RESEARCH DOSSIER
Disaster Memory Decay and Renewal of Coverage
Abstract. Catastrophe experience changes the information households possess, the risks they perceive, and the institutions governing their insurance choices. Across flood-insurance studies, disaster exposure often precedes a temporary rise in purchases, while the durability of that response differs markedly by geographic risk classification, previous experience, market rules, and public assistance. An observed decline in aggregate policies following an event is sometimes described as disaster-memory decay. Yet an aggregate decline in take-up, a household's declining subjective probability of loss, and the lapse of an existing contract are three separate outcomes. This monograph examines the causal relationships among those outcomes and specifies which claims the existing evidence can establish. It critically evaluates U.S. flood-insurance panel studies, administrative policy-tenure research, property-price responses, longitudinal German mitigation evidence, behavioral experiments and surveys, and contemporary institutional experience in Japan and the United Kingdom. Findings from studies of the 2001–2009 U.S. National Flood Insurance Program portfolio, Georgia counties, Atlantic and Gulf Coast hurricanes, and recently flooded property zones establish that a single universal post-disaster decay duration is empirically indefensible. The relevant mechanisms include learning and forgetting, salience, price and liquidity constraints, mortgage and disaster-aid requirements, coverage design, migration, adverse selection, and insurer-side decisions. The study derives a research-ready measurement and identification agenda separating new purchases, continuous renewal, switching, reinstatement, coverage limits, and effective protection. It concludes that disaster memory is consequential when it changes decisions at actual insurance choice points, while durable protection additionally depends on accurate risk information, affordable and reliable contracts, administrative continuity, and exposure reduction. These conclusions constitute an original integrative analysis of existing scholarship, not a newly named theory or doctrine.
Keywords: Disaster memory; risk perception; flood insurance; coverage renewal; lapse; policy tenure; salience; learning; forgetting; natural catastrophe; event study; difference-in-differences; survival analysis; National Flood Insurance Program; insurance demand; insurance affordability; adverse selection; repeated hazards; risk communication; disaster assistance; property insurance; resilience.
Introduction
An inundated neighborhood can make the possibility of another flood vivid at once. Insurance purchases may rise during the following year; several years later, the observed number of new policies may return toward the pre-event level. This sequence raises a substantive scientific question: how far does a disaster change what people know about hazard, how long does the change influence their judgments, and under which conditions does it survive a policy-renewal decision? The issue matters because an annual insurance contract requires a decision repeatedly, while the underlying exposure may persist for decades and may itself change through development, mitigation, climate conditions, and infrastructure.
Gallagher's (2014) national flood-insurance event study supplies influential evidence that U.S. flood-insurance take-up rises after major floods and then decays toward baseline. Atreya, Ferreira, and Michel-Kerjan (2015) report a three-year disappearance of a recent-flood association in a Georgia county panel. Michel-Kerjan, Lemoyne de Forges, and Kunreuther (2012) instead analyze the tenure of individual policies and find a median of two to four years among new NFIP policies in 2001–2009. Those results address related but nonidentical statistical objects. Their similarities are worth investigating; their differences are essential to explanation.
Later results put any simple expiration account under pressure. Petkov and Ortega (2025) document persistent take-up outside mapped high-risk zones following a major flood, even as the response inside the mapped zone faded after three years. Kousky's (2017) hurricane and flood-insurance analysis shows that disaster-aid conditions explain much of the post-hurricane purchase response. The observed outcome can therefore depend on geographic surprise and a formal purchase requirement as much as on private recollection.
This dossier treats renewal as the decisive link between psychological response and lasting financial protection. It examines how a loss event enters perception, reaches the household's information and budget constraints, and becomes a purchase, renewal, modification, transfer, or lapse under specific legal and market circumstances. Its international horizon is comparative: national institutions are studied on their own terms; estimated effects from American flood policies are not silently imposed on earthquake insurance in Japan or household flood reinsurance in Britain.
Scientific Object, Scope, and Evidentiary Method
The object is the persistence and fading of perceived natural-hazard risk after an event and its relation to the continued availability and maintenance of property-risk insurance. An event may be a local physical loss, a near miss, a highly publicized catastrophe elsewhere, or repeated exposure. The insurance behaviors examined are: first-time purchase, subsequent renewal, coverage adjustment, voluntary cancellation, failure to pay, insurer nonrenewal, switching, and restoration after an interruption. These outcomes are measured separately because they produce different evidence about household decisions and market supply.
The unit of explanation is a temporally situated choice by an identifiable risk-bearing household or insured property. The units of observation in the literature vary. Some papers track state- or county-level insurance policies and flood events; some follow policies or properties; others use household surveys or property transactions. A county panel estimates a change in a population rate. A contract-level longitudinal dataset measures continuity of a specified policy or insured property. A survey estimates associations with beliefs and stated intentions. A property-price study measures capitalization of perceived and expected losses through a market equilibrium. None can be substituted for another without stating the identification assumptions.
