Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Behavioral Demand for Index-Based Insurance

Abstract. Index-based insurance pays according to an observable rainfall, yield, vegetation, mortality, or other contractual index rather than the individual policyholder's independently adjusted loss. Its administrative advantages coexist with basis risk: a covered household may suffer a loss without receiving a payout, or receive a payout without experiencing a corresponding loss. The resulting demand problem requires joint analysis of contract performance, subjective beliefs, preferences under risk and ambiguity, liquidity, provider credibility, informal transfers, social learning, and experience with actual payouts. This monograph critically synthesizes theoretical results, observational evidence, randomized field experiments, and comparative institutional research, primarily from agricultural and pastoral insurance in Asia and Africa. It develops an explicit distinction between the value of actuarial protection, the perceived effectiveness of protection, and the feasibility of purchase at a given moment. The evidence establishes that price and basis risk materially influence demand, while the causal effects of education, framing, peer information, and payout experience depend upon their design and setting. A high willingness to pay is neither a sufficient measure of risk reduction nor evidence of sustained voluntary coverage. The concluding research assessment specifies what must be measured before behavioral interventions can be credited with improving household welfare: loss-index correspondence, actual and expected payout reliability, informed choice, durable renewal, and the distribution of residual risk.

Keywords: Index-based insurance; parametric insurance; behavioral demand; basis risk; rainfall insurance; agricultural microinsurance; index-based livestock insurance; willingness to pay; ambiguity aversion; compound risk; trust; liquidity constraints; social networks; insurance learning; renewal; consumer protection; randomized field experiments; welfare assessment.

Introduction

The insured event in an index contract is an objectively specified measurement. In rainfall insurance, it may be the amount of precipitation observed at a designated station during a defined crop-growth interval. In area-yield insurance, it may be an officially estimated regional yield. In livestock insurance, it may be an index that predicts drought-associated mortality from remote-sensing or related environmental data. A purchaser must understand two distinct relationships: the index must be observable under the contract, and its movements must correspond sufficiently closely to the purchaser's actual financial vulnerability. Demand depends on both relationships, as well as on the purchaser's capacity to pay and belief that the insurer will honor the agreement.

The rainfall participation study in rural India by Gine, Townsend, and Vickery (2008) linked observed purchases with basis risk, household wealth, credit constraints, village networks, and familiarity with the vendor. The randomized Indian demand experiments of Cole et al. (2013) subsequently identified causal responses to pricing and marketing interventions. The Ghanaian insurance experiments reported by Karlan et al. (2014) connected insurance provision with farm investment and found subsequent purchases associated with personal and network payout experience. Taken together, these studies place demand within an interaction among the quality of contingent protection, financial constraints, and expectations about performance.

The object of this dossier is the individual's and household's decision to purchase, maintain, change, or reject index-based insurance. Formal welfare theory and physical index calibration are indispensable inputs, but the primary question is how people interpret and act upon a contingent contract whose payout can diverge from their lived loss. The separate Insurance Economics and Parametric and Index-Based Insurance Science areas provide the broader pricing, market, legal, and index-design context. The present analysis addresses behavior and evidentiary identification across those boundaries, retaining one canonical scholarly identity.

This is a critical synthesis of published research, not an original household survey, randomized trial, systematic meta-analysis, or new authored theory. The source set consists of publisher-verified journal records, author or institutional manuscripts where available, and research-institution reports. Inference is limited to the findings and designs actually described by these sources; randomized effects are distinguished from correlations, simulated welfare is distinguished from measured welfare, and laboratory elicitation is distinguished from paid purchase. The bibliographic review is current through November 14, 2026.

Analytical terms and scope

Let a household's uncompensated loss be L, the contractual index be I, and the nonnegative payout function be P(I). Its post-event financial loss before other transfers is L - P(I). Index-related residual exposure can be represented by Var[L - P(I)] or, more meaningfully for severe-loss protection, by a tail-loss measure such as E[(L - P(I) - k)_+], for a specified threshold k. These are alternative descriptive measures; neither should be labeled a universal index-quality score. The distribution of residual losses across households is as important as an average fit statistic.

For demand research, three events need separate observation: an offer of an actual contract, a paid purchase, and effective protection in a later state of nature. A respondent who says a policy is desirable has not necessarily purchased it; a purchaser has not necessarily understood it; and a timely, contractually correct payout has not necessarily reduced an individual's loss. The conceptual distinctions structure all chapters.

