RESEARCH DOSSIER
The Behavioral Insurance Paradox of Small and Catastrophic Losses
Abstract. Insurance demand displays a striking divergence across risk scales. In documented settings, households select costly protection against modest, relatively familiar losses, while large and infrequent exposures remain uninsured or incompletely insured. This monograph investigates the empirical content, mechanisms, and welfare interpretation of that divergence. It integrates research on household deductible choices, behavioral probability weighting, health-plan comprehension, flood-insurance demand, disaster learning, contract accessibility, liquidity, and cross-jurisdictional protection gaps. The central methodological distinction is between observing two patterns in different populations or insurance lines and proving that the same households exhibit both. A state-contingent financial benchmark identifies when paying to reduce a deductible is expensive relative to expected indemnity and when a catastrophic policy would be valuable, while preserving the roles of premium loads, household wealth, background risk, contract exclusions, and liquidity. Evidence from the United States, Germany, India, and Pakistan reveals heterogeneous causes: biased assessment and attention, information frictions, credit and trust constraints, imperfect index coverage, regulatory mandates, and supply-side availability. The resulting synthesis identifies a conditional, institutionally mediated pattern rather than a universal psychological law. It establishes testable comparative predictions and implications for transparent product design, catastrophe risk communication, public protection, and rigorous future research without claiming a new theory, an original data estimate, or externally reviewed findings.
Keywords: Behavioral insurance; small-loss overinsurance; catastrophe underinsurance; deductible choice; flood insurance; probability weighting; loss salience; risk perception; financial protection gap; household liquidity; insurance demand; bounded attention; basis risk; contract comprehension; causal identification; public policy; insurance welfare.
Introduction
Insurance reallocates uncertain financial obligations across time and states of the world. Its economic purpose is usually clearest where a loss would seriously impair consumption, threaten solvency, interrupt income, or require liquidation of valuable assets. Yet observed demand sometimes appears strongest where possible losses are financially modest and weakest where uninsured losses could be devastating. An expensive low-deductible contract and an uncovered flood peril represent different ways in which the realized allocation of protection may diverge from the allocation suggested by a simple model of risk aversion and administrative cost. Understanding that divergence is central to behavioral insurance economics, catastrophe finance, household welfare, and market design.
The historical research program on hazard protection predates contemporary experimental behavioral economics. Kunreuther and colleagues' 1978 disaster-insurance investigation combined field interviews, experimental reasoning, and institutional analysis of flood and earthquake coverage. The account of insurance and behavioral economics developed by Kunreuther, Pauly, and McMorrow (2013) places the coexistence of modest-risk coverage and catastrophe protection gaps within a broad analysis of consumers, sellers, incentives, and policy. These contributions motivate a research question that can be stated more precisely than the familiar impression that individuals fear minor losses but ignore major ones: under what conditions does measured insurance choice protect small retained losses more intensively than large uncovered exposures, and what explains that allocation?
Evidence on the first side of the question comes from actual contractual choices. Sydnor's homeowners deductible analysis (2010) documents widespread selection of low deductibles whose incremental premiums greatly exceeded expected incremental claims payouts. In a separate health-insurance setting, Bhargava, Loewenstein, and Sydnor's employee plan-choice study (2017) shows that many workers selected financially dominated lower-deductible plans. These findings concern different markets, contract menus, periods, and populations; their identification of unexpected small-loss protection is stronger than any inference they can establish about those households' own catastrophe coverage.
Evidence on the second side is also concrete, although its denominator requires careful handling. Gallagher's flood take-up panel study (2014) documents a rise in flood insurance purchases after flooding and a subsequent return toward baseline. A 2026 US government audit reports that, as of April 2026, 86 percent of properties classified as high flood risk in its analysis lacked National Flood Insurance Program coverage. That is a statement specifically about NFIP coverage, not direct proof that every such property lacked all private, residual, or other forms of protection. Both observations motivate study of large-loss underprotection; neither alone establishes a common behavioral trait shared with low-deductible buyers.
The present research object is the relation between retained-loss magnitude, insurance expenditure, and protection intensity under measurable risk and institutional constraints. The core questions are whether apparently inconsistent choices are observed at the individual level; which psychological and structural mechanisms can generate them; how competing explanations may be distinguished; and what follows for economically and ethically defensible policy. The principal Science is Behavioral Insurance and Risk Perception. Its cross-disciplinary relevance extends to Catastrophe Risk, Insurance Economics, Risk Theory, and Inclusive Insurance and Microinsurance. The separate questions of choosing a particular deductible and of momentary disaster salience are related scientific objects, but this dossier concentrates on their interaction across loss scales.
The method is a critical synthesis of primary journal studies, institutional data reports, transparent financial examples, and conceptual comparisons. Numerical illustrations are constructed for this manuscript and are not measured household parameters. Publisher, journal, and issuing-agency records support article identities and the narrow empirical findings reported. Where a publisher abstract or summary is the examined source layer, claims are confined to its displayed methods and results. No private insurer microdata, household-level linked policy panel, new field experiment, or original causal estimation has been performed. Inferences are therefore graded by design: administrative contract data establish observed menus and selections; experiments identify responses in their particular settings; observational event studies constrain event-time mechanisms; and country statistics describe coverage distributions subject to changing definitions.
The Small-Loss Coverage Pattern
Defining what is actually protected
A small loss is a claim or retained liability whose absolute magnitude is limited relative to the policyholder's relevant liquid resources, consumption obligations, and longer-term financial position. A catastrophic loss is a severe, potentially ruinous, or recovery-disrupting event relative to that same economic position. Frequency concerns the probability over a specified period, such as one policy year; severity concerns the distribution of consequences conditional on an event. These dimensions interact but are analytically separate. A low-frequency event need not be catastrophic, and a common event may impose serious repeated or correlated losses.
Insurance protection intensity must also be specified. A policy's headline insured sum is insufficient: the effective distribution of net claims depends on covered perils, deductibles, waiting periods, exclusions, coinsurance, coverage limits, claims performance, and insurer solvency. A homeowners policy may insure frequent property damage while excluding flood, creating high protection against familiar nonflood claims and potentially extreme uninsured flood loss. Such a contract structure might be chosen, mandated, bundled, or simply the only feasible option. Therefore exposure to the two loss types cannot be inferred from a single annual premium.
For an annual loss variable X, let retained loss under contract c be R_c(X), including deductibles, coinsurance, and excluded events. A risk-sensitive household evaluates not just the probability of any claim but the distribution of R_c(X), annual premium P_c, timing of payments, and its ability to survive a severe retained loss. A contract that lowers a deductible on routine claims while leaving flood entirely excluded can reduce frequent retained payments without appreciably changing the worst relevant economic outcome. This is a useful representation of the empirical object, although the valuation of that difference requires a behavioral and economic model.