The dossier uses source-critical comparative synthesis. Peer-reviewed empirical articles receive priority for causal and statistical claims, with their reported samples, periods, and identification strategies identified explicitly. Government and program records establish operative contract rules, administrative data definitions, contemporary regulatory context, and reported coverage patterns. Institutional and industry documents are used as evidence of their authors' estimates or program structures. The account distinguishes directly reported estimates from interpretations offered here. In particular, the GAO's October 2026 flood-insurance report provides a contemporary national market description, while the earlier academic studies supply the relevant historical identification designs. The GAO report cannot retroactively establish a household memory trajectory during a historical panel.
Two interpretive rules govern quantitative comparisons. First, a conditional insurance-demand change is not automatically an annual renewal hazard: the former depends on all entry and exit flows. Second, different durations such as three years, four to nine years, and an eventual return to baseline represent distinct estimands, samples, models, and outcome variables. A research conclusion can compare them only after naming these differences. No original microdata are claimed to have been estimated for this monograph. Its methodological equations specify testable analyses, not fitted empirical results.
Disaster Event Memory: Conceptual and Research History
A disaster may influence action through personal recollection of danger, retained information about property exposure, social communication, televised or digital accounts, revised local expectations, and practical experience with claims or recovery. Those channels persist differently. A person can vividly remember a flood yet lack money to renew coverage; another can maintain coverage under a mortgage covenant while assigning little subjective attention to flood risk. Institutional memory may survive in updated flood maps, premiums, claims records, building regulations, and lender requirements after individual attention wanes. The research task is to identify the particular channel through which a remembered event affects a contractual decision.
Tversky and Kahneman (1973) characterized availability in frequency judgments as reliance on how easily examples come to mind. Disaster imagery plausibly increases the accessibility of severe events. A later reduction in media attention or autobiographical vividness could weaken that effect. Yet availability is a judgment process, not direct evidence that hazard information was erased, and a rise or fall in policy purchases alone cannot identify availability. Hertwig, Barron, Weber, and Erev (2004) find in decisions from experience that rare outcomes can be underweighted when decision makers sample limited personal histories. That result clarifies why recency and rarity matter; the laboratory tasks do not reproduce an insurance contract, mortgage requirement, or housing budget.
The study of risk perception also cautions against assuming that beliefs mechanically translate into protective choices. Wachinger, Renn, Begg, and Kuhlicke (2013) synthesize the natural-hazard risk-perception paradox: recognized risk sometimes coexists with limited preparation because trust, efficacy, resources, and other factors intervene. A household may therefore accurately assess flood danger and still lack insurance. This is especially important when renewal is affected by premium increases, deductibles, misunderstanding of exclusions, a previous claim, or expectations of assistance.
Meyer, Baker, Broad, Czajkowski, and Orlove (2014) document that residents facing hurricanes Isaac and Sandy could misjudge wind and water threats. This research distinguishes perceived event probability from perceived severity and peril-specific understanding. In insurance, confusing wind with flood risk can produce a persistent mismatch between what a household fears and what an existing policy actually covers. A renewal may then preserve an inadequate product or fail to address the most relevant hazard. Memory must therefore be measured with multiple questions: event recall, predicted chance of damage, expected severity, efficacy of insurance, knowledge of the policy's covered peril, and recollection of previous claim settlements.
LaCour-Little, Palim, and Pan (2025) examine consumer flood-insurance and mitigation choices. Their survey-based evidence gives an additional reason to distinguish reported knowledge from the budgetary and institutional feasibility of coverage, without treating a cross-sectional survey as a longitudinal measure of forgetting.
The literature's historical progression moves from behavioral accounts of vivid hazards to administrative estimates of uptake and policy tenure, and subsequently to designs that distinguish geographic and legal sources of persistence. This development suggests a scientific agenda: connect information acquisition after an event to the actual series of contract decisions, while allowing contractual and environmental conditions to change across years.
Time Since Loss, Information, and Perceived Risk
Time since the latest damaging event is an observable variable; memory is a latent characteristic inferred from responses or behavior. The distinction matters because the calendar also governs rebuilding, financial recovery, policy expiration, mortgage refinancing, disaster-aid conditions, premium changes, and demographic mobility. An event-time coefficient can absorb all these concurrent changes. Proper measurement requires evidence capable of distinguishing a fading response from a changing constraint.
A useful description of perceived annual loss risk at time t includes an underlying belief about hazard probability, a belief about conditional damage given the hazard, and an assessment of insurance effectiveness. If the perceived annual probability is denoted p(t) and subjective loss severity is L(t), then p(t)L(t) describes an expected-loss component of a decision; it does not exhaust expected utility, risk aversion, contract loadings, deductible exposure, liquidity, or ambiguity. Moreover, a household may know the annual chance of a flood and nevertheless misinterpret what a "100-year flood" means. The insuring decision is consequently sensitive both to what is remembered and to how statistical information is communicated.
Gallagher (2014) interprets observed take-up patterns as consistent with Bayesian learning with forgetting. This is a model comparison in that empirical setting; it is not proof that all households follow the same learning rule. Residents in nonflooded communities sharing a television media market increased take-up at approximately one-third the rate in flooded communities, providing evidence that information can travel beyond physical damage. Media geography is thus analytically relevant, but shared television-market status may also correlate with economic and geographic characteristics requiring research controls.
