identified preference and information model |
The table records analytical implications of established frameworks and the original comparisons developed in this dossier. It is not a proposed new theory or a claim that every row has been quantitatively estimated for all insurance markets.
Implications for benefit design
A benefit designer should start with the policy's economic object. What loss is being insured? Who bears the premium? What states make a payout valuable? Can eligibility and loss severity be verified? Which actions will change after coverage? Who supplies the capital and bears residual catastrophe risk? Once these objects are described, a deductible or coinsurance percentage becomes a meaningful parameter rather than an isolated design convention.
For low-frequency severe losses with a reliable private loss distribution, risk aversion can make upper-layer insurance valuable. A deductible may conserve premium expenditure by leaving low-severity fluctuations to household or firm resources. This logic must be modified when the first units of loss produce liquidity distress or when protection against moderate losses has nonmonetary value. A deductible set by a wealth-rich representative consumer can be unsuitable for a liquidity-constrained insured who otherwise has similar expected loss.
For a health insurance contract, a single constant coinsurance rate across all services is unlikely to capture differences in medical value, price sensitivity, and future health consequences. The evidence of Manning and colleagues and Brot-Goldberg and colleagues justifies careful estimation of utilization responses, while welfare analysis must still evaluate health and financial security. In property or casualty settings, prevention effort, adjustment cost, recovery time, limits, and correlated catastrophe exposure may be more decisive than service utilization.
Risk classification affects attainable designs. If expected losses vary across insureds and information is incomplete, a flat price cannot simply be taken as actuarially fair for each participant. Contract menus and regulatory restrictions may change the nature of market selection. Einav and Finkelstein's review of selection patterns counsels against assuming that more insurance necessarily selects more expensive types in all markets.
Practical comparison should report the whole function I(l), the premium function P(I), the retained loss distribution R(L), the insurer's expected and tail liabilities, and identified behavioral responses. A headline deductible or coinsurance rate is an insufficient welfare statistic.
International applicability and research agenda
The theoretical benchmark is jurisdiction-neutral because it defines preferences, losses, and feasible transfers abstractly. Actual implementation is institution-dependent. International research should compare contracts only after aligning the measurement of covered loss, benefit basis, price, beneficiary, claims obligations, aggregate limits, and public subsidy. Public health insurance, commercial liability insurance, and reinsurance can all be studied as contingent transfers, but each requires an institutional model appropriate to its financing and behavioral environment.
Three research priorities follow from the reviewed literature. The first is the joint estimation of risk preferences, subjective beliefs, and the distribution of losses that would be observed without insurance or with a different benefit schedule. The second is the separate estimation of frequency prevention, severity mitigation, claim reporting, and utilization, since a change in observed paid claims can arise through any of these channels. The third is the estimation of insurer capital and settlement-performance costs across the distribution of losses, particularly for correlated extreme events. These are proposals for empirical inquiry, not claims of completed new research.
A well-identified comparison of indemnity structures therefore requires a clearly specified counterfactual menu and a common social accounting boundary. It must incorporate the benefit and the cost of risk transfer, along with changes in the behavior and composition of insured populations. Under those conditions, optimal indemnity analysis can inform premiums, contract regulation, welfare evaluation, and reinsurance arrangements while maintaining the distinction between a model's theorem and its empirical applicability.
Conclusion
Optimal indemnity is a state-contingent allocation problem. Its solution depends on the loss distribution, insured preferences, the premium schedule, insurer financing, feasible claim verification, and the behavioral and market responses induced by coverage. Under the classical one-period benchmark, actuarially fair coverage permits complete monetary loss insurance, and positive uniform loading with a flexible indemnity set supports a deductible-type allocation. This result follows from an explicit expected-utility comparison and the equalization of marginal utility where additional coverage is purchased.
Deductibles, proportional coinsurance, and indemnity caps assign different marginal exposures to the insured. A restricted optimum within one contract family does not establish the global optimum across all possible payment functions. The numerical binary-loss example shows that a deductible and proportional contract can produce exactly the same payments and welfare when only one positive loss size is possible. A richer loss distribution reveals their differences in retention and tail protection.
The principal sources of divergence from the benchmark are economically identifiable. State-dependent utility changes the value of money in insured states. Unobservable prevention and medical utilization make the loss or expenditure distribution endogenous. Verification costs and insurer capital alter the marginal price of payments. Adverse selection changes who buys coverage and the equilibrium premium. Behavioral heterogeneity and choice frictions complicate inference from selected deductibles to welfare.
