Insuring home from afar: Experimental evidence on migrants’ demand for weather index insurance

This study examines whether rural–to-urban migrants are willing to take up formal, weather-indexed financial products on behalf of their rural origin households, and how such products interact with existing informal risk-sharing arrangements such as remittances. The intervention targets migrants residing in urban locations who originate from agricultural households that are exposed to climate-related risks, including droughts and floods.
Smallholder households in developing countries frequently rely on informal insurance mechanisms—most notably financial support from migrant family members (“remittances”) —to cope with agricultural income (weather) shocks. While such transfers play an important role in household risk management, they may be unreliable under covariate shocks and can place substantial financial pressure on migrants. Formal agricultural insurance offers a potential alternative, but uptake remains persistently low. Beyond high costs, limited trust in insurers, and high discounting of future payment, one substantial reason may be that households perceive themselves as sufficiently insured by their family network, notably migrants.
This creates externalities for urban migrants: they act as de facto risk-bearers for their rural households but are often not formally involved in insurance adoption decisions. The present intervention seeks to explore how migrants make choices about taking up formal, weather-indexed insurance for their origin households through a set of experimentally structured financial decisions. Specifically, the study investigates (i) how migrants’ uptake of insurance is related to migrant, household, and migrant–household dyad characteristics, including past exposure to shocks, socio-demographics, interlinkages between the household and migrant, and past integration in the financial system; (ii) how time preferences, altruistic motives, and risk perceptions shape financial decisions; (iii) how formal insurance interacts with existing informal support mechanisms, such as remittances; and (iv) whether uptake is influenced by trust in insurance providers and by alternative product framings (“insurance” versus “protection product”).

The intervention is conducted within a migrant tracking survey that is part of the long-running TVSEP panel in Thailand and Vietnam. Eligible participants are adult rural–urban migrants whose households of origin reside in six rural provinces across the two countries and who were interviewed during the 2024 household wave (and in most cases in multiple prior waves of the 18-year-long panel). During the migrant interviews in 2025/2026, respondents are presented with a series of structured financial decision scenarios involving real monetary stakes (via a Random Incentive System). These scenarios allow migrants to allocate a lump-sum endowment either to themselves or toward financial products that benefit their rural household.

A central component of the intervention is an offer to purchase a weather-indexed, insurance-like product for the migrant’s household of origin. If selected and taken up, the product pays out automatically to the household in the event of an extreme weather shock occurring in the household’s village within a 6-month future period. Weather shocks are measured using a publicly available, objective precipitation–evapotranspiration index (SPEI-6), ensuring transparent and verifiable trigger conditions. The product is designed to provide protection against both drought and excessive rainfall, depending on local climatic conditions.
In addition to the insurance offer, migrants are asked to make comparable decisions involving an immediate transfer to the household and a delayed but guaranteed savings payout to the household. These additional options provide benchmarks to understand how migrants value immediate support, future certainty, and contingent risk protection when making financial decisions for their families.
To ensure incentive compatibility while managing the long-term panel structure of TVSEP, only a randomly selected subset of respondents and decision scenarios are implemented with real financial payouts. All participants are informed about the random selection process before making their choices, and are told at the end of the survey whether they have been selected for a payout. Clear explanations and comprehension checks are used to ensure respondents understand the nature of the products, the payout conditions, and the timing of any transfers.
In addition to the incentivized decision tasks, migrants report the amount of remittances they would send back to their rural household under two hypothetical scenarios: (a) the household experiences a shock and is insured, or (b) the household experiences a shock but is not insured. The order of these scenarios is randomly varied.
The study also includes a framing variation to test whether trust and perceptions influence uptake. Migrants are randomly assigned to receive the same weather-indexed product described either using standard “insurance” terminology or using neutral language that frames the product as a weather “protection product,” while keeping all contractual features identical.

By directly involving migrants in formal risk management decisions for their origin households, the intervention aims to assess (i) the willingness of migrants to purchase insurance-like products, (ii) behavioral trade-offs in the decision to purchase insurance products for the household; (iii) how formal insurance interacts with informal remittance-based support, and (iv) whether trust and framing affect demand. While implemented in the context of rural–urban migration in Southeast Asia, the intervention speaks more broadly to settings where informal family networks play a central role in coping with economic risk and where formal insurance markets face persistent adoption barriers.