Contract-menu evidence and the Sydnor result
Sydnor (2010) examined homeowners' choices among actual deductibles. The published article's account reports that most observed customers purchased relatively low deductibles despite incremental prices substantially exceeding the expected indemnity advantage. Imposing conventional expected utility on these selections required implausibly large local risk-aversion parameters. This finding is an important empirical challenge to the idea that demand for expensive protection against limited risks can always be interpreted as standard smooth concavity of utility for final wealth.
The finding is also bounded. It does not establish that every low deductible is suboptimal, that every consumer has the same probability beliefs, or that all customers had access to costless and transparent switching. A customer may incorrectly overestimate claim frequency, have strong reference-dependent aversion to paying a deductible, regard administrative hassle as costly, or use the default displayed by an intermediary. Policyholders also differ in wealth, expected losses, insurance knowledge, and prior experience. A credible explanation must retain these possibilities until the data or a design can distinguish them.
Cohen and Einav's automobile deductible estimation (2007) provides a complementary, structurally estimated account. Their model allows unobserved heterogeneity in both claims risk and risk aversion; the analysis finds a substantial and skewed distribution of inferred risk attitudes. That result matters because a homogeneous representative consumer is a poor empirical substitute for the actual market. Moreover, estimated willingness to avoid a deductible is sensitive to assumed beliefs, the feasible choice set, and heterogeneity. The resulting estimates are conditional objects rather than direct psychological measurements.
The later analysis of insurance choices and probability distortion by Barseghyan, Molinari, O'Donoghue, and Teitelbaum (2013) estimates a model combining conventional risk aversion and nonstandard perception or weighting of probabilities. Its reported result assigns an important role to overweighting of small probabilities in explaining observed deductible choices. This is a significant qualification to the naive interpretation of low-deductible demand: willingness to pay extra for modest protection can be high even where smooth utility curvature would predict a much smaller benefit.
Dominance and comprehension in health-plan choices
Bhargava, Loewenstein, and Sydnor (2017) investigate a setting with a particularly informative financial comparison. The health-plan dominance study uses choices by 23,894 employees at one US firm. Many menu options differed in cost-sharing and premium, while lower-deductible options were often financially dominated by higher-deductible choices for all relevant medical-spending outcomes. The majority selected dominated plans, and the researchers estimated excess spending equal to approximately 24 percent of selected plan premiums. Their experiments and analysis emphasized failure to understand health-insurance contracts rather than an unusually strong rational preference for avoiding small deductibles.
Dominance is stronger evidence of a decision failure than a premium that merely exceeds the expected reduction in claims. Suppose plan A costs more than plan B in annual premiums, and every possible medical-use state produces total premiums plus out-of-pocket spending under A at least as high as under B. If all other plan attributes, risk, and payment timing genuinely match, A offers no financial state-contingent advantage. Risk aversion by itself cannot rescue a dominated choice, because the supposed insurance improvement never arrives in any state. The strength of this conclusion depends on the study's documented equivalence of contract attributes and on whether relevant nonfinancial benefits are genuinely absent.
This evidence is especially valuable because it separates psychological fear of a large deductible from misunderstanding how a deductible interacts with premiums, coinsurance, and maximum out-of-pocket exposure. It also supplies a warning against inferring true risk preferences from a survey answer or selected plan without verifying whether the buyer understood the contractual consequences. The concern aligns with Handel and Kolstad's information-friction analysis (2015), which combines administrative health-choice information with survey measures and shows that knowledge and hassle costs substantially alter estimated risk-preference interpretations.
A disciplined statement of the observed pattern
Small-loss protection can be expensive, but the price comparison must be normalized by the incremental transfer being purchased. The relevant quantity is the difference between two contracts in premiums, relative to the induced change in state-contingent retained loss, rather than the mere fact that low-deductible insurance has a positive premium. A person willing to pay a moderate extra premium for a low deductible may act rationally if the probability of a claim is unusually high, liquidity costs from a deductible are consequential, or the policy avoids substantial transaction costs. A health plan that is strictly dominated under verified equivalence presents a distinct and stronger case.
Accordingly, the first side of the paradox rests on several levels of evidence. Actual deductible selections demonstrate market behavior, structural studies model how risk and beliefs can generate it, and dominated-plan comparisons isolate choices whose financial ordering can be evaluated without selecting a particular smooth risk-aversion coefficient. None of those designs alone says how much flood, earthquake, wildfire, crop-failure, or systemic-event protection the identical households purchase. That within-household contrast remains a separate empirical hypothesis.
The Financial and Decision-Theoretic Benchmark
A transparent two-contract calculation
Consider a household choosing between two annual policies that cover the same peril, offer otherwise identical settlement terms, and differ solely in deductible and price. Let the high deductible be 1,000 monetary units and the low deductible be 250. Suppose the annual extra premium for the low deductible is 150 and the probability of exactly one covered claim is 0.05. For a claim of at least 1,000, the maximum additional insurance payment from choosing the lower deductible is 750; for smaller claims, the incremental payment is lower. The actuarial expected extra indemnity is therefore at most 0.05 times 750, or 37.5. In this constructed benchmark, the extra premium is at least four times the maximum expected payout. The calculation is conditional on the assumptions stated; it is not a reconstruction of any particular insurer's observed menu.
A useful extension writes the difference in retained loss as D(X) = min(X,1,000) - min(X,250) for a nonnegative covered claim X, with D(X) between zero and 750. If at most one claim occurs in the policy period, the expected indemnity gain is E[D(X)]. If a claim arrives with probability q and severity distribution F conditional on arrival, the gain equals q times the conditional expectation of D(X). It cannot exceed 750q. Under risk neutrality and absent transaction costs, paying an incremental premium ΔP for reduced retained risk is beneficial only if ΔP is no greater than E[D(X)]. With risk aversion and state-contingent consumption, a positive risk premium may justify ΔP above the expected indemnity; the size of that justification depends on wealth, background risk, risk preferences, and financing restrictions.
A second constructed exposure involves a 100,000-unit disaster loss with probability 0.01 during the year. Expected loss is 1,000. If a fully collectible policy with no relevant exclusions is offered for 1,400, its actuarial load is 400. Whether the contract is optimal depends on the buyer's risk-bearing capacity and preferences; a risk-neutral household without financing or transaction benefits declines it, whereas a risk-averse household facing devastating loss may rationally purchase it. The severity ratio is economically important: the initial small-risk contract changes a maximum 750-unit exposure, while the disaster contract may protect 100,000. A buyer who pays 150 to avoid a possible 750 yet declines to pay 1,400 against a possible 100,000 creates the appearance of a behavioral paradox, but that appearance is not automatically an identified preference inconsistency.
