## _wp12195

## Source details

**Canonical URL:** [_wp12195](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12195.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12195.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12195.pdf.json)

---

### Introduction and research question
- Purpose: study an innovative financial contract (rainfall index insurance) designed to insure rural Indian households against rainfall variation during the monsoon season.
- Context: drought cited by 89 percent of households in the sample as the most important risk they face.
- Product accessibility: policies sold in unit sizes as small as US$1.
- Core research question: What frictions limit adoption of financial products that pool important sources of household income risk?

### Main empirical findings on demand and frictions
- Price elasticity:
  - A ten percent price decline leads to a ten to twelve percent increase in take-up.
  - Point estimates imply demand would increase by 36 to 66 percent if priced at payout ratios similar to US retail insurance contracts.
- Persistent non-price frictions:
  - Even with very large price discounts, less than half of households purchased insurance in a subset of the sample despite an expected return of around 70 percent.
  - Households almost universally purchase only one policy unit, indicating non-price barriers limit scale of uptake per adopter.
- Trust and information:
  - Households do not fully trust or understand the insurance product; trust significantly affects demand.
  - Randomized endorsement: demand is 36 percent higher when an insurance educator is first recommended by a trusted local agent.
  - Households answer simple arithmetic questions correctly 60 percent of the time (panel evidence), indicating limited numeracy/financial literacy.
  - Demand higher in villages that previously experienced a payout and among households with prior insurance experience, higher financial literacy, and greater facility with probability concepts.
- Liquidity constraints and salience:
  - Providing enough cash to buy one policy increases take-up by 140 percent.
  - The cash reward effect is magnified among poor households.
  - Adopters generally buy a single policy; non-purchasers most frequently cite “lack of funds” as reason for not buying.
  - Receiving a product flyer or a visit from an insurance educator increases take-up significantly, consistent with limited attention or salience effects.
- Education and framing interventions:
  - A short insurance education module has no significant effect on demand.
  - Point estimates for the framing effects considered are generally close to zero; standard-error bounds imply smaller effects than those found in Bertrand et al. (2010), where comparable.

### Insurance contract design and delivery
- Contract type: index insurance (payouts linked to publicly observable rainfall index).
- Underwriters:
  - ICICI Lombard in Andhra Pradesh.
  - IFFCO-Tokio in Gujarat.
- Measurement: payoffs calculated based on measured rainfall at a nearby government rainfall station or an automated rain gauge operated by a private third-party vendor.
- ICICI policy structure (Andhra Pradesh):
  - Monsoon divided into three contiguous phases.
  - Phases I and II cover deficit rainfall: pay Rs. 10 for each mm below a pre-specified strike; if rain is below the exit level, a contract pays Rs. 1,000.
  - Phase III covers excess rainfall: pays Rs. 10 for each mm above the strike, with a maximum payout of 1,000 if rainfall meets or exceeds the exit level.
  - Over 60 percent of adopters in the sample purchased Phase I.
- IFFCO-Tokio policy structure (Gujarat):
  - Based on cumulative rainfall over the entire monsoon season (June 1 to August 31).
  - Payout is a non-linear function of the percentage shortfall from a specified "normal" rainfall level.
- Marketing and sales channels:
  - Gujarat: marketed by SEWA.
  - Andhra Pradesh: sold by BASIX via local Livelihood Services Agents (LSAs).
- Payout disbursement:
  - Gujarat: SEWA anticipated visiting households individually to make payouts.
  - Andhra Pradesh: payouts disbursed in a public ceremony.