Index Products and Consumer Expectations

The economic attraction of an observable trigger

Traditional indemnity insurance adjusts claims against individual covered losses. That process can be expensive, slow, and informationally difficult when insured values are small, losses are dispersed, and independent assessment is costly. An index-based contract uses a pre-agreed formula and an external measurement. The provider can determine an obligation without investigating every policyholder's specific crop or livestock loss. This can lower loss-verification expenses and permit faster settlement when the underlying data arrive promptly. The index insurance reassessment by Carter, de Janvry, Sadoulet, and Sarris (2017) documents both the promise of this instrument and the continuing constraints on voluntary take-up.

For a buyer, the product is a financial claim on a measurement process. The buyer therefore needs an accurate understanding of the insured variable, measurement location, observational window, data authority, payout threshold, payout function, maximum benefit, exclusions, timing, and remedies for data errors. The natural-language assertion that a policy "covers drought" can be materially misleading if the operative contract covers only rainfall below a specified station threshold during particular weeks. A drought can cause losses with no contractual index trigger, and an index trigger can occur without the purchaser's corresponding loss. Both outcomes follow from the contract's structure.

The early Indian experience is instructive. Gine, Townsend, and Vickery (2008) describe rainfall contracts linked to three monsoon phases and report that buyers' connection with the selling microfinance institution and village networks was strongly associated with purchase. These associations do not establish that every network member understood the trigger. Familiarity, access, trust, and marketing intensity can move together. The decision to buy is therefore best viewed as a choice among expected contingent transfers, information burdens, and alternative uses of scarce cash.

Expectations have more than one object

The household can hold different beliefs about the probability of crop failure, the probability that the index crosses its trigger conditional on crop failure, the probability of administrative or commercial nonperformance, and the time needed to obtain a payment. A single question such as "Do you trust the insurance?" combines these distinct beliefs. If a respondent expects the weather station to be poorly correlated with the farm's conditions, the perceived payout probability following personal loss is low even when the insurer is perfectly solvent and pays according to contract. Conversely, strong confidence in a known seller does not correct a physically weak index.

A useful elicitation approach asks for separate judgments about a named loss scenario and the resulting contractual index. It then asks about likely contract performance once the trigger is met. These judgments should be compared with a documented trigger rule and, where possible, independent historical loss-index pairs. Errors may reflect lack of understanding, ambiguity about environmental patterns, distrust of the data provider, distrust of the insurer, or rational knowledge that the trigger is imperfect.

The behavioral meaning of product simplicity

A simple illustration of rainfall falling below a threshold can reduce the burden of learning. Simplicity may also conceal uncertainty about the location and timing of measurement. Product explanations should permit buyers to identify both favorable and unfavorable contract scenarios. A successful knowledge test should include loss-without-payout examples, not only examples in which drought and payment coincide.

In the southern Ethiopian livestock insurance experiment, Takahashi, Ikegami, Sheahan, and Barrett (2016) found that randomly supplied learning kits improved product knowledge but did not produce strong evidence that knowledge improvements themselves raised take-up. Randomly supplied discounts did increase immediate demand without the hypothesized later reference-price penalty. This result directly challenges the presumption that comprehension alone is the binding constraint. Better understanding can reveal a policy's weaknesses as well as its benefits.

Basis Risk and Perceived Protection

Basis risk is residual exposure experienced by a specific household

A regional index necessarily aggregates over different farms, herds, soil types, cropping patterns, and management practices. Even a well-calibrated environmental index captures only part of the variation in household outcomes. The Kenyan product-quality analysis of Jensen, Barrett, and Mude (2016), using eight semiannual seasons of longitudinal household data, found that the studied livestock index substantially reduced exposure to covariate high-loss events, while significant idiosyncratic residual risk remained. Its reported average reduction in covariate high-loss exposure was 63 percent, yet the reported remaining exposure to high-loss risk averaged 69 percent of the original because noncovariate risks were substantial. These statistics characterize that particular product and sample; they are not universal performance rates for index insurance.

Physical fit is not identical to behavioral response. A household may know from experience that its herd condition differs from the regional proxy. A farmer may cultivate a crop whose critical growth period falls outside the selected rainfall window. The buyer may also misunderstand an otherwise well-calibrated index. The analyst must distinguish objective basis risk, perceived basis risk, and the information used to form the perception. Observed demand can rationally decline with actual poor index fit without indicating any psychological bias.

The theory of rational index-insurance demand developed by Clarke (2016) makes this point especially clear. With basis risk, optimal insurance demand can be nonmonotonic in wealth, price, and risk aversion, and can reach zero for an extremely risk-averse decision maker. Greater aversion to risk therefore does not automatically imply higher purchases of every product bearing the word insurance. The incomplete hedge creates additional uncertainty about the payoff just when protection is needed.