The two products may be offered in different markets and on different terms. Their periods, claim correlations, actual probabilities, policy exclusions, disaster assistance expectations, liquidity demands, and administrative costs may differ. A formal claim of inconsistency requires an integrated feasible-choice model with common monetary units, budget and time horizon, actual contract attributes, and a single representation of the household's beliefs and preferences. Without those conditions, the comparison is a research stimulus rather than a normative proof.
Expected utility and catastrophic severity
Let W denote liquid-and-realizable wealth relevant to a one-year decision, L a possible financial loss, p its probability, and P the premium for full, certain, immediate indemnity. With a strictly increasing utility function u, no insurance yields expected utility (1-p)u(W) + p u(W-L). Full insurance yields u(W-P). The household is willing to buy when the latter expression is at least the former, assuming no taxes, default, other policy attributes, or intertemporal complications. The certainty-equivalent premium is defined by equality. For increasing concave u, complete insurance can be worth more than pL because it smooths consumption across states.
The comparative statics of this model must be handled with care. A fixed 750-unit retained difference may have negligible utility consequences to a wealthy household, but material consequences to a liquidity-constrained household. A rare 100,000-unit loss can place W-L below a subsistence threshold or force highly costly asset disposal. It may therefore deserve insurance at a loaded premium. Yet a sufficiently high load can also rationally deter insurance for catastrophes. A low-probability peril outside a policy's exclusions might be impossible to insure privately, and a policyholder cannot purchase an absent contract. Classical expected utility provides a useful benchmark for testing affordability and coverage; it does not, without institutional inputs, predict uniform full insurance across every hazard.
The welfare problem becomes more complicated when postdisaster transfers exist. A household may anticipate public relief, family assistance, emergency credit, or debt restructuring, changing the private benefit of a purchased policy even if the social cost of the disaster remains enormous. Government protection can alleviate hardship while also changing private demand incentives. Those channels should be represented explicitly, rather than put into a residual category labeled irrationality.
Probability weighting is conditional on attention
Kahneman and Tversky's original prospect-theory analysis (1979) describes decisions in terms of gains and losses relative to a reference point and decision weights that differ from objective probabilities. Their model was explicitly descriptive. Tversky and Kahneman's cumulative prospect-theory development (1992) extends the framework with rank-dependent weighting for multiple outcomes and distinguishes positive and negative outcomes. Both inform the interpretation of low-probability insurance decisions, but neither licenses assuming that each objectively rare event is always mentally overweighted in practice.
An essential logical observation follows. If a single stable weighting function overweights both a 0.05 small-claim probability and a 0.01 disaster probability, and if both prospects are noticed, properly understood, comparably insured, and evaluated under the same reference-point and contract assumptions, probability overweighting by itself does not generically explain expensive protection for the small event together with neglect of the catastrophe. Such a model may support positive willingness to insure both. The divergent choice requires further explanation, for example different reference points, zero or suppressed subjective catastrophe probability, limited attention, premium framing, a contract supply restriction, or unequal access to cash at the time of purchase. This is an analytical implication of the stated assumptions, not a new behavioral law.
The price of certainty and the timing of money
Insurance premiums are certain costs now, whereas compensation is contingent on an uncertain event in the future. A deductible reduction produces a visible, familiar promise: the customer expects to pay less when a relatively routine claim occurs. A catastrophe contract may require evaluating a distant, poorly imagined loss, with possible claims delays and complicated exclusions. Mental budget categories, payment timing, and uncertainty about performance can therefore affect the two decisions differently.
A rigorous assessment distinguishes current disposable income from lifetime wealth, the annual premium from the present value of multiyear premiums, and physical catastrophe loss from the net financial consequences after other transfers. A large physical loss is not always the same as a large uninsured financial loss for a particular consumer. A homeowner with a high-value property but strong liquid assets faces a different economic exposure than a highly leveraged household with the same house. Similarly, an index-insurance contract may promise a payout triggered by rainfall but fail to pay when the farmer suffers an income shock unrelated to the index. The relevant insured object is the contractual payoff, not the disaster's dramatic description.
Low-Frequency Disaster Underprotection
Historical evidence and the institutional question
Research on disaster insurance has long documented the distance between potential loss and voluntary coverage. The early disaster protection research program combined roughly 3,000 interviews with behavioral and organizational methods concerning flood and earthquake hazards. Its historical importance lies in recognizing that information, hazard perception, insurance distribution, relief expectations, and institution-building interact. Treating a single low purchase rate as pure consumer indifference would discard much of what that program examined.
Kunreuther, Pauly, and McMorrow (2013), in their insurance design discussion, explain why observed demand for some small-loss covers may coexist with limited catastrophe protection. Their work supplies the organizing comparison for this dossier, but the empirical causal composition must be reconstructed from more focused studies. Disaster policies differ across countries and time in law, eligibility, mandatory coverage, risk classification, pricing, distribution, and subsidy. In some areas, the relevant risk is excluded from basic home insurance; in others, catastrophe cover is bundled with standard property protection or actively required by mortgage rules. A household that appears to have decided against coverage may have faced an infeasible purchase set.
Flood-insurance participation and observed event responses
Michel-Kerjan and Kousky's Florida flood-insurance study (2010) draws on more than 7.5 million NFIP policy-in-force records during 2000-2005 and associated Florida claims. The unit is a record of a policy in force, not 7.5 million distinct sampled individuals. Its scale enables contract and market analysis, but administrative take-up estimates still require an appropriate population denominator. A concentration of policy purchases in flood-prone areas cannot by itself separate risk recognition from insurance mandates, lender screening, and differences in insured property values.
Gallagher's 2014 nationwide flood panel finds that purchases increase in the aftermath of a large flood and subsequently decline toward a prior baseline. The reported spillover in communities outside the flood zone but inside the same television market, at about one-third the flooded community's response, suggests that received information and observed nearby experience matter. The study finds the pattern most consistent with learning subject to forgetting or incomplete recall. It does not demonstrate that all changes are caused by an irrational availability heuristic; event experience may also reveal real information about poorly measured hazard probabilities.
The magnitude of today's protection gap depends on definitions. The October 2026 GAO flood-insurance report, using FEMA administrative data, National Association of Insurance Commissioners private-market data, and external property-level flood-risk estimates, reports that 86 percent of assessed high-risk properties lacked NFIP coverage as of April 2026. The report estimates insurance participation at about 43 percent inside FEMA special flood hazard areas and 2 percent outside those areas, and identifies significant high-risk exposures beyond the mapped mandatory-purchase perimeter. These statistics should not be mechanically combined: the 'high-risk' population defined by external risk estimates and the statutory FEMA mapped zone are different denominators, and NFIP nonparticipation does not prove the absence of all private flood policies.