2025-10-01

2026-02-15

Our primary measures of interest are:

(1) Insurance take-up by migrants (variable s1_2) (descriptive)
a. with heterogeneity analysis by migrant, household and migrant-household-dyad characteristics

(2) Relative preferences across products (s1_1, s1_2, s1_3) to measure behavioral trade-offs that influence the decision to take up any of the products (causal, within-subject)
a. Intertemporal preferences: s1_1 versus s1_3
b. Risk-related trade-offs: s1_1 versus s1_2

(3) Expected remittance response to insurance under a hypothetical weather shock (variables s2_1 and s2_2) (causal, within-subject)

(4) Insurance take-up by migrants (variable s1_2), impacted by framing as “insurance” or “protection product” (causal, between-subject)

All primary outcomes are drawn from the Financial Decisions Module (Section 13.1) of the migrant questionnaire. Heterogeneity variables are sourced from other sections of the Migrant Tracking Survey, as well as multiple waves of the household survey panel.

(1) Insurance take-up is measured using a binary indicator capturing whether the migrant allocates the experimental endowment to the weather-indexed insurance product (variable s1_2). This outcome captures migrants’ revealed demand for formal weather risk protection for their origin households under incentive-compatible conditions.

a. We will analyze heterogeneity in insurance take-up descriptively by linking migrant responses to longitudinal household data from the TVSEP panel. This linkage allows us to relate migrants’ insurance choices to the household’s historical exposure to shocks, the migrant’s own socio-demographic characteristics; as well as dyad-specific characteristics, like prior reliance on remittances, and the strength of migrant–household ties.
In particular, heterogeneity analyses will consider variation along the following dimensions:
- Migrant’s socio-demographic characteristics (e.g. income group, gender, age group)
- Migrant’s financial literacy with respect to insurance (v53010) and degree of integration into formal financial systems (savings/ insurance from section 8; v53004 for insurance)
- Migrant’s trust in insurance companies (v53009)
- Migrant’s time preferences (v72221*)
- Duration of the migrant’s absence from the origin household (v31001)
- Past intensity of migrant–household interactions, including visit frequency and remittance behavior (v21020 - v21022; v51003 and v51003a)
- Migrant’s knowledge about farming practices of his household (v21023)
- Exposure of the household to weather shocks prior to the migrant’s departure; as well as after the migrant’s departure (linking shock history of household panel with migrant ID); and shortly before the migrant interview
- The extent of informal insurance provided by the migrant in response to past shocks (linking past migrant’s remittances from household panel with migrant ID)
- The relative importance of the migrant’s financial support for the household (proxies: share of remittances provided by the migrant; single-migrant versus multi-migrant households)
- Household financial literacy and access to formal financial services (savings/ insurance modules from household survey 2024)
These analyses are intended to characterize patterns of insurance demand and to provide context for the interpretation of the causal estimates described below.


(2) Behavioral trade-offs aim to capture mechanisms behind migrant financial decisions. Using comparisons between s1_1 and s1_3 (remittance today versus in the future), as well as s1_3 and s1_2 (future payment for sure versus future payment contingent on a weather shock), we quantify two key behavioral trade-offs:

a. Intertemporal preferences, corresponding to migrant’s willingness to delay transfers and commitment to the household over time (present bias)
→ comparison between immediate remittance (s1_1) and delayed savings product (s1_3)

b. Risk-related trade-offs, measuring whether conditional on having delayed the pay-out to the future, the migrant values protection against weather shocks
→ comparison between delayed savings (s1_3) and insurance (s1_2)

These comparisons are implemented in a joint regression where each decision is considered as a separate observation (with three observations per migrant). The dependent variable indicates the willingness to transfer resources to the rural household under each of the three scenarios (s1_1, s1_2, s1_3), and the explanatory variables record either the different scenarios (s1_2, s1_3), or indicators for future payouts (s1_2, s1_3) and uncertain future payouts (s1_3).