Flood risk may also be capitalized in property values. Atreya, Ferreira, and Kriesel (2013) examine the 1994 flood in Georgia and report a temporary floodplain price discount, disappearing between four and nine years under different specifications. Kousky (2010), studying St. Louis County property sales after the 1993 flood, finds heterogeneous property-price responses across flood zones and river municipalities. These are evidence about market valuations after disasters, not direct measurements of insurance renewal or forgetting. Property prices aggregate buyers, sellers, credit constraints, local amenities, and observed physical changes. Calling the entire valuation effect psychological decay would assign too much causal force to memory.
An increase in objective risk over time further complicates interpretation. Historical recurrence frequencies may be inadequate when exposure, drainage, sea level, or precipitation patterns change. An observed return of perceived risk toward the pre-event level could imply an increasing mismatch if physical risk rises. Conversely, successful defenses or relocation might reduce actual exposure. A scientifically defensible account reports both subjective responses and a time-indexed physical-risk measure, and it distinguishes increased recall from justified belief updating.
Flood Insurance Take-Up and the Empirical Decay Record
The early observational foundation comes from Browne and Hoyt (2000), who use U.S. NFIP state-level information from 1983–1993 and find that purchases correlate with prior-year flood losses, alongside income and price effects. The level of aggregation limits identification of household beliefs: state-level damages may alter market attention, eligibility, administrative activity, and the composition of exposures. Yet the work establishes an important empirical regularity: previous losses predict subsequent insurance purchasing within a market.
Gallagher (2014) advances the evidence using a national panel of large regional floods and policies. The post-flood rise and decline are statistically consistent with gradual forgetting or incomplete knowledge within the author's learning framework, and media spillovers occur beyond flooded communities. The result concerns take-up after a disaster, not an observed psychological half-life for each household or the exact expiration time of individual contracts. One community can have persistently insured households even when the marginal post-event demand shock has vanished.
Using 153 Georgia counties over 1978–2010, Atreya, Ferreira, and Michel-Kerjan (2015) estimate flood-insurance demand with county and time variables and conclude that the recent-event association fades after three years. They also find relatively price-inelastic take-up in that observed market. The three-year finding is conditional on their setting and statistical specification; it cannot be equated to the median tenure of individual contracts, nor should the finding of price inelasticity be treated as zero price sensitivity.
Kousky (2017) uses a panel of Atlantic and Gulf Coast policies between 2001 and 2010 to separate direct storm exposure and government action. Her post-hurricane purchase estimates include an estimated 7.2% increase in net purchases after a hurricane, with the response gone by three years; federal disaster declarations and grant-conditioned insurance requirements are also associated with additional purchases. In the author's interpretation, removing purchases tied to aid conditions leaves a much smaller estimated voluntary increase. These percentages apply to different treatments and model conditions, and should not be summed as independent effects. The central implication is that apparent behavioral learning can partly be an administrative response to public policy.
The clearest reason to resist a universal decay law comes from Petkov and Ortega (2025). Their flood-zone and periphery analysis, which links building footprints and inundation from a major flood, shows that the increase in flood-insurance take-up disappeared after three years inside the special flood hazard area but remained highly persistent outside it. Their interpretation stresses that unexpectedly flooded peripheral residents learned something new about actual exposure. This is a substantive modification of the simplest recency narrative: learning from a surprising event can outlast the immediate burst of attention, especially where previous official mapping or public understanding understated risk.
The empirical studies support at least four distinct temporal phenomena: a near-term event-triggered purchase spike, eventual decline in an aggregate response, continued insurance by some initial purchasers, and durable changes in beliefs or take-up among groups receiving especially informative signals. Treating these as one process would erase the most valuable distinctions in the evidence. A careful future panel analysis must follow the cohort that first purchased after the flood and compare its retention with both continuing incumbents and newly insured later entrants.
Contract Renewal, Policy Tenure, and Administrative Continuity
Renewal is a distinct insurance event. An individual can remember a flood vividly while failing to renew because payment was missed, an intermediary relationship ended, coverage became unaffordable, or a financing condition changed. Another policyholder may forget the disaster while maintaining a contract through escrow or habit. The outcome can be an active policy, a lapse, a cancellation, a switch to another insurer, a migration into a different insurance product, or a period without adequate peril coverage. Any attempt to estimate memory decay from renewal must account for these competing pathways.
Michel-Kerjan, Lemoyne de Forges, and Kunreuther (2012) analyze the tenure of NFIP policies during 2001–2009 and estimate median tenure of new policies between two and four years. Small prior flood claims are associated with longer retention while large claims are associated with earlier lapses. These findings complicate an intuitive claim that greater personal loss always strengthens insurance persistence. Severe claims may be accompanied by relocation, changes in property ownership, substantial rebuilding costs, insurance dissatisfaction, or changes in premiums. The data pattern alone cannot determine which mechanism predominates in each case, but the asymmetry is compelling evidence that claim severity should enter any longitudinal design separately from the binary experience of damage.