This denominator discipline affects the scientific claim. An observed uninsured structure might be located outside a compulsory zone, have private insurance, lack an eligible lender or homeowner decision maker, or face a premium that exceeds feasible liquidity. Even a correctly measured uninsured property may reflect a landlord's choice, a tenant's limited insurable interest, a policy discontinuance, or coverage supplied through another instrument. Proper behavioral identification begins after these institutional and contractual alternatives are made observable.
The consequences of underprotection
Large insured and uninsured catastrophes can have very different recovery trajectories. Kousky's 2019 disaster-insurance review finds evidence that insurance improves recovery outcomes, while calling attention to the limitations of rigorous empirical evaluations of risk-reduction incentives. Thieken and coauthors' German flood household study (2006), following the severe 2002 flood, reports earlier compensation for insured households and modest differences in awareness and mitigation. Such associations are important for understanding vulnerability, but insured and uninsured households can differ in resources, location, and preparedness before the flood.
Financial recovery capacity extends beyond indemnity. Households can repair from savings, receive aid, borrow, sell assets, reduce consumption, or leave an area. Those methods differ in timing, redistribution, and welfare consequences. A policy's value may arise as much from the speed and reliability of post-event funds as from the expected monetary transfer. Catastrophe risk management should therefore be assessed through net financial resilience and actual payout performance rather than a binary policy indicator alone.
Why population-level underinsurance does not finish the argument
The evidence of substantial flood protection gaps establishes a public-policy problem and a plausible second component of the behavioral paradox. It does not show that individuals who selected expensive low-deductible automobile or health policies are personally those who left flood risk uninsured. Nor can a flood take-up rate be directly compared to a homeowners deductible share without matching ownership, peril exposure, time, contract availability, and coverage status. A scientifically defensible paradox must preserve this distinction even as it develops hypotheses about the behavior of people who face both menus simultaneously.
Probability Weighting, Attention, and Learning
What experimental evidence does and does not identify
The role of low-probability decision weighting is experimentally plausible but empirically heterogeneous. McClelland, Schulze, and Coursey's low-probability insurance experiments (1993) examined real-money insurance decisions with repeated experience. For a loss probability of 0.01, bids showed a notably bimodal distribution, with some participants bidding nothing and others bidding considerably above expected indemnity. The bimodality persisted in the second experiment with larger losses and repeated rounds. This result demonstrates the possibility of polarized response to rare events rather than a single representative overweighting or underweighting coefficient.
The relevance to catastrophe markets is suggestive and limited. A laboratory participant's repeated choices for a specified small monetary loss are not the same as a homeowner's multiyear purchase of legally complex flood coverage, and an experimentally disclosed probability may differ substantially from the probability the homeowner can reasonably know. But bimodality helps explain why two people exposed to similar rare events might adopt apparently opposite insurance positions. It also implies that an average willingness to pay can conceal distinct behavioral groups and should not be presented as the individual decision pattern.
Johnson, Hershey, Meszaros, and Kunreuther's 1993 insurance-framing experiments documented changes in hypothetical purchase preferences from altered framing and probability presentation. Such experiments isolate causal effects of the presented decision description more cleanly than field correlations, although they need external validation for costly real policies. They indicate that insurance demand can depend on how a loss, premium, and promised payment are described. Their implications for choice architecture must be tested against ethical standards: providing truthful comparisons improves decision quality, while exploiting fear or ambiguity can make demand more expensive without improving protection.
Stable preferences versus unstable exposure representations
Insurance choices depend on the objective loss distribution, what the buyer believes, and which aspects of that distribution enter the active decision. A person may report concern about earthquakes in a survey yet omit earthquake insurance from the next budget. Another may misunderstand a nonflood homeowners contract as providing broad disaster protection. These are distinct mechanisms: subjective probability error, restricted attention, contract miscomprehension, and choice-set unavailability. Each has a different measurement signature and a different feasible remedy.
Barseghyan and colleagues' 2013 deductible model offers evidence for probability distortions in actual insurance purchases. Their 2018 field-estimation review surveys identification problems when risk preferences are inferred from observed decisions. The main interpretive challenge is observational equivalence. A consumer's high price paid for protection can be fit by strong utility curvature, high subjective claim probability, probability weighting, fear of deductible payment, or information and intermediary frictions. In a single binary choice, these parameters generally cannot all be separately recovered without restrictions or added observations.
The insured and uninsured catastrophe states can be represented as separate internal decision tasks. The consumer might receive a home-insurance renewal quote in a familiar format, understand its deductible, and accept the default, while never receiving a plainly priced flood insurance offer. Under that information environment, explaining divergence exclusively through preferences would confuse what was evaluated with what was available for evaluation. Modeling conditional attention is therefore often as important as selecting a functional form of risk utility.
Availability, learning, forgetting, and repeated experience
Rare disasters supply bursts of information followed by long periods during which the event becomes less prominent. Gallagher's flood learning evidence allows both real Bayesian updating and imperfect information retention to be considered. A rise in purchases after a flood is not inherently excessive: a flood may reveal an actual hazard not previously recognized. A subsequent decline may reflect rational revision, changes in risk prices, mortgage turnover, forgetting, or reduced attention. Distinguishing these requires direct information about perceived probability, exposure, and contract offerings at each point in time.
Turner, Said, and Afzal's Pakistan flood microinsurance study (2014) compares 384 participants from affected and comparison villages after the severe 2010 floods, combining surveys and framed experiments. The authors report higher insurance demand among those affected, with both direct and observed loss experience associated with demand after controls. This suggests that experience matters outside wealthy insurance markets, although the matched-village experimental context does not provide a universal effect size for real annual purchases.
The combination of these studies supports a nuanced temporal picture. Rare losses may be under-attended before an event, receive concentrated attention immediately afterward, and decline in salience as time passes. The size and duration of that cycle depend on media exposure, hazard maps, premium changes, past experience, liquidity, and communication. A credible policy can maintain access to clear hazard information without constructing artificial alarm, using regular risk disclosures and concrete coverage illustrations rather than a one-time emotionally amplified campaign.
Why a single weighting parameter is insufficient
Consider two rare losses with probabilities p_s and p_c, where both are small but the catastrophe is much more severe. Standard inverse-S probability weighting increases the relative decision weight of small positive probabilities, potentially making the smaller risk appear worth insuring. If the same weighting curve is applied consistently to both states, the catastrophe's psychological evaluation depends additionally on loss severity and reference-dependent valuation. Consequently the mere presence of small-probability overweighting in deductible data cannot logically entail neglect of a similarly rare larger insured loss. A satisfactory explanation needs source-specific beliefs, choices made under different frames, restricted opportunities, or other independently identified preference components.