(3) Expected remittance response to insurance under a hypothetical weather shock captures how migrants adjust intended remittances when their household is insured versus uninsured.
Specifically, we regress the remittance value (variables s2_1 and s2_2) on the presence of insurance in hypothetical scenario s2_2. A positive coefficient indicates that insurance crowds in additional remittances, whereas a negative value indicates potential crowd-out.


(4) The effect of framing on insurance take-up will be analyzed using variable s1_2, comparing take-up between the two experimental conditions (“insurance” vs. “protection product”). This analysis is causal and uses a between-subject design, allowing us to assess whether the way the product is presented influences migrants’ decisions to take up insurance.

(1) Heterogeneity analyses of primary outcomes across migrant, household, and migrant–household dyad characteristics

(2) Order effects in decision-making
a.Order of financial decision tasks (s1_1: immediate remittance, s1_2: insurance, s1_3: delayed savings)
b. Order of hypothetical remittance scenarios with and without insurance (s2_1: without insurance, s2_2: with insurance)

Secondary outcomes are intended to support the interpretation and internal validity of the primary findings.


(1) Heterogeneity analyses will examine how the primary outcomes—insurance take-up, behavioral trade-offs, and intended remittance responses—vary across observable migrant, household, and migrant–household dyad characteristics. While these analyses are not intended to detect causal subgroup effects, they provide descriptive evidence on which types of migrants and households are more or less responsive to the experimental interventions.
In particular, heterogeneity analyses will again consider variation along the following dimensions:
- Migrant’s socio-demographic characteristics (e.g. income group, gender, age group)
- Migrant’s financial literacy with respect to insurance (v53010) and degree of integration into formal financial systems (savings/ insurance from section 8; v53004 for insurance)
- Migrant’s trust in insurance companies (v53009)
- Migrant’s time preferences (v72221*)
- Duration of the migrant’s absence from the origin household (v31001)
- Past intensity of migrant–household interactions, including visit frequency and remittance behavior (v21020 - v21022; v51003 and v51003a)
- Migrant’s knowledge about farming practices of his household (v21023)
- Exposure of the household to weather shocks prior to the migrant’s departure; as well as after the migrant’s departure (linking shock history of household panel with migrant ID); and shortly before the migrant interview
- The extent of informal insurance provided by the migrant in response to past shocks (linking past migrant’s remittances from household panel with migrant ID)
- The relative importance of the migrant’s financial support for the household (proxies: share of remittances provided by the migrant; single-migrant versus multi-migrant households)
- Household financial literacy and access to formal financial services (savings/ insurance modules from household survey 2024)


(2) Order effects capture whether the sequence in which decision tasks are presented influences respondents’ choices. We will examine whether insurance take-up, remittance decisions, and behavioral trade-offs vary systematically with the randomized order of the financial decision modules (s1_1, s1_2, s1_3), as well as with the order in which the hypothetical remittance scenarios with and without insurance (s2_1 and s2_2) are shown. These analyses serve to assess potential framing or anchoring effects induced by task ordering and to verify that the primary results are not driven by the order in which scenarios were presented.

The intervention leverages the long-running TVSEP migrant tracking survey to implement a multi-component, incentive-compatible experiment on migrants’ financial choices and demand for weather risk protection. The design is tailored to uncover how migrants trade off their own consumption, support to their origin households, and formal insurance. The intervention further allows analysis of whether formal insurance and informal support through remittances act as substitutes or complements, and whether framing and trust-related perceptions influence migrants’ willingness to allocate resources to formal weather risk protection.

(1) At the core of the design are three incentivized financial decision tasks: an immediate transfer, a delayed transfer (savings product), and a weather-indexed product. The order of these tasks is randomly assigned to each respondent to mitigate ordering effects and to enable within-person comparisons across different financial contexts.
​
(2) To ensure incentive compatibility, the study employs a random incentive system in which a subset of respondents (20 per country) is randomly selected to receive real monetary payouts based on one of their decisions from step (1), with insurance-related payouts determined ex post by realized weather outcomes. This maintains a meaningful probability that any given choice is payoff-relevant, while limiting financial expenditures and avoiding spillovers to the ongoing household panel survey.
​
(3) Beyond these incentivized choices, migrants state hypothetical remittance amounts under two scenarios—household shock with insurance and household shock without insurance—with the scenario order again being randomized to avoid systematic ordering effects.