Distinguishing basis risk from insurer default

Let C be a binary indicator that the contractual index triggers, D an indicator of a qualifying household loss, and H an indicator that an insurer pays when C equals one. Three separate probabilities are relevant: Pr(C=1 | D=1), the conditional coverage sensitivity; Pr(C=0 | D=1), the missed-payout probability; and Pr(H=1 | C=1), the perceived or observed reliability of performing a triggered obligation. A fourth quantity, Pr(D=0 | C=1), captures payments with no matched individual loss. These descriptions deliberately avoid assuming that a household's loss definition and the contract's stated coverage boundary are identical.

A purchaser can regard the first probability as inadequate while trusting the insurer's payment administration. Another may regard the trigger as excellent while distrusting settlement. Pooling these perceptions into one trust score creates measurement error and risks recommending advertising for a poor index or expensive index redesign for a settlement-credibility failure.

The global review of basis-risk evidence by Yuzva, Botzen, Brouwer, and Aerts (2018) emphasizes systematic measurement, index design, stakeholder participation, and integration with other risk-management arrangements. Such integration is especially relevant when an index can cover systemic drought but cannot reimburse every individual crop or herd shock.

Compound lotteries and ambiguity

A household faces uncertainty over the loss itself and over the extent to which a contractual index will respond. Under an expected-utility specification with known joint probabilities, these uncertainties can be represented as a single state-contingent distribution. Real buyers often know the direct loss risk imperfectly and know even less about the relation between their individual loss and the index.

Elabed and Carter's compound-risk experiment in Mali (2015) used framed field experiments among cotton farmers to examine the effect of compound-risk aversion on willingness to pay. Approximately 57 percent of participating farmers displayed compound-risk aversion under the study's elicitation method. A simulated contract comparison using estimated preferences implied materially lower purchase propensity when compound risk was incorporated. These are elicited preference estimates and model-based demand implications, rather than observed market-wide purchase rates.

Bryan's study of ambiguous partial insurance (2019), drawing on randomized trials in Malawi and Kenya, further shows why partial risk protection can be difficult to value when the underlying production technology is itself uncertain. A household considering unfamiliar seeds or cultivation practices must assess both a novel productive return and an imperfect safeguard against loss. A simple premium discount need not remove this informational ambiguity.

A behavioral explanation becomes credible only after a suitable benchmark is specified. The researcher must first show what purchasing behavior is consistent with the observed contract, plausible beliefs, and budget constraints. A residual deviation may then be studied through elicitation of ambiguity, probability weighting, attention, or trust. Treating every unsold index policy as evidence of irrationality assumes the very product quality that should be measured.

Severity and distribution matter more than a correlation alone

High correlation between an index and average regional loss can coexist with dangerous conditional nonpayment for the most vulnerable households. An index that tracks moderate fluctuations well but misses the bottom tail may be least effective when insurance is most valuable. Product evaluation should report performance conditional on large individual losses, differences across location and economic groups, and the probability of receiving a useful payment before a distressed household liquidates productive assets.

The northern Kenya demand investigation by Jensen, Mude, and Barrett (2018) links longitudinal purchase data with measured basis risk and discusses spatial and temporal selection. The study's findings indicate that demand responds to basis risk and that index contracts are not automatically immune to informed selection. Their external validity is confined by the product, ecology, informational environment, and measured household population.

Price, Trust, and Salience

A premium is paid with cash available now

Insurance may have positive expected risk-reduction value while remaining financially inaccessible at the purchase date. Agricultural cash flows are seasonal; households may need to buy seeds and fertilizer before harvest and may have little liquid cash when premiums are due. A long-run capacity to finance a small contract is different from liquidity on the enrollment day. These constraints also affect demand for bundled credit and insurance products.

Cole et al.'s price and liquidity interventions (2013) used randomized field experiments for rainfall insurance in rural India. Demand was significantly price sensitive, but the article concluded that price reductions alone were unlikely to yield widespread participation. The study identified evidence consistent with liquidity, trust, and salience as additional constraints. The earlier Federal Reserve working-paper analysis reports a baseline purchase range of 5 to 10 percent in the studied setting; that figure should not be generalized to all products or countries.

The relevant counterfactual separates a cash transfer, a premium subsidy, a payment-date adjustment, and a credit offer. A cash transfer relaxes household resources and may affect multiple expenditures. A subsidy changes the policy's relative price. A deferred premium changes liquidity timing and may alter default risk. A loan can create obligations that reduce the value of insurance for a credit-constrained borrower. These are distinct causal treatments despite superficially similar effects on initial purchases.