The survival perspective starts at the correct denominator. Let a policy's discrete conditional lapse hazard at the end of insurance year k be the probability that the existing policy ceases during k, conditional on that policy remaining in force at the start of the year. The corresponding continuation probability is one minus that hazard. Multiplying annual continuation probabilities yields the observed retention curve. This is a standard survival-analysis accounting identity, used here as an analytic description rather than a proposed new model. Tenure estimates that ignore incomplete observation at the end of a dataset will understate or distort duration; a switch to private flood coverage should be treated as a competing outcome rather than automatic disappearance of all protection.
Policy anniversaries are particularly important because they impose recurrent decisions on households whose attention to disaster risk is episodic. According to the National Flood Insurance Program's renewal instructions, NFIP policies generally require annual renewal; they do not automatically renew. Mortgage lenders and escrow arrangements may facilitate payments. The program's policy purchase guidance describes a one-year policy term and a conditional 30-day renewal grace period. The availability of a grace period is a contractual and administrative rule, not evidence that a delayed renewal reflects restored risk perception. The specific governing terms and temporary program directives should be checked for the historical cohort under investigation.
A rigorous renewal study must distinguish five dates: the catastrophe, policy inception, renewal notice, contractual expiration, and effective date of any substitute coverage. An event immediately before a renewal deadline could affect the near-term continuation decision much more than an identical event occurring just after renewal, even if the household's risk belief changed equally on both dates. This creates potential quasi-experimental variation around anniversaries, subject to seasonality and policy-selection tests. The analysis also needs to distinguish failure to respond to a notice from a deliberate decision to decline the offer. Observed cancellation codes and payment records rarely reveal the psychological reason unambiguously.
Coverage limits matter alongside a binary active-policy outcome. A household may maintain a policy while reducing limits, raising deductibles, excluding contents, or switching to a product with narrower protection. Alternatively, its premium may increase while coverage remains stable. Renewal quality should thus be evaluated as continuity of adequate insurance for the actual hazard exposure. Measuring only policy count rewards apparent persistence that could conceal progressive underinsurance.
The program also supplies a direct NFIP renewal channel, illustrating that contract continuation depends partly on renewal infrastructure and the availability of an actionable payment route. This institutional fact should be measured separately from disaster recall.
Administrative intervention can stabilize retention without altering beliefs. Automatic premium financing, multiyear arrangements, reminders, lender verification, and transparent peril descriptions can reduce inadvertent gaps. The policy-tenure study's suggestion of multiyear property-attached policies is an institutional proposal, and its feasibility depends on contract law, affordability, insurer capital management, underwriting flexibility, and borrower protection. Such interventions should be tested against actual continuation, cost, and welfare, rather than evaluated solely by the number of reminder messages sent.
Repeated Hazard, Learning, and Heterogeneous Persistence
Repeated hazards create complicated information histories. A resident who has survived several storms without damage may discount warnings, while another who experienced a flood outside the official high-risk map may sharply revise expectations. The physical experience of the same storm is therefore not equivalent across households, and identical media coverage need not produce identical learning.
Petkov and Ortega's (2025) periphery persistence result is especially consequential because it contradicts the presumption that all post-event demand mechanically returns to its baseline in three years. Residents outside special flood hazard areas had more reason to be surprised by actual inundation. The observed durable uptake is compatible with information about previously underestimated hazard. Inside mapped high-risk areas, the increase in take-up was temporary, despite continued physical exposure. One possible explanation is that information provides diminishing marginal news to already-informed residents, while payment and administrative constraints affect the observed market response. This is an analytical interpretation of heterogeneous evidence; the study does not independently observe and experimentally manipulate each household's latent belief.
The presence of a disaster also changes perceived severity and perceived control. Homeowners may regard mitigation as a substitute for insurance, a complement to insurance, or a competing use of limited cash. Osberghaus (2017) examines flood experience and private adaptation in longitudinal German household evidence. This literature is relevant because prior disaster experience can affect precautionary actions beyond policy purchase. Investments in flood barriers or property elevation might lower physical losses while changing the optimal insurance decision; observing decreased insurance demand after effective mitigation would therefore have a different interpretation from observing a lapse with unchanged exposure.
Claim severity provides another dimension of heterogeneity. The Michel-Kerjan, Lemoyne de Forges, and Kunreuther (2012) severity–tenure relationship suggests that a small claim may reinforce the value of coverage while a large claim interacts with more disruptive financial and housing choices. Direct damage, vicarious experience, media exposure, aid receipt, and the perceived quality of claims service should be separately identified. For households displaced by catastrophic events, an apparently missing renewal can reflect the end of insurable exposure at the previous address. A research design organized only by geographic policy counts would confound this with behavioral forgetting.
Overlapping catastrophes complicate event-time models. A household exposed to a flood and subsequently a wildfire, hurricane, or new flood may update different beliefs about different insured perils. Identifying the effect of the first disaster from the second requires an exposure history, attention to simultaneous disasters in nearby counties, and controls for common insurer or regulator responses. Recurrent disaster timing may be correlated with underlying climatic conditions, so assuming independent events will typically be untenable. Hazard-specific outcome measurement and appropriately clustered uncertainty estimates are necessary.