(4) A key treatment concerns how the weather-indexed product is described. Migrants are randomly assigned to see the same contract presented either with standard “insurance” terminology or as a neutral “weather protection product,” with price, trigger, and payout features held constant to isolate the causal effect of terminology and perceived trust on uptake.

Combining randomization over task order, framing, and payout selection generates exogenous variation that supports estimation of framing effects on uptake, the impact of insurance availability on expected remittances (link between formal and informal insurance mechanisms), and the relationship between preferences and behavior across tasks.

Not available

Computer-based
Ex-ante at individual migrant level for task order, framing, and RIS payout.

Individual migrants (sole unit of randomization). All treatments—task order, framing, and RIS payout selection—are independently randomized at the individual level with equal probability across groups.

No

-

- Thailand: from 1,115 migrants ca. 300 to be reached - Vietnam: from 697 migrants, ca. 200 to be reached The sampling frame consists of migrants reported as absent by their origin households and residing in the study’s target urban areas at the time of the 2024 TVSEP household survey. Randomization is implemented ex ante on this full frame. Substantial attrition is expected from three stages: (i) households providing valid migrant contacts, (ii) migrants remaining absent in target urban areas one year later, and (iii) migrants consenting to interviews once contacted. Common challenges include scam-call reluctance, frequent phone number changes, and mediation through origin households Based on prior TVSEP migrant tracking waves, we expect a final analytic sample of approximately 150–400 completed migrant interviews per country. Power calculations will be based on a realistic combined sample of about 500 migrants across both countries. To ensure data quality, we collected survey process indicators such as time spent on the Financial Decisions Module and performance on comprehension questions to assess whether respondents meaningfully engaged with the experimental tasks. As a robustness check, we may exclude respondents who exhibit clear difficulties in understanding the decision scenarios or who complete the module implausibly quickly from the main analysis sample. These criteria can shrink the sample further.

We analyze four separate aspects:
(i) Willingness of migrants to purchase insurance-like products
(ii) Behavioral trade-offs in decisions to purchase insurance products
(iii) Interaction between formal insurance and informal remittance-based support
(iv) Effects of trust and framing on insurance demand

For (i), we conduct a descriptive analysis of insurance uptake using the full analytic sample, comprising approximately 500 individuals in total.
Analyses (ii) and (iii) rely on a within-subject design, comparing each individual’s decisions across different questions and scenarios. The sample for these analyses also includes approximately 500 individuals, with 3 decisions per individual in (ii) and 2 decisions per individual in (iii).
Analysis (iv) employs a between-subject design, comparing equally sized treatment groups. Given the randomization, each treatment arm includes roughly 250 individuals.

We calculated the minimum detectable effect sizes (MDEs) for our main outcomes, accounting for sample design, clustering, and within-respondent correlations. Detailed calculations are provided in the attached PDF. (i) For descriptive analyses of insurance or protection product uptake, we examine correlations with baseline characteristics of respondents, households, and respondent-household dyads. These analyses are purely descriptive and are not intended for causal inference. (ii) In the within-subject behavioral choice task, respondents make repeated decisions between immediate transfers, savings, or insurance under scenarios varying in payment delay and uncertainty. With a planned sample of 500 respondents and a conservative within-individual correlation of 0.5, the study is powered to detect changes in choice probabilities of approximately 4 to 6 percentage points, depending on the baseline uptake rate. (iii) For the within-subject formal insurance and remittance decision, respondents report remittances to a rural household under scenarios with and without insurance. Using the same sample size and correlation assumptions, the study can detect a within-individual difference of roughly 0.13 standard deviations in remittances. (iv) Finally, for the between-subject framing experiment, respondents are randomly assigned to view the product as either “insurance” or a “protection product,” with 250 respondents per group. The study is powered to detect absolute differences in take-up of approximately 8 to 13 percentage points, depending on baseline uptake rates.