Relative price and effective coverage

The ratio of premium to expected payout is a weak standalone indicator of customer value because it ignores the temporal and household-specific correlation of payouts with loss. A policy with generous average payouts can deliver relatively little benefit to an individual whose livelihood loss does not align with the index. A more relevant comparison examines expected utility or an explicitly chosen residual-risk metric at the purchaser's actual exposure level.

Hill, Robles, and Ceballos (2016), in their Indian weather-hedge trial, found demand decreasing with price and basis risk and a hump-shaped relationship with risk aversion. The study provides experimental support for aspects of the rational partial-hedge account. It also indicates why comparisons across price treatments must control for index quality and heterogeneity rather than treating the purchase percentage as an isolated quality indicator.

Trust has identifiable institutional objects

Trust can attach to a seller, a community representative, a weather station, the calculation rule, a reinsurer, an administrative process, or an eventual payment. These objects can vary independently. A buyer may rely on a familiar local agent without understanding the carrier's obligations; a highly accurate index can fail commercially when settlement is delayed or payment records are opaque; a clearly regulated provider can still offer a trigger poorly matched to local losses.

In the Indian observational study, association with the existing sales institution was positively associated with purchase. Because customers of that institution may differ in credit access, social position, and sales exposure, the association should not be read as the causal effect of trust alone. Randomized communication or seller assignment is needed for more precise attribution.

Information about the insurer's previous payouts has two roles: it supplies evidence of contract performance and it changes the salience of a previously abstract contingency. A reliable audit separates "the index triggered" from "the insurer honored the trigger" and from "this farmer suffered a loss." Each observation answers a different question.

Salience is an attention mechanism, not a measure of risk

A vivid recent drought, a salesperson's example, a community payout ceremony, or visible damage on neighboring farms can make insurance more mentally available. That salience may help buyers attend to neglected protection needs. It can also temporarily overweight a rare event or produce miscalibrated expectations about contractual payment frequencies.

The Ghana farm investment and insurance study found later demand associated with own payouts, others' payouts, and recent adverse rainfall. That pattern is consistent with learning about trustworthiness, basis risk, and recent losses, but a simple correlation of renewed purchases with recent weather should not be treated as a separately randomized behavioral mechanism. The event both changes information and may change household resources or actual perceived exposure.

Farm Insurance Demand

Why agricultural evidence is central and bounded

The empirical literature on index insurance has concentrated on smallholder cropping and pastoral livestock in settings where independent loss adjustment can be costly and covariate climatic risk is economically consequential. These cases illuminate broad demand mechanisms but do not supply a universal elasticity for nonagricultural parametric products, sovereign protection, enterprise catastrophe policies, or advanced financial hedges.

Farm decisions are intertemporal. Insurance purchased before planting can change the perceived downside risk of a later decision about seed varieties, fertilizer, hired labor, or herd management. Outcomes should therefore include both insurance purchases and the subsequent productive choices that protection enables. Observed investment responses are informative, although stronger agricultural investment may not by itself prove that households are better protected in every possible state.

India: revealed choices and controlled interventions

The Gine, Townsend, and Vickery 2008 rainfall study found lower participation where basis risk was greater, along with positive associations for wealth and vendor familiarity. Its survey design provides rich correlational evidence, not randomized identification of every explanatory variable. Cole et al.'s 2013 experiments then used randomized marketing and pricing variation to identify purchase responses more rigorously. They found strong price responses but persistence of nonprice barriers.

A third Indian randomized weather-hedge study by Hill, Robles, and Ceballos (2016) helps adjudicate the role of basis risk and risk aversion under a comparatively simple contract. Across these studies, the analytical lesson is the value of multiple designs. Observational evidence identifies predictors and possible mechanisms; random variation in price or marketing supports specified causal effects; structural interpretation still depends on product quality, expectation measures, and the choice model.

Ghana: risk reduction and productive response

The northern Ghana cash-grant and rainfall-insurance experiments of Karlan et al. (2014) assigned cash grants, insurance access or grants, and combinations of these treatments. The study reports larger farm investments and riskier agricultural production choices among insured households, supporting the interpretation that uninsured risk can be a binding constraint on investment. This contrasts with a simple account in which poverty necessarily prevents all meaningful insurance demand. At the same time, treatment design and contract terms matter for distinguishing responses to gifted protection from voluntary commercial willingness to pay.