The scientific conclusion is a conditional one: prior experience can make risk more salient, furnish new information, and alter the economics of rebuilding. The temporal persistence of coverage varies according to what the event reveals, which constraints bind, and whether the household continues to possess an insurable interest. These mechanisms are testable, but they cannot be inferred directly from the slope of aggregate policy counts.
Alternative Explanations and Causal Identification
The most serious threat to a causal claim about memory is the many simultaneous changes following a disaster. Flood damage changes household wealth; it may also change perceived probability, lender requirements, local prices, migration, disaster relief, insurer availability, and public risk maps. These treatments interact. A design that attributes all additional purchases in the following year to memory has estimated a total event response rather than the psychological contribution of remembering.
A disaster can change insurance enrollment through public assistance. In Kousky's (2017) policy-shock analysis, aid-related coverage requirements account for much of the measured post-hurricane purchase increase. The study estimates approximately 7.2% additional net purchases in the year after a hurricane, yet roughly 1.5% additional voluntary purchases when accounting for requirements associated with aid. It also reports a 6.7% association with disaster declarations and a 5% association with aid-grant availability. Those figures come from separate statistical contrasts; their additive interpretation would be incorrect. This evidence establishes the necessity of tracing explicit legal enrollment conditions before attributing policy changes to emotion or learning.
Kousky, Michel-Kerjan, and Raschky (2018) investigate how disaster assistance relates to private flood insurance decisions. Their work raises the possibility that expected ex post public assistance changes insurance incentives. This is a separate economic mechanism from memory decay. A household may retain a precise understanding of hazard but expect the government to bear part of the consequences. Empirical identification is difficult because aid can be targeted to places experiencing more serious harm. Comparisons across aid regimes require careful consideration of eligibility rules and distribution mechanisms.
Kousky (2018) reviews NFIP financing and risk communication, linking take-up, rate setting, public-private market roles, mitigation and catastrophe financing. This broader institutional analysis provides a basis for identifying policy and price mechanisms that empirical event studies need to control.
At the market level, the price and availability of coverage can change after a large event. Risk-based repricing, revised deductibles, insurer withdrawal, new reinsurer prices, or changed risk classifications may alter renewal for reasons unrelated to forgetting. A decline in take-up as risk salience wanes could therefore be a demand response to improved information, a demand response to price, an insurer-supply constraint, or the simultaneous movement of all three. Longitudinal policy data should include premium offers, rate revisions, deductible changes, coverage limits, insurer identifiers, and market entry or exit.
Several identification strategies are credible if their assumptions are explicitly tested. Event-study difference-in-differences estimates can compare exposed and plausibly comparable unexposed areas before and after disasters, with pre-event trends and anticipation effects assessed. An event-study graph provides valuable descriptive dynamics even when causal attribution remains conditional on parallel-trend and composition assumptions. Geographic boundaries can separate contrasting exposure classifications but may coincide with systematic housing and lending differences. Media-market comparisons can identify spillovers, provided common market boundaries do not proxy for unmeasured socioeconomic changes. An instrument based on exogenous information exposure requires a strong exclusion argument: storms themselves, physical damage, and associated public interventions may affect coverage directly.
Policyholder fixed effects can absorb time-invariant household characteristics, but not endogenous moves, evolving resources, or changes in contractual eligibility. Survival regressions of lapse must define whether the unit is a policy number, person, property, or household. Administrative identifiers might change after carrier transitions, creating apparent churn. Dynamic selection is particularly acute: an insured cohort observed after three years consists disproportionately of early survivors. The survival probability of that cohort cannot be interpreted as the renewal probability of all initially exposed residents.
Measurement error also follows geography. Official high-risk classifications are informative regulatory categories, not complete physical-risk measurements. Building-level inundation often differs from coarse flood-zone indicators; flood-zone revisions might occur after the disaster. Petkov and Ortega (2025) address this by connecting building footprints and inundation records, thereby furnishing more direct exposure evidence. For robust causal inference, researchers should examine whether control areas experience media spillovers, whether geographic spillover boundaries are stable, and whether claims and aid data identify genuine damage at the right location.
International Comparisons and Institutional Design
The U.S. evidence is unusually rich because the National Flood Insurance Program produces administrative policy histories and repeated disaster events. Cross-national comparison is nevertheless essential: legal coverage requirements, risk financing, consumer beliefs, and contract terms vary. Transferring a U.S. three-year decay coefficient into another system would ignore these institutional conditions.
Japan's public–private earthquake insurance arrangement offers a contrasting contract structure. The Japanese Ministry of Finance's earthquake insurance outline describes coverage tied to fire insurance, a government reinsurance role, and policies available for one-year or longer terms within the relevant contractual rules. An analysis of renewal would need to examine whether the decision is joint with a home fire policy, how rate classifications and premium deductions work, and how households evaluate earthquake coverage after widely observed disasters. These institutional features can affect persistence even when two populations have similar risk perceptions.
The United Kingdom's Flood Re arrangement offers a further institutional comparison. The government describes the Flood Re scheme as a mechanism intended to improve the availability and affordability of flood insurance for eligible homes while facilitating a transition to a risk-reflective market, with the scheme scheduled to end in 2039. Flood Re's published annual reports describe scheme development and household access. A subsidy or reinsurance facility may maintain coverage through adverse market conditions, but the existence of insurance after a disaster does not establish whether affected households' risk memories persisted. A credible British study would examine household eligibility, premium changes, insurer participation, flood history, and property-level coverage separately.