The empirical literature documents meaningful utilization responses, heterogeneous deductible preferences, market-selection mechanisms, and important identification constraints. Those studies support differentiated contract analysis, while their settings limit quantitative generalization. The appropriate scientific conclusion is conditional and operational: an optimal indemnity schedule is one that maximizes a declared objective over an accurately specified feasible contract set, with the causal response of insureds and the cost of risk bearing included wherever material.
The research yields a source-grounded comparative synthesis and transparent mathematical illustrations. It does not introduce a new theoretical system, assert an unverified scientific priority, or assign an unpublished manuscript a public publication date.
References
Arrow, Kenneth J. (1974). “Optimal Insurance and Generalized Deductibles.” Scandinavian Actuarial Journal 1974 (1): 1–42. Publisher article and DOI.
Barseghyan, Levon, Francesca Molinari, Ted O'Donoghue, and Joshua C. Teitelbaum. (2013). “The Nature of Risk Preferences: Evidence from Insurance Choices.” American Economic Review 103 (6): 2499–2529. Journal article and DOI.
Borch, Karl. (1962). “Equilibrium in a Reinsurance Market.” Econometrica 30 (3): 424–444. Archival article and DOI.
Brot-Goldberg, Zarek C., Amitabh Chandra, Benjamin R. Handel, and Jonathan T. Kolstad. (2017). “What Does a Deductible Do? The Impact of Cost-Sharing on Health Care Prices, Quantities, and Spending Dynamics.” Quarterly Journal of Economics 132 (3): 1261–1318. Publisher article and DOI.
Cohen, Alma, and Liran Einav. (2007). “Estimating Risk Preferences from Deductible Choice.” American Economic Review 97 (3): 745–788. Journal article and DOI.
Ehrlich, Isaac, and Gary S. Becker. (1972). “Market Insurance, Self-Insurance, and Self-Protection.” Journal of Political Economy 80 (4): 623–648. Publisher article and DOI.
Einav, Liran, and Amy Finkelstein. (2011). “Selection in Insurance Markets: Theory and Empirics in Pictures.” Journal of Economic Perspectives 25 (1): 115–138. Journal article and DOI.
Einav, Liran, Amy Finkelstein, and Mark R. Cullen. (2010). “Estimating Welfare in Insurance Markets Using Variation in Prices.” Quarterly Journal of Economics 125 (3): 877–921. Publisher article and DOI.
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. Journal article and DOI.
Handel, Benjamin R. (2013). “Adverse Selection and Inertia in Health Insurance Markets: When Nudging Hurts.” American Economic Review 103 (7): 2643–2682. Journal article and DOI.
HealthCare.gov. (n.d.). “Coinsurance.” Glossary, US Department of Health and Human Services. Accessed November 14, 2026. Official coinsurance definition.
HealthCare.gov. (n.d.). “Deductible.” Glossary, US Department of Health and Human Services. Accessed November 14, 2026. Official deductible definition.
HealthCare.gov. (n.d.). “Out-of-Pocket Maximum/Limit.” Glossary, US Department of Health and Human Services. Accessed November 14, 2026. Official out-of-pocket definition.
Holmström, Bengt. (1979). “Moral Hazard and Observability.” Bell Journal of Economics 10 (1): 74–91. RePEc journal bibliographic record.
Manning, Willard G., Joseph P. Newhouse, Naihua Duan, Emmett B. Keeler, Arleen Leibowitz, and M. Susan Marquis. (1987). “Health Insurance and the Demand for Medical Care: Evidence from a Randomized Experiment.” American Economic Review 77 (3): 251–277. National Library of Medicine record.
Mossin, Jan. (1968). “Aspects of Rational Insurance Purchasing.” Journal of Political Economy 76 (4, Part 1): 553–568. Publisher article and DOI.
Pauly, Mark V. (1968). “The Economics of Moral Hazard: Comment.” American Economic Review 58 (3): 531–537. JSTOR archival article.
Raviv, Artur. (1979). “The Design of an Optimal Insurance Policy.” American Economic Review 69 (1): 84–96. RePEc article bibliographic record.
Rothschild, Michael, and Joseph Stiglitz. (1976). “Equilibrium in Competitive Insurance Markets: An Essay on the Economics of Imperfect Information.” Quarterly Journal of Economics 90 (4): 629–649. Publisher article and DOI.
Shavell, Steven. (1979). “On Moral Hazard and Insurance.” Quarterly Journal of Economics 93 (4): 541–562. Publisher article and DOI.
Sydnor, Justin. (2010). “(Over)insuring Modest Risks.” American Economic Journal: Applied Economics 2 (4): 177–199. Journal article and DOI.