A further complication is ambiguity over catastrophe frequency. Insurers use hazard models and geographical exposure maps, while households may not possess comparable location-specific information. A known 5 percent ordinary-claim probability is not behaviorally equivalent to a subjectively unknown 1 percent flood probability, even if their objective rates are correctly estimated. Ambiguity preferences and reliance on expert information may change demand independently of the probability weighting inferred from ordinary deductible choices. The appropriate evidence base therefore includes elicited beliefs, objective and perceived hazard information, and the details of how probabilities were communicated.
Liquidity, Trust, Affordability, and Contract Design
Cash-flow constraints and household insurance budgets
The difference between willingness to pay in a hypothetical elicitation and actual purchase often reflects available cash rather than a preference for exposure. A household can perceive a disaster as devastating and still be unable to advance the premium without reducing essential spending. Insurance against a rare large event may be especially challenging when premiums are annual and financing is constrained, while a lower deductible may be bundled into a mortgage-required or employer-assisted product. Comparing contracts solely by expected loss can thus miss the affordability constraint that determines feasible demand.
In rural India, Cole and colleagues' randomized rainfall-insurance experiments (2013) identify strong price sensitivity and evidence for liquidity constraints, limited trust, and salience as barriers to enrollment. The authors emphasize that high take-up would remain elusive even with favorable payout ratios. The inference is particularly relevant because demand does not become high merely when an economist identifies actuarial value: the consumer must believe the payout will arrive, understand the trigger, and find a way to finance the premium.
Giné, Townsend, and Vickery's rainfall insurance participation study (2008) reports lower take-up with greater basis risk and credit constraints, higher take-up with wealth, and strong relationships with social networks and vendor familiarity. Strikingly, their observed measures of risk aversion did not produce the simple predicted positive insurance-demand association. The evidence illustrates why broad preferences cannot be read directly from purchases when contract quality, credit access, social information, and trust differ across households.
Basis risk and the effective insurance promise
A catastrophe or rainfall-index policy may not indemnify the actual household loss. Basis risk arises when the insured trigger or index fails to match the relevant income or physical damage. A severe drought can reduce agricultural income even where a contractual weather index does not cross its threshold; an index payout may also occur without major loss. The buyer is therefore choosing a contingent financial instrument with residual risk, not purchasing perfect insulation from the physical hazard. This is a rational reason for lower take-up in some environments.
Contractual complexity is especially consequential in catastrophic insurance. Coinsurance, deductibles, disaster-specific sublimits, waiting periods, eligibility maps, and proof-of-loss requirements make the nominal sum insured an incomplete measure of protection. A low-cost, small-risk product that pays predictably may be more attractive than a broad-sounding catastrophe product with uncertain performance. Trust in the provider and in enforcement institutions has economic value. Where claim payment is delayed or disputed, the effective insurance promise is discounted by recovery costs and timing; these frictions cannot automatically be classified as psychology.
The World Bank disaster risk finance overview emphasizes constraints arising from weak data, institutional capacity, risk pricing, affordability, and changing hazard exposure. It describes layered public and private financing, including reserves for frequent losses and risk transfer for severe events. This perspective reveals why the consumer-level paradox also has a supply-side dimension: catastrophic risk may be highly correlated across insured properties, challenging diversification and making affordable long-tail coverage difficult to supply without reinsurance, capital markets, or public participation.
Premium burden, distribution, and the price of access
Affordable catastrophe protection is not equivalent to actuarially underpriced protection. A peril's annual expected loss might be small in probability terms but economically substantial if asset exposure is high and reinsurance, claims adjustment, and capital costs are large. Subsidized premiums may expand coverage, but they transfer cost to taxpayers or other policyholders and may alter location and mitigation incentives. Premium support and risk-reduction requirements must therefore be evaluated jointly, with attention to incidence and distributional effects.
Distribution arrangements matter as well. A policyholder may find the standard homeowners renewal convenient because it is presented at the same moment as mortgage or escrow payment, while the additional catastrophe policy requires searching for an unfamiliar vendor. A generic comparison of uptake by loss size would confound decision cost with valuation of loss size. A useful test varies the presentation of catastrophe cover while preserving contractual terms and measures whether take-up changes, ideally randomizing both information and purchase assistance.
Trust and legal enforceability across jurisdictions
The same psychological narrative does not transport automatically from US homeowners to Indian rainfall microinsurance, German property insurance, or a disaster-finance program with public assistance. Product reliability, contract enforcement, regulatory supervision, and availability vary. The US NFIP combines a public insurance mechanism with mortgage-linked mandatory purchase in mapped hazard areas; private German insurers have offered flood supplements with regionally different penetration; index contracts can face performance uncertainty unrelated to legal indemnity. These differences alter both the cost and the meaning of purchasing insurance.
A comprehensive model includes legal eligibility, claims reliability, contract terms, and social protection alongside attention and preferences. A low catastrophe take-up rate may be simultaneously affected by beliefs and nonbehavioral barriers. The relative contribution of each factor must be established with experiments, linked microdata, or well-supported institutional counterfactuals. Explanations that attribute every uncovered disaster loss to a single cognitive bias lack this identification.
Comparative Evidence Across Insurance Lines and Countries
United States: deductibles, medical plans, and flood exposure
The US evidence has unusually strong but fragmented empirical components. Homeowners and automobile deductible studies examine revealed preferences within ordinary personal insurance. Employer health-plan administrative records permit more demanding tests of financial dominance. Flood insurance studies examine participation, event response, geographic regulation, and disaster exposure. Each component supports a different proposition: expensive deductible protection is chosen; some lower-deductible plans are dominated; catastrophe insurance participation is incomplete; and learning following major floods is dynamic. Their combination motivates the paradox but is not a longitudinal observation of the same consumer choosing all three products.
The 2026 GAO flood coverage analysis provides a particularly timely illustration of why market structure and legal geography must be considered. Its finding that about 43 percent of properties in FEMA special flood hazard areas had flood insurance, compared with about 2 percent outside those areas, is consistent with a substantial role for mandatory purchase rules. The report's assessment that approximately 13 million externally identified high-risk properties lie outside those FEMA special flood hazard areas further demonstrates why official maps and modeled risk exposures should not be treated as interchangeable. A risk map changes the universe of properties considered high-risk; a mortgage mandate changes the probability that a household purchases coverage.