These comparisons reveal why insurance institutions belong inside a behavioral analysis. Some markets provide direct public catastrophe cover; others use private policies supported by a reinsurance pool; still others permit or require specific peril riders. Information campaigns, lender rules, subsidy arrangements, insurer renewal offers, and policy duration change the opportunity set in which memory operates. The comparative unit should therefore be the insured decision under a specified contract and institutional regime, rather than a generic count of insured properties.
International protection-gap evidence provides a macroeconomic frame, not a measurement of lapse. Swiss Re's natural-catastrophe protection-gap research reports substantial uninsured losses globally. A protection-gap figure depends on disaster losses, asset exposure, insured claims, deductibles, and regional insurance penetration. It cannot identify the share of uncovered losses resulting from fading disaster memory. The gap may be driven by uninsurable property, low incomes, incomplete public markets, limited insurer capacity, or exclusions even among households who accurately anticipate the hazard.
Longitudinal Empirical Tests and Measurement Protocol
The evidence suggests a series of sharply defined questions rather than a single global decay parameter. The first asks whether information acquired through disaster experience persists in expressed risk beliefs. The second asks whether a household carries the same insurance through subsequent renewal dates. The third asks whether hazard-specific coverage remains adequate as premium, exposure, and household conditions change. Estimating each independently permits substantive statements about memory, contract behavior, and financial protection.
A prospective observational study would identify a pre-event cohort of properties and residents in a mapped hazard region. It would record building exposure, previous claims, mortgage status, premiums, deductibles, insured sums, insurer, and duration of existing coverage. After an event, respondents would periodically report perceived likelihood of future damage, perceived severity, knowledge of insured and excluded perils, recollection of the event, efficacy expectations, and willingness to pay. Survey responses would be linked, with appropriate consent and privacy protection, to administrative policy and property data. Surveys and administrative data would cover at least several renewal anniversaries, not merely the first post-event year.
Outcome definitions would be prespecified. The initial uptake rate is the share of eligible initially uninsured households obtaining appropriate new cover in a fixed interval. Renewal persistence is the share of contracts active at each anniversary among policies at risk of expiration, with transfers or replacements separately identified. Continuous property protection is the fraction of property-years with suitable hazard coverage, regardless of carrier. Coverage adequacy measures the insured amount and relevant perils against a common exposure definition. Subjective risk memory is measured through comparable repeated questions, with attention to survey attrition and how wording itself can prime recall.
Analysis should use mutually informative but distinct estimands. A difference-in-differences event study estimates the change in coverage relative to a counterfactual trajectory under its identification assumptions. Survival analysis estimates renewal or lapse hazards among the original insured cohort. Repeated-measures models describe beliefs over time, with attrition weights or bounds assessed. Where plausible, joint analysis can ask whether changes in expressed beliefs mediate policy choices, while reporting that mediation estimates require additional untestable assumptions about omitted confounders. A convincing finding of memory decay requires evidence on the belief trajectory itself, not only a negative coefficient on time since disaster.
Repeated shocks, household moves, disaster aid, and pricing reforms should be recorded as time-varying covariates. A physical-hazard model independent of insurance transactions should establish whether objective exposure was steady or changed. Geographic clustering must respect spatially correlated disasters; insurer or media-market clustering may also matter depending on treatment assignment. Robustness tests should exclude disaster-aid-mandated policies, examine properties outside mapped high-risk zones, allow separate effects by first-time versus repeated damage, and treat policy switching as a competing risk.
A research protocol should publish a transparent flow of its sample from eligible properties to observed contracts, with separate attrition for surveys, moves, canceled records, and unmatched carriers. The central comparison is between the same residents' risk beliefs and their verified protection through time. When that linkage is unavailable, the results should be labeled aggregate event responses or policy-duration associations, not direct measurements of individual memory.
Economic Welfare, Equity, and Public Policy
A temporary increase in uptake followed by cancellation matters because insurance promises compensation for future losses, while disasters typically recur unpredictably. A household newly insured only in the year immediately following a rare flood may remain exposed during much of the subsequent risk horizon. The financial cost of that exposure depends on hazard, policy terms, alternative recovery mechanisms, and the household's capacity to absorb losses. A policyholder with stable coverage can smooth some financial consequences of shocks, although exclusions, deductibles, claims delays, and policy limits remain relevant.
The economics of renewal also concerns equity. Low-income households may face severe liquidity constraints precisely when disaster salience is strongest. Spending on repairs, temporary housing, and debt service can crowd out renewal premiums. A lapse after major damage then may reflect depleted resources rather than inaccurate risk perception. The relationship in the NFIP tenure analysis between larger claims and earlier lapses points to the need for financial as well as psychological explanations. Any intervention should be assessed for the distribution of premiums, subsidies, and residual risk across homeowners, renters, and vulnerable communities.