The Ghana findings on later demand connect insurance to an ongoing learning process: the event that makes protection observable may arrive only after purchase. An early absence of payouts is ambiguous to a customer. It may mean benign weather, a policy that did not trigger, or a mismatch between a personal loss and the measured index. Designs that observe both actual local losses and trigger performance are best equipped to interpret this experience.

Ethiopia: information, risk sharing, and price

Dercon, Hill, Clarke, Outes-Leon, and Taffesse (2014) offered rainfall insurance through informal risk-sharing groups and randomized training content provided to group leaders. Their Ethiopian field experiment found greater take-up when training emphasized the value of sharing residual idiosyncratic risk. The authors note that the measured effect can reflect selection of training participants as well as direct effects of learning. The distinction matters when assigning causal credit to a message.

The southern Ethiopian learning-kit and discount experiment by Takahashi et al. (2016) found improved knowledge without strong evidence of a resulting purchase increase, while price discounts had an immediate positive effect. The pair of studies demonstrates that different informational interventions should not be grouped together under an undifferentiated "financial education" label.

Kenya: measured basis risk and livestock protection

Pastoralist demand for index-based livestock insurance depends on drought conditions, regional vegetation signals, herd condition, liquidity, expected mortality, and remaining uninsurable idiosyncratic exposure. Jensen, Barrett, and Mude's longitudinal basis-risk assessment places contract quality at the center of evaluation. Jensen, Mude, and Barrett's subsequent household demand study associates lower basis risk with greater demand and reports evidence consistent with spatiotemporal selection. Information can allow households to respond more accurately to both index quality and anticipated weather conditions.

In a related but distinct comparative impact study, Jensen, Barrett, and Mude (2017) assessed cash transfers and livestock insurance in northern Kenya. Their reported effects differ by intervention, including outcomes in livestock management and household welfare. These results help place insurance among alternative protection instruments, but the impact estimates should not be reinterpreted as direct estimates of voluntary demand or as proof that one instrument universally dominates another.

Social networks and risk sharing

Formal indexed insurance can complement informal transfers when systemic drought payouts protect against broad common shocks while relatives or neighbors help offset household-specific failures. Dercon et al.'s Ethiopian risk-sharing intervention examines this complementarity directly. Mobarak and Rosenzweig's Indian network study (2013) reports preliminary findings on interactions among rainfall insurance, caste-based informal sharing, and risk-taking. Because the AER article is short and expressly preliminary, it cannot bear claims about universal welfare effects.

Social diffusion is also a separately testable information mechanism. Cai, de Janvry, and Sadoulet (2015) randomized intensive information sessions about weather insurance in rural China. Their social-network experiment found that an additional informed friend affected an untreated farmer's purchase probability by an amount equivalent to a 13 percent premium reduction. Additional experiments pointed to diffusion of product knowledge rather than imitation of peers' purchase decisions as the operative channel in that setting. Such a monetary equivalence is an estimate for that treatment and population, not a price elasticity transferable to all markets.

Payout Experience and Renewal

Insurance is an experience good with unusual feedback

A policyholder cannot fully observe contract quality at the moment of sale. A weather event, trigger determination, and final settlement may be necessary to learn whether a promise translates into financial protection. Yet many seasons pass without an informative test. A paid premium followed by no payout is often consistent with proper contract operation. A personal crop loss followed by no payout may likewise be contractually correct while revealing severe basis risk. Learning about product quality therefore requires more than comparing premiums and payouts over a single year.

The Ghanaian renewal evidence in Karlan et al. (2014) links subsequent demand with own and peer payout experience and recent poor rainfall. Because these events may change both beliefs and resources, a robust interpretation must inspect whether the observed information was experimentally varied, whether contemporaneous losses were controlled, and whether the buyer actually understood the trigger.

Good experiences and bad experiences have different informational content

A payout received after personal loss is evidence of successful protection in that state, although it cannot establish future reliability. A payout received without personal loss may help confirm that the payment formula is real while also generating concerns about fairness or the value of funds spent. A loss with no payout is especially informative about basis risk only when the contract's index, location, and period are known. A promised triggered payment that is delayed or withheld is evidence concerning fulfillment, a different dimension of quality.

A recent Ethiopian fairness experiment by Kramer, Porter, and Wassie (2025) examined basis risk, social comparisons, and the introduction of more complete crop insurance in a lab-in-the-field design. The study reports that information about neighbors' payout experiences increased perceived fairness; the more complete insurance option improved fairness assessments and willingness to pay without detecting the hypothesized jealousy in that setting. These are experimentally elicited perceptions and valuations; they are not estimates of long-run market renewal after real losses.