Even within the same jurisdiction, a consumer's purchase can be the outcome of an intermediary's presentation, a mortgage covenant, an employee benefit plan, and an individual choice at different stages. Compulsory standard property insurance or employer-sponsored medical coverage can supply a convenient menu of small-loss options, whereas a supplemental flood product may require a separate action. Observed intensity of protection therefore captures the institutional design of offers and defaults as well as household preferences.
Germany: peril-specific coverage and geographic differences
Germany provides a complementary comparison because the difference between ordinary storm coverage and flood or extreme-rainfall coverage has been documented directly. The OECD's Germany environmental review (2023) reports that approximately 94 percent of residential buildings were insured against storm and hail, while roughly 45 percent were insured against flood and extreme precipitation in the cited reference period. The OECD also records large regional variation in flood coverage, reflecting the significance of institutions, historical conditions, and local risk.
These figures compare coverage of different perils in a national property market rather than exact deductible decisions for the same owner. Storm and hail losses are not necessarily 'small' events in severity, and flood losses are not always catastrophic. The comparative value lies in demonstrating that a high rate of insurance against one familiar property peril can coexist with much lower coverage against another, often severe peril. A general conclusion requires caution about product availability, local underwriting, pricing, and differences in the underlying hazard distributions.
The 2006 German flood study emphasizes insurance eligibility and postevent compensation after the 2002 floods, reinforcing the role of contract availability and recovery institutions. Its findings should be read alongside the much later OECD national figures without implying that the samples, time periods, and denominators are identical. Historical comparison is informative when the underlying definitions are stated, but changes over time do not establish causal effects of any single public measure.
India and Pakistan: inclusive insurance and catastrophe learning
The India evidence places cash, product quality, and local distribution at the center of the decision. The 2008 rainfall insurance study and the 2013 randomized field experiments examine a class of contracts in which insured payouts need not match individual realized losses. Their findings identify the potential importance of basis risk, trust, liquidity, network familiarity, price sensitivity, and limited salience. The scientific implication is that the gap between catastrophic exposure and coverage can be materially structural, even when the consumer comprehends that weather risk is severe.
The Pakistan postflood microinsurance experiment adds a behavioral learning dimension. Flood experience and observed losses were associated with greater demand in the reported sample, consistent with experience increasing subjective relevance. Yet the product and welfare question remain separate. Higher willingness to buy after a flood may reflect better hazard information or stronger attention, while actual enrollment may still be constrained by premium, trust, and contract usefulness. A policy that teaches risk more effectively but leaves premiums unaffordable may not appreciably close the protection gap.
These studies challenge the presumption that the same mechanism dominates across income groups. In one setting, a consumer can spend too much because health-plan terms are misunderstood; in another, a household can remain exposed because the relevant catastrophe contract is untrusted, imperfectly matched to loss, or cash-flow infeasible. A global research synthesis must allow the mixture of mechanisms to vary and resist importing a parameter estimated in a US deductible market into a rural index-insurance population.
Aggregate comparisons and their denominators
The comparative evidence has four distinct statistical units: policies and claims; individuals choosing from a known menu; properties located in a modeled risk area; and households responding to surveys or experiments. These cannot be pooled as if each were one household observed over a common year. A property may be owned by a business or landlord, one household may hold multiple policies, a policy may renew repeatedly, and a farmer may insure an income index rather than a building. Both the numerator and denominator must match the question being answered.
For an observed insurance penetration statistic, the investigator must define whether it is the share of all households, exposed households, insurable households, eligible policies, or properties above a risk threshold. For a deductible selection statistic, the denominator is customers with a particular insurer and menu. A defensible comparison of the paradox requires measuring each household's effective coverage against both small and severe losses conditional on feasible contract options, rather than juxtaposing two national percentages.
Empirical Identification and Measurement
The within-household test
The strongest empirical test would link the same households across insurance lines and time, observe their assets and liabilities, and reconstruct the contracts actually offered. For each household h, period t, and peril k, the researcher would need objective hazard estimates, subjective beliefs, potential loss distributions, premium menus, deductibles, excluded losses, contract limits, insurance claims, household liquid wealth, and relevant government assistance. The source of the menu is important: observing purchased cover alone does not reveal whether a protective alternative was available and affordable.
One descriptive statistic could compare the incremental premium loading paid to reduce modest retained losses with protection against a large uninsured loss, both expressed as functions of annual disposable resources. A second statistic could measure the fraction of a household's catastrophe exposure that remains net of insurance and credible transfers. These are different objects; neither should be compressed into a single unvalidated 'rationality score.' A within-household association would show co-occurrence, while causal explanation would still require exogenous variation in information, price, contract offers, or default positions.
A decisive empirical result would show that the same buyer knowingly accepts a costly low-loss policy while declining a feasible, adequately performing catastrophe policy whose welfare value is greater under an explicitly specified preference and wealth model. Such evidence would be stronger than comparing homeowners deductible data and flood insurance statistics drawn from different people. It would also require checking whether the alleged catastrophe policy was excluded, unavailable, or otherwise materially different from the modeled promise.
Identification hierarchy
Administrative insurance records measure selected contracts, renewal, and claims accurately where reporting is complete, yet cannot by themselves identify the customer's full beliefs or unoffered alternatives. Elicitation experiments can randomize wording, probabilities, and framing, but hypothetical choices may differ from decisions involving actual, substantial premiums. Field randomized experiments can estimate local effects of price, information, endorsement, and assistance, with limits from sample, product, and institutional context. Event studies such as Gallagher (2014) provide a temporal response to flooding, but require assumptions about contemporaneous prices, geographic sorting, and changing risk.
The field risk-preference methodology review by Barseghyan and coauthors (2018) reinforces the importance of restrictions for identifying risk preferences from market data. In the present problem, the relevant identification threats include unobserved heterogeneity, omitted contract attributes, endogenous risk information, selection into insurance markets, defaults, differential search, and correlation between disaster exposure and income. Results from a selected insured sample cannot automatically generalize to households never offered a product.
Researchers should also distinguish objective probabilities from modeled probabilities. Flood maps and catastrophe models incorporate uncertainties about climate, terrain, exposure, engineering, and data quality. Perceived probabilities are elicited using questions whose framing and numerical comprehension may influence answers. A gap between reported beliefs and modeled estimates is evidence of disagreement under a specified metric; it becomes evidence of systematic misperception only after the model's uncertainty and question design are addressed.
Specific falsifiable explanations
The behavioral-attention account predicts that otherwise comparable households increase catastrophe uptake when accurate risk and product information is delivered at a salient decision point, with premium, eligibility, and liquidity held constant. The liquidity account predicts that installment payment, liquidity support, or premium scheduling may increase uptake even without substantial change in reported hazard belief. The comprehension account predicts that simplified state-by-state contract comparisons reduce financially dominated small-loss choices. The trust account predicts that independently verifiable claims performance, credible oversight, or known provider endorsement changes demand more strongly where insurer credibility is low.