Accurate risk communication is useful when it identifies location-specific exposure and specific insured perils. Petkov and Ortega (2025) emphasize more granular flood-risk communication, especially for households outside the traditional flood-zone boundary. A public warning that all residents face some flood risk can create general awareness; a building-level account of plausible inundation and potential uncovered damages can better inform the actual purchase. Communication should state uncertainty and avoid translating a rare historical occurrence into a fixed calendar promise of safety.
Renewal reminders can address inattention, while contractual support can address complexity and transaction costs. The NFIP's annual renewal procedure makes the anniversary an identifiable intervention point. An experimental evaluation might randomize clarity, timing, and hazard-specific information in renewal notifications, subject to consumer-protection rules and equal access. Its primary outcome would be continuous appropriate coverage, not click-through rate. Consumer comprehension and complaint outcomes are important safeguards against designing messages that induce expensive but poorly matched products.
Policymakers must also distinguish stronger private protection from shifted public liabilities. A subsidy that prevents lapses may improve recovery but changes who finances the premium. A lender enforcement measure can stabilize coverage while affecting affordability and housing access. Disaster assistance can reduce immediate suffering yet alter ex ante insurance incentives; the public-aid and purchase evidence demonstrates how large this channel may be in observed U.S. policy data. Cost-effectiveness therefore requires estimating uninsured loss reductions, risk-reduction investments, administrative expenditure, distributional incidence, and effects on the availability of future insurance.
The larger empirical purpose is to locate when coverage gaps arise from preventable lapses, when they arise from prices and resources, and when they arise from missing or unsuitable insurance markets. The right intervention depends on that diagnosis. Maintaining a contract through an automatic mechanism could reduce forgetfulness-related lapse while leaving an unaffordable premium unresolved. Subsidizing a premium could address affordability while leaving the insured peril misunderstood. A successful policy integrates measurement of beliefs, conditions of purchase, and the actual financial protection that remains after several years.
Integrated Findings and Research Limits
The literature is most consistent on the existence of an event-triggered response. Floods and hurricanes often increase near-term purchases or interest in coverage. Gallagher (2014), Atreya, Ferreira, and Michel-Kerjan (2015), and Kousky (2017) identify substantial temporal attenuation in particular observational settings. The precise three-year pattern must remain attached to those locations, samples, and definitions. Petkov and Ortega (2025) document a durable peripheral-zone response that demonstrates the inadequacy of a universal three-year rule.
The second finding concerns policy survival. Michel-Kerjan, Lemoyne de Forges, and Kunreuther (2012) observe a two-to-four-year median tenure of new NFIP policies in their sample and heterogeneous consequences of prior claims. Tenure is a directly measured coverage trajectory. It is related to, yet distinct from, the time series of aggregate take-up or recalled disaster salience. Its interpretation depends on annual contracts, renewals, market switching, relocation, and administrative financing conditions.
The third finding is causal. Risk perceptions can fade without coverage lapsing, and coverage can lapse while risk remains salient. Claims, aid conditions, credit arrangements, price changes, and physical mitigation are mechanisms that can move these outcomes in different directions. This is the central analytic contribution of the dossier: organizing documented evidence around three separately measurable trajectories—risk beliefs, contract continuation, and adequacy of protection—and specifying the empirical conditions under which their association becomes a defensible causal explanation. It is an evidence-based synthesis of existing methods and research, and claims no new theory or conceptual system.
Several limitations qualify the synthesis. Much of the directly relevant quantitative evidence derives from U.S. flood insurance and selected storms. Studies of insurance demand frequently examine purchases rather than verified subsequent individual renewal; psychological studies examine perception or recall without linked contract records. Institutional records describe present and historical program rules but cannot independently establish behavioral motivation. Publisher abstracts and accessible scholarly text, rather than unrestricted raw administrative datasets, constitute part of this review's verification base. The dossier has not performed original statistical estimation or reproduced the underlying proprietary datasets. These limits should shape future research priorities and prevent claims of globally established causal coefficients.
Conclusion
Disasters alter insurance decisions through information, experience, public response, and changes to the contractual environment. The resulting rise in demand can recede, while some individual policies remain in force and other households preserve a newly acquired understanding of hazard for longer. Evidence from U.S. flood insurance documents both temporary increases in take-up and policy tenures measured over several years; recent geographic research shows persistent post-flood coverage beyond mapped high-risk areas. These observations fit a conditional account of learning and institutional constraint in which the history of a disaster remains financially consequential after its prominence in public attention has declined.
The scientific task is to establish which part of the observed change reflects belief, which reflects contract administration, and which reflects changes in real exposure or affordability. Event-time studies of aggregate take-up, contract-level survival analysis, and direct repeated measurement of perceived risk answer different questions. Treating them as complementary evidence produces a more accurate understanding of coverage renewal and policy lapse.
Insurance institutions can improve continuity through clear risk information, usable renewal procedures, accurate coverage descriptions, and financial arrangements that preserve appropriate protection. These measures require evaluation across income groups, hazard zones, countries, and policy designs. The generalizable finding is methodological and substantive: disaster memory must be studied alongside the legal, economic, and operational life of an insurance contract. The persistence of protection is ultimately determined by the interaction of learned risk, available coverage, and the household's continuing capacity and opportunity to insure.