Renewal as a selection and learning process

Renewal decisions are conditioned on survival of the original policy, continued access to the product, changes in premium, weather, income, producer assets, and expected opportunity costs. Comparing repeat buyers with lapsed buyers without addressing these changing conditions risks attributing every lapse to dissatisfaction. A person can discontinue valuable insurance because of liquidity stress, a change of crop, sale of livestock, or migration. A person can renew a low-quality product under subsidy, compulsory bundling, or misunderstanding.

A credible renewal study records enrollment, paid renewal, sum insured, premium, discount, index outcomes, individual losses, actual payment dates, knowledge of conditions, and reason for lapse. Baseline expectations should be elicited before the season; otherwise recollection after a drought may be shaped by its outcome. Where random incentives affect initial purchase, downstream renewal analyses should distinguish the causal effect of the original offer from the selected behavior of those who took up coverage.

Trust, fairness, and institutional records

When policyholders see others paid, social comparison may update beliefs about the institution's reliability. It may also lead to incorrect expectations of personal eligibility. A transparent program should publish aggregate payout triggers and examples of both triggered and untriggered individual-loss scenarios in accessible language. Fairness has at least a contractual aspect, where all eligible policyholders are treated consistently, and a welfare aspect, where people suffering comparable losses may receive unequal protection because their indices differ.

The 2025 Ethiopian fairness and demand study provides empirical reason to measure perceived fairness independently of insurer-payment trust. Formal legality of the index rule does not answer the behavioral question of whether households view the resulting allocation as protective or legitimate.

Behavioral Interventions and Evaluation

Start by identifying the binding constraint

An intervention succeeds in a scientifically meaningful sense only if it addresses a demonstrated mechanism. A reduction in premium can alleviate relative-price constraints. A payment-date change can alleviate immediate cash shortages. More accurate satellite data, denser observation, or revised index timing can improve actual loss-index correspondence. A credible payment demonstration can update beliefs about settlement performance. Plain-language examples can improve comprehension. Those interventions may interact, but they target different margins.

The index insurance reassessment by Carter et al. (2017) points toward improved contract design, measurement technology, market support, and complementary savings and credit arrangements. Greatrex and colleagues' cross-country implementation synthesis (2015) describes varied scale-up arrangements. Such implementation cases document feasible delivery models and institutional conditions; they do not automatically establish the causal superiority of any particular distribution arrangement.

Price and payment experimentation

A randomized discount can identify the effect of a lower consumer price on current demand at that product's existing quality. It does not identify willingness to pay for a redesigned index. When price is randomized, estimates should report the subsidy distribution, baseline price, purchase window, and whether the discount persisted into renewals. To separate liquidity from price, an experiment can randomize payment timing holding present-value price approximately constant, while explicitly reporting any credit cost and default exposure.

The Indian randomized pricing study and the Ethiopian coupon experiment show that price effects can be detected, while large remaining nonprice constraints can persist. The correct inference is a conditional treatment effect, not a universal formula that a fixed subsidy percentage will create enduring financially protective coverage.

Information and trust interventions

Information campaigns should be tested on both knowledge and behavior. An improved quiz score may not produce demand if respondents already correctly perceive that the product leaves large residual loss exposure. Conversely, a rise in purchases after persuasive communication is not sufficient evidence of informed welfare improvement. Evaluation should disclose whether buyers understand the states in which they will be unpaid despite suffering loss.

For peer-based outreach, the Chinese social-information experiment demonstrates the value of separating information diffusion from purchase imitation. The Ethiopian group-training experiment further demonstrates that group composition and leader selection can influence estimates. Effective field design must track who received information, who transmitted it, whose decision is observed, and whether untreated households were indirectly exposed.

Change the contract when the contract is the obstacle

If verified records show high missed-payout rates conditional on severe individual loss, improved sales technique may increase short-term purchases without providing adequate protection. The first intervention should examine the match between the index and the risk to be covered. This could involve geographic refinement, multiple hazards, more appropriate timing, individual-loss components, or supplemental coverage. Every design change has cost, complexity, and possible moral-hazard implications. The analysis therefore requires a comparison of additional risk reduction with the additional premium and administrative burden.

A Kenyan hybrid insurance and seed experiment by Bulte, Cecchi, Lensink, Marr, and van Asseldonk (2020) found increased investments following free provision of a hybrid crop-insurance product conditional on certified seed purchase and reported higher subsequent stated willingness to pay. Because the product included index and indemnity elements and was bundled with seed acquisition, the results are specific to that treatment; they cannot be isolated as the effect of pure index insurance or of an insurance-only marketing campaign.