The institutional-choice-set account predicts that differences in bundling, default inclusion, mortgage requirements, and accessible offers explain part of the protection asymmetry even after accounting for subjective risk beliefs. The basis-risk account predicts that improved alignment of contract payouts with actual household loss increases the value and demand for index insurance. Each prediction requires an appropriately controlled test and can produce different outcomes across populations. They should not be collapsed into one umbrella causal effect.
A robust empirical design would pre-register the contract and risk definitions; use comparable annual horizons; report both take-up and effective financial protection; include placebo peril and timing tests; examine heterogeneity by wealth and exposure; and separately measure price, eligibility, and plan comprehension. It would follow insured and uninsured households through realized catastrophes when feasible, because the benefits of a policy depend on actual performance and recovery as well as purchase. Distributional incidence, insurer failure, and claims disputes must enter any full welfare analysis.
Welfare Analysis and Policy Implications
The normative evaluation problem
A household's choice can be descriptively unexpected without proving that it lowers welfare. An expensive incremental low deductible can buy liquidity relief, convenience, or confidence; a catastrophe policy can carry real price loading, exclusions, basis risk, or doubtful payment. The first task of welfare analysis is to construct the actual state-contingent financial consequences. The second is to define whose welfare counts and how preferences informed by incomplete understanding are treated. Only then is it defensible to characterize a particular allocation of insurance as inefficient.
Dominated health-plan choices are unusually informative because they permit direct comparisons of total costs across all possible spending states under otherwise matching benefits. By contrast, an uninsured flood exposure may be the optimal private response to a severely constrained or overpriced market. Welfare judgments that ignore the feasibility set risk attributing supply failure to consumer psychology. Both failures may deserve institutional remedies, but their policy targets differ.
The 2026 GAO assessment of flood insurance purchase requirements points to the role of maps, mortgage-linked obligations, and information. It offers recommendations to Congress regarding risk identification, disclosure, and program coverage. These are jurisdiction-specific proposals whose public-finance, distributional, and implementation consequences require separate assessment. They cannot be treated as a universally portable insurance mandate.
Insurance literacy and transparent comparison
A practical lesson of the health plan-choice evidence is that consumers benefit from comparisons showing annual premium, the realized total financial cost in representative states, maximum possible out-of-pocket exposure, and exclusions. The goal of disclosure should be to make the same decision problem comprehensible across all offered contracts. A low deductible should never be marketed as inherently better protection without showing the incremental price and the states in which it pays.
Catastrophe communication should similarly identify the peril, the source and range of risk estimates, whether the existing policy covers it, expected financial consequences, the premium, limits, deductible, and claim triggers. A visual flood-risk indicator can orient attention, but an intelligible account of contract performance remains necessary. The welfare objective is informed choice and feasible protection, not maximization of sales for any particular product.
Affordability, access, and long-tail risk finance
The World Bank disaster risk finance perspective supports layering instruments according to loss size and economic efficiency. Households can retain manageable frequent losses where liquidity permits, while severe correlated exposures may require insurance, reinsurance, contingency resources, or public financial mechanisms. The design challenge is to make catastrophe protection available and understandable without encouraging excessive location risk or removing incentives to mitigate losses.
Premium affordability programs require explicit attention to distribution. Lower-income households may derive substantial welfare value from catastrophe coverage while being unable to finance annual premiums. Subsidies or automatic enrollment may change take-up, but their benefits must be weighed against budget cost, targeting, insurer pricing incentives, and land-use effects. Where catastrophe risk is increasing or becoming difficult to model, straightforward information alone cannot ensure affordable private capacity. Research on insurance supply and public-private risk finance belongs alongside the behavioral literature.
The 2019 natural disaster insurance review documents evidence of improved recovery with coverage while stressing the need for stronger studies on risk reduction. Policymakers should therefore avoid asserting that higher insurance take-up alone will always produce greater preventive investment. Incentives to reduce damage depend on deductibles, risk-rated premiums, inspections, building standards, and the expected availability of disaster assistance.
Guardrails against manipulative design
Behavioral insurance findings can be used either to improve decisions or to exploit predictable confusion. A short default enrollment process may expand essential protection, but only if terms, premiums, and withdrawal rights remain conspicuous. Framing a low deductible as 'peace of mind' while obscuring its high incremental cost is inconsistent with informed choice. Conversely, frightening catastrophe imagery without reliable personalized risk information can exaggerate vulnerability and encourage misallocation of limited funds.
Institutional recommendations should therefore be assessed against independent standards of transparent disclosure, contract comparability, affordable access, effective claim settlement, and documented consumer outcomes. Randomized communication trials should measure the cost of resulting decisions, persistence through renewal, and actual recovery protection. These criteria are valid across diverse markets even where precise legal duties differ.
Integrated Analytical Synthesis
The literature supports a layered interpretation of the behavioral insurance paradox. At the level of observed markets, some people purchase costly protection against modest risks and many properties remain exposed to catastrophic hazards. At the level of behavioral explanation, probability weighting, attention, reference points, learning, and comprehension can alter the value assigned to insured states. At the level of feasibility, liquidity, contract availability, trust, regulation, basis risk, and insurer capacity determine which choices can actually be made. A comprehensive explanation requires the interaction of all three levels rather than a universal coefficient attached to 'human irrationality.'
The strongest contribution of the comparative synthesis is methodological. Small-loss purchases and catastrophe gaps are facts drawn from separate empirical structures, and the joint-household claim must be identified independently. Existing evidence makes the phenomenon scientifically credible as a cross-market regularity, while leaving open the prevalence and causal composition of its within-person form. An apparent paradox can also dissolve when excluded perils, cost-sharing, liquidity, mandated coverage, or the timing of premiums are incorporated into the actual choice set.
The literature further implies a constraint on psychological explanations. A stable overweighting of small probabilities can contribute to expensive low-deductible protection, but it cannot by itself predict that the same attentive consumer ignores a comparably rare severe exposure under identical terms and beliefs. Divergent choices require additional assumptions about attention, reference points, beliefs, price, product quality, or feasibility. This conclusion follows from comparative model logic and the reviewed experiments; it is presented as an analytical inference, not a newly proposed theory or empirical discovery.
Different interventions consequently address different margins. Contract comparison tools target comprehension and dominance; transparent hazard disclosures target information; installment options and targeted assistance address liquidity; credible claims administration addresses trust; index redesign reduces basis risk; and public-private catastrophe arrangements address systemic capacity. These instruments should be evaluated for actual risk transfer, household recovery, and distributional welfare, not simply their ability to increase an aggregate sales metric.