References
Atreya, A., Ferreira, S., & Kriesel, W. (2013). Forgetting the flood? An analysis of the flood risk discount over time. Land Economics, 89(4), 577–596. https://doi.org/10.3368/le.89.4.577
Atreya, A., Ferreira, S., & Michel-Kerjan, E. (2015). What drives households to buy flood insurance? New evidence from Georgia. Ecological Economics, 117, 153–161. https://www.sciencedirect.com/science/article/pii/S0921800915002876
Browne, M. J., & Hoyt, R. E. (2000). The demand for flood insurance: Empirical evidence. Journal of Risk and Uncertainty, 20(3), 291–306. https://doi.org/10.1023/A:1007823631497
Flood Re. (n.d.). Reports and publications [Institutional reports and annual-report collection]. https://www.floodre.co.uk/about-us/reports/
Gallagher, J. (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://www.aeaweb.org/articles?id=10.1257/app.6.3.206
Hertwig, R., Barron, G., Weber, E. U., & Erev, I. (2004). Decisions from experience and the effect of rare events in risky choice. Psychological Science, 15(8), 534–539. https://journals.sagepub.com/doi/10.1111/j.0956-7976.2004.00715.x
Kousky, C. (2010). Learning from extreme events: Risk perceptions after the flood. Land Economics, 86(3), 395–422. https://www.rff.org/publications/journal-articles/learning-from-extreme-events-risk-perceptions-after-the-flood/
Kousky, C. (2017). Disasters as learning experiences or disasters as policy opportunities? Examining flood insurance purchases after hurricanes. Risk Analysis, 37(3), 517–530. https://doi.org/10.1111/risa.12646
Kousky, C. (2018). Financing flood losses: A discussion of the National Flood Insurance Program. Risk Management and Insurance Review, 21(1), 11–32. https://doi.org/10.1111/rmir.12090
Kousky, C., Michel-Kerjan, E. O., & Raschky, P. A. (2018). Does federal disaster assistance crowd out flood insurance? Journal of Environmental Economics and Management, 87, 150–164. https://www.sciencedirect.com/science/article/pii/S0095069617303479
LaCour-Little, M., Palim, M., & Pan, Y. (2025). Why don’t consumers purchase flood insurance or mitigate flood risk? Journal of Housing Research, 34(1), 60–89. https://doi.org/10.1080/10527001.2024.2425230
Meyer, R. J., Baker, J., Broad, K., Czajkowski, J., & Orlove, B. (2014). The dynamics of hurricane risk perception: Real-time evidence from the 2012 Atlantic hurricane season. Bulletin of the American Meteorological Society, 95(9), 1389–1404. https://doi.org/10.1175/BAMS-D-12-00218.1
Michel-Kerjan, E., Lemoyne de Forges, S., & Kunreuther, H. (2012). Policy tenure under the U.S. National Flood Insurance Program (NFIP). Risk Analysis, 32(4), 644–658. https://onlinelibrary.wiley.com/doi/10.1111/j.1539-6924.2011.01671.x
Ministry of Finance, Japan. (2026). Outline of Japan’s earthquake insurance system [Government explanatory page]. https://www.mof.go.jp/english/policy/financial_system/earthquake_insurance/outline_of_earthquake_insurance.html
National Flood Insurance Program. (n.d.). Buy a flood insurance policy. https://www.floodsmart.gov/get-insured/buy-a-policy
National Flood Insurance Program. (n.d.). How to renew your flood insurance policy. https://www.floodsmart.gov/get-insured/renew-a-policy
National Flood Insurance Program. (n.d.). Renew my flood insurance. https://www.floodsmart.gov/renew-direct
Osberghaus, D. (2017). The effect of flood experience on household mitigation—Evidence from longitudinal and insurance data. Global Environmental Change, 43, 126–136. https://www.sciencedirect.com/science/article/pii/S0959378016302394
Petkov, I., & Ortega, F. (2025). Learning from experience: Flooding and insurance take-up in the flood zone and its periphery. Journal of Risk and Insurance, 92(2), 312–356. https://onlinelibrary.wiley.com/doi/10.1111/jori.70002
Swiss Re Institute. (2026). The natural catastrophe protection gap [Research overview]. https://www.swissre.com/institute/research/topics-and-risk-dialogues/climate-and-natural-catastrophe-risk/growing-exposure/Natcat-protection-gap.html
Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207–232. https://www.sciencedirect.com/science/article/pii/0010028573900339
U.S. Government Accountability Office. (2026, October 1). Flood insurance: Congressional action could help increase coverage for at-risk properties (GAO-27-108012). https://www.gao.gov/products/gao-27-108012
United Kingdom Government. (n.d.). Flood Re: About us. https://www.gov.uk/government/organisations/flood-re/about
Wachinger, G., Renn, O., Begg, C., & Kuhlicke, C. (2013). The risk perception paradox—Implications for governance and communication of natural hazards. Risk Analysis, 33(6), 1049–1065. https://onlinelibrary.wiley.com/doi/10.1111/j.1539-6924.2012.01942.x