Evaluation beyond the purchase rate

The minimum research outcome set should include: verified purchase and premium payment; accurate comprehension of trigger and residual risk; contract-level loss-index performance; individual-loss and payout histories; payment latency; net household exposure after protection; renewal under commercially specified terms; and household production, consumption, or distress responses. Multiple outcome measures prevent a high enrollment rate from substituting for the product's central purpose.

Causal study designs should pre-specify the primary treatment, unit of randomization, spillover structure, attrition handling, statistical estimand, and key comparison groups. If communities are randomized, errors and inference should accommodate clustering. If households are exposed to friends' information, interference between units must be incorporated into the design. Where the purchaser can buy different coverage amounts, extensive and intensive margins should be reported separately. Stated willingness to pay should be identified as an elicitation result; actual demand requires real payment or a validated incentive-compatible mechanism.

Heterogeneity needs deliberate attention. Poorer households may face more binding liquidity and higher welfare stakes from residual losses. Different crops or herd types may have different index fit. Women and men may differ in control of premium funds, information channels, and asset ownership. An average causal effect can hide harms to those whose losses are least correlated with the index. Subgroup claims should be motivated, statistically supported, and treated cautiously when multiple comparisons are uncorrected.

Evidence strength and interpretation

Observational regressions are valuable for identifying correlated features of demand, particularly measured basis risk, wealth, and network ties, but face selection and omitted-variable problems. Randomized pricing and information interventions offer stronger causal identification of the manipulated margin, while remaining local to the assigned contracts and populations. Framed laboratory experiments clarify elicited preferences under constructed uncertainty, but external validity depends on how choices map to real money and real losses. Comparative program assessments and reviews establish design considerations and institutional experience without substituting for a randomized test of every recommended practice.

The review of basis-risk research and reassessment of agricultural indices converge on the need to assess actual coverage quality. The independent analytical implication is that demand evaluation should be performed on a documented joint distribution of household losses and index outcomes whenever such data are available, with claims about behavioral mechanisms made only to the extent the research design can distinguish them.

Conclusion

Demand for index-based insurance is a decision about the distribution of future household resources under an imperfect contractual hedge. Premiums, liquidity, physical basis risk, subjective understanding, ambiguity, institutional credibility, social relations, and past payout experience are jointly relevant. The literature demonstrates that demand can be low even among households exposed to severe weather risk, and that both rational assessment of an imperfect product and behavioral frictions contribute to observed patterns. A diagnosis that attributes weak purchases to a single cause is methodologically inadequate.

The sources establish several durable findings. In the Indian setting, randomized evidence identifies price, liquidity, and trust-related barriers; in Ghana, experimental insurance provision is associated with more productive risk-taking and payout-related learning; in Ethiopia, greater knowledge does not invariably produce more sales, while specific discounts and group risk-sharing information affect uptake; in Kenya, direct measurement confirms that material basis risk remains and influences demand; and in China, network information can transmit insurance knowledge with economically meaningful effects. These results are conditional on product design and study context, and none warrants extrapolation of a single demand estimate to every country or index type.

The governing research conclusion is that purchase, informed consent, contractual fulfillment, and effective loss reduction must be evaluated separately and then connected. An increase in take-up is beneficial only insofar as households obtain useful, comprehended, sustainable protection on terms consistent with their actual risk and financial circumstances. The primary institutional priority is to document loss-index fit, clearly explain nonpayment states, honor triggered obligations, and measure whether repeated transactions improve household financial resilience. These conclusions are analytical synthesis of verified scholarship; no new named theory, original field estimate, or external peer-review finding is claimed.

References

  • Bryan, G. (2019). Ambiguity aversion decreases the impact of partial insurance: Evidence from African farmers. Journal of the European Economic Association, 17(5), 1428-1469. https://doi.org/10.1093/jeea/jvy056

  • Bulte, E., Cecchi, F., Lensink, R., Marr, A., & van Asseldonk, M. (2020). Does bundling crop insurance with certified seeds crowd-in investments? Experimental evidence from Kenya. Journal of Economic Behavior & Organization, 180, 744-757. https://doi.org/10.1016/j.jebo.2019.07.006

  • Cai, J., de Janvry, A., & Sadoulet, E. (2015). Social networks and the decision to insure. American Economic Journal: Applied Economics, 7(2), 81-108. https://doi.org/10.1257/app.20130442