Conclusion
The behavioral insurance paradox of small and catastrophic losses is a rigorous research question about the allocation of financial protection across the scale and frequency of losses. It is supported by concrete but differently structured evidence: homeowners deductible choices, dominated health plans, field-estimated decision weights, flood-insurance protection gaps, postdisaster learning, and insurance access constraints in several countries. These sources establish the ingredients of the phenomenon, and they show that no single behavioral factor uniformly accounts for them.
Three conclusions govern its interpretation. First, identifying small-risk overprotection requires comparing incremental premiums with the actual state-dependent financial benefit; identifying catastrophe underprotection requires measuring the remaining severe exposure and the genuinely feasible coverage set. Second, cross-market coexistence does not prove that the same households exhibit both choices, nor that every instance is a preference error. Third, the policy response should integrate trustworthy, comprehensible insurance choices with affordable and effective catastrophe finance, because consumer behavior operates inside legal, economic, and institutional structures.
Future research should join household-level insurance menus across products, estimate subjective and objective hazard beliefs, distinguish defaults and attention from premiums and constraints, and follow actual recovery after losses. The scientific value of the paradox lies in directing inquiry toward those identifiable mechanisms and measurable welfare consequences. The present monograph offers an original source-critical integration and explicit conditional deductions without asserting a new theory, an original causal estimate, or completed external peer review.
References
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. https://doi.org/10.1257/aer.103.6.2499.
Barseghyan, Levon, Francesca Molinari, Ted O'Donoghue, and Joshua C. Teitelbaum. 2018. "Estimating Risk Preferences in the Field." Journal of Economic Literature 56 (2): 501-564. https://doi.org/10.1257/jel.20161148.
Bhargava, Saurabh, George Loewenstein, and Justin Sydnor. 2017. "Choose to Lose: Health Plan Choices from a Menu with Dominated Option." Quarterly Journal of Economics 132 (3): 1319-1372. https://doi.org/10.1093/qje/qjx011.
Cohen, Alma, and Liran Einav. 2007. "Estimating Risk Preferences from Deductible Choice." American Economic Review 97 (3): 745-788. https://doi.org/10.1257/aer.97.3.745.
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.
Gallagher, Justin. 2014. "Learning about an Infrequent Event: Evidence from Flood Insurance Take-Up in the United States." American Economic Journal: Applied Economics 6 (3): 206-233. https://doi.org/10.1257/app.6.3.206.
Giné, Xavier, Robert Townsend, and James Vickery. 2008. "Patterns of Rainfall Insurance Participation in Rural India." World Bank Economic Review 22 (3): 539-566. https://doi.org/10.1093/wber/lhn015.
Handel, Benjamin R., and Jonathan T. Kolstad. 2015. "Health Insurance for 'Humans': Information Frictions, Plan Choice, and Consumer Welfare." American Economic Review 105 (8): 2449-2500. https://doi.org/10.1257/aer.20131126.
Johnson, Eric J., John Hershey, Jacqueline Meszaros, and Howard Kunreuther. 1993. "Framing, Probability Distortions, and Insurance Decisions." Journal of Risk and Uncertainty 7 (1): 35-51. https://business.columbia.edu/faculty/research/framing-probability-distortions-and-insurance-decisions.
Kahneman, Daniel, and Amos Tversky. 1979. "Prospect Theory: An Analysis of Decision under Risk." Econometrica 47 (2): 263-291. https://doi.org/10.2307/1914185.
Kousky, Carolyn. 2019. "The Role of Natural Disaster Insurance in Recovery and Risk Reduction." Annual Review of Resource Economics 11: 399-418. https://doi.org/10.1146/annurev-resource-100518-094028.
Kunreuther, Howard C., with Ralph Ginsberg, Louis Miller, Philip Sagi, and Paul Slovic. 1978. Disaster Insurance Protection: Public Policy Lessons. New York: Wiley. Research-catalog record: https://nehrpsearch.nist.gov/article/PB-289%20950/8/XAB.
Kunreuther, Howard C., Mark V. Pauly, and Stacey McMorrow. 2013. Insurance and Behavioral Economics: Improving Decisions in the Most Misunderstood Industry. Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9781139050319.
McClelland, Gary H., William D. Schulze, and Don L. Coursey. 1993. "Insurance for Low-Probability Hazards: A Bimodal Response to Unlikely Events." Journal of Risk and Uncertainty 7 (1): 95-116. https://ideas.repec.org/a/kap/jrisku/v7y1993i1p95-116.html.
Michel-Kerjan, Erwann O., and Carolyn Kousky. 2010. "Come Rain or Shine: Evidence on Flood Insurance Purchases in Florida." Journal of Risk and Insurance 77 (2): 369-397. https://doi.org/10.1111/j.1539-6975.2009.01349.x.
Organisation for Economic Co-operation and Development (OECD). 2023. OECD Environmental Performance Reviews: Germany 2023. Paris: OECD Publishing, section "Climate Change Adaptation and Nature-Based Solutions." https://www.oecd.org/en/publications/oecd-environmental-performance-reviews-germany-2023_f26da7da-en/full-report/component-7.html.
Sydnor, Justin. 2010. "(Over)insuring Modest Risks." American Economic Journal: Applied Economics 2 (4): 177-199. https://doi.org/10.1257/app.2.4.177.
Thieken, Annegret H., Theresia Petrow, Heidi Kreibich, and Bruno Merz. 2006. "Insurability and Mitigation of Flood Losses in Private Households in Germany." Risk Analysis 26 (2): 383-395. https://doi.org/10.1111/j.1539-6924.2006.00741.x.
Turner, Ginger, Farah Said, and Uzma Afzal. 2014. "Microinsurance Demand After a Rare Flood Event: Evidence from a Field Experiment in Pakistan." Geneva Papers on Risk and Insurance: Issues and Practice 39 (2): 201-223. https://ideas.repec.org/a/pal/gpprii/v39y2014i2p201-223.html.
Tversky, Amos, and Daniel Kahneman. 1992. "Advances in Prospect Theory: Cumulative Representation of Uncertainty." Journal of Risk and Uncertainty 5 (4): 297-323. https://doi.org/10.1007/BF00122574.
United States Government Accountability Office (GAO). 2026. Flood Insurance: Congressional Action Could Help Increase Coverage for At-Risk Properties. GAO-27-108012. Washington, DC: US Government Accountability Office, October 1. https://www.gao.gov/products/gao-27-108012.
World Bank Group. n.d. "Disaster Risk Finance and Insurance." World Bank Group, accessed November 15, 2026. https://www.worldbank.org/ext/en/topic/financial-sector/disaster-risk-finance-and-insurance.