  • Carter, M., de Janvry, A., Sadoulet, E., & Sarris, A. (2017). Index insurance for developing country agriculture: A reassessment. Annual Review of Resource Economics, 9, 421-438. https://doi.org/10.1146/annurev-resource-100516-053352

  • Clarke, D. J. (2016). A theory of rational demand for index insurance. American Economic Journal: Microeconomics, 8(1), 283-306. https://doi.org/10.1257/mic.20140103

  • Cole, S., Gine, X., Tobacman, J., Topalova, P., Townsend, R., & Vickery, J. (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

  • Dercon, S., Hill, R. V., Clarke, D., Outes-Leon, I., & Taffesse, A. S. (2014). Offering rainfall insurance to informal insurance groups: Evidence from a field experiment in Ethiopia. Journal of Development Economics, 106, 132-143. https://doi.org/10.1016/j.jdeveco.2013.09.006

  • Elabed, G., & Carter, M. R. (2015). Compound-risk aversion, ambiguity and the willingness to pay for microinsurance. Journal of Economic Behavior & Organization, 118, 150-166. https://doi.org/10.1016/j.jebo.2015.03.002

  • Gine, X., Townsend, R., & Vickery, J. (2008). Patterns of rainfall insurance participation in rural India. The World Bank Economic Review, 22(3), 539-566. https://doi.org/10.1093/wber/lhn015

  • Greatrex, H., Hansen, J., Garvin, S., Diro, R., Le Guen, M., Blakeley, S., Rao, K., & Osgood, D. (2015). Scaling up index insurance for smallholder farmers: Recent evidence and insights (CCAFS Report No. 14). CGIAR Research Program on Climate Change, Agriculture and Food Security. https://cgspace.cgiar.org/items/34d65f48-ad14-409b-9e4e-b6da7d28adda

  • Hill, R. V., Robles, M., & Ceballos, F. (2016). Demand for a simple weather insurance product in India: Theory and evidence. American Journal of Agricultural Economics, 98(4), 1250-1270. https://doi.org/10.1093/ajae/aaw031

  • Jensen, N. D., Barrett, C. B., & Mude, A. G. (2016). Index insurance quality and basis risk: Evidence from northern Kenya. American Journal of Agricultural Economics, 98(5), 1450-1469. https://doi.org/10.1093/ajae/aaw046

  • Jensen, N. D., Barrett, C. B., & Mude, A. G. (2017). Cash transfers and index insurance: A comparative impact analysis from northern Kenya. Journal of Development Economics, 129, 14-28. https://doi.org/10.1016/j.jdeveco.2017.08.002

  • Jensen, N. D., Mude, A. G., & Barrett, C. B. (2018). How basis risk and spatiotemporal adverse selection influence demand for index insurance: Evidence from northern Kenya. Food Policy, 74, 172-198. https://doi.org/10.1016/j.foodpol.2018.01.002

  • Karlan, D., Osei, R., Osei-Akoto, I., & Udry, C. (2014). Agricultural decisions after relaxing credit and risk constraints. The Quarterly Journal of Economics, 129(2), 597-652. https://doi.org/10.1093/qje/qju002

  • Kramer, B., Porter, M., & Wassie, S. B. (2025). Basis risk, social comparison, perceptions of fairness, and demand for insurance: A field experiment in Ethiopia. Journal of Risk and Insurance, advance online publication, July 31. https://doi.org/10.1111/jori.70015

  • Mobarak, A. M., & Rosenzweig, M. R. (2013). Informal risk sharing, index insurance, and risk taking in developing countries. American Economic Review, 103(3), 375-380. https://doi.org/10.1257/aer.103.3.375

  • Takahashi, K., Ikegami, M., Sheahan, M., & Barrett, C. B. (2016). Experimental evidence on the drivers of index-based livestock insurance demand in southern Ethiopia. World Development, 78, 324-340. https://doi.org/10.1016/j.worlddev.2015.10.039

  • Yuzva, K., Botzen, W. J. W., Brouwer, R., & Aerts, J. C. J. H. (2018). A global review of the impact of basis risk on the functioning of and demand for index insurance. International Journal of Disaster Risk Reduction, 28, 845-853. https://doi.org/10.1016/j.ijdrr.2018.01.001

  • Cole, S., Gine, X., Tobacman, J., Topalova, P., Townsend, R., & Vickery, J. (2009). Barriers to household risk management: Evidence from India (Staff Report No. 373; working-paper version of the 2013 article). Federal Reserve Bank of New York. https://www.newyorkfed.org/research/staff_reports/sr373.html