### Actuarial values, observed payouts, and pricing
- Expected payout estimates (historical calculation): point estimates range from 33 to 57 percent of premiums, averaging 46 percent.
- Comparison to U.S. retail insurance:
  - U.S. retail insurance contract payout ratios average 65 to 75 percent (examples cited: private passenger auto liability 76.2 percent, private passenger auto physical damage 68.4 percent, homeowners insurance 64.7 percent; earthquake 20.4 percent; crop insurance aggregate claims-to-premiums ratio 244 percent).
- Observed payouts in study:
  - In Gujarat, sufficient rain in 2006 and 2007 meant no payout was triggered.
  - In Andhra Pradesh, at least one positive payout was observed for each rainfall station between 2004 and 2006, with payouts ranging from Rs. 40 to Rs. 1,796 per policy.
- Historical administrative evidence:
  - Giné et al. (2007): policy-holders obtain a positive return in only 11 percent of phases; maximum return observed in about 1 percent of phases is 900 percent.
  - Giné et al. (2012) (BASIX data 2003–2009): average ratio of total insurance payouts to total premiums of 138 percent for BASIX policies.

### Sample, setting, and descriptive statistics
- Andhra Pradesh sample:
  - Districts: Mahbubnagar and Anantapur.
  - Summary statistics based on a survey of 1,047 landowner households in 37 villages (survey conducted in 2006; original selection in 2004 from a census of approximately 7,000 landowner households).
  - Same set of households used for field experiments.
- Gujarat sample:
  - Districts: Ahmedabad, Anand, and Patan.
  - Villages selected where SEWA operated and within 30 km of a rainfall station.
  - Summary statistics based on a baseline survey of 1,500 SEWA members in 100 villages (conducted in May).
- Selected household statistics (exact figures preserved):
  - Household size: Andhra Pradesh mean 6.26, St. Dev. 2.82; Gujarat mean 5.85, St. Dev. 2.39.
  - Scheduled Caste or Scheduled Tribe (1=Yes): Andhra Pradesh 11.60%, St. Dev. 32.04%; Gujarat 43.70%, St. Dev. 49.60%.
  - Monthly per capita food expenditures: Andhra Pradesh mean 310.53, St. Dev. 126.89; Gujarat mean 555.37, St. Dev. 417.42.
  - Total value of all savings deposits: Andhra Pradesh mean 1,030.42, St. Dev. 2,891.43; Gujarat mean 1,060.13, St. Dev. 2,314.97.
  - Land holdings (in acres): Andhra Pradesh mean 6.31, St. Dev. 6.17; Gujarat mean 4.11, St. Dev. 5.49.
  - Pct. of cultivated land that is irrigated: Andhra Pradesh 43.93%, St. Dev. 43.26%; Gujarat 43.70%, St. Dev. 47.10%.
  - Household bought weather insurance in 2004 (1=Yes): Andhra Pradesh 25.31%, St. Dev. 43.50% (Gujarat n.a.).
  - Household has some type of insurance (1=Yes): Andhra Pradesh 80.54%, St. Dev. 39.25%; Gujarat 63.78%, St. Dev. 48.08%.

### Education, financial literacy, cognition, and comprehension
- Education levels (selected):
  - Primary school or below: Andhra Pradesh 66.8%; Gujarat 42.0%.
  - College or above: Andhra Pradesh 7.4%; Gujarat 17.6%.
- Test scores and comprehension (exact figures preserved):
  - Average Score, Math Questions (Gujarat): 61.7% and another entry 71.8% (table layout preserved).
  - Average Score, Financial Literacy (Gujarat): 35.8%.
  - Average Score, Insurance Questions: Andhra Pradesh 79.3%; Gujarat 68.2%.
  - Average Score, Probability Questions (Gujarat): 59.1%.
- Specific insurance comprehension (Andhra Pradesh exact shares):
  - a) It rains 120 mm. Will you get an insurance payout? [Ans: No] Correct: 85.8%.
  - b.i) It does not rain at all: Will you get an insurance payout? [Ans: Yes] Correct: 83.0%.
  - b.ii) How much of a payout would you receive? [Ans: Rs. 500] Correct: 80.6%.
  - c.i) It rains 20mm: Will you get an insurance payout? [Ans: Yes] Correct: 81.5%.
  - c.ii) How much of a payout would you receive? [Ans: Rs. 200] Correct: 76.0%.
- Interpretation: substantial fractions understand contract mechanics, but math/probability skills are limited and correlated with purchase.

### Experimental design — Andhra Pradesh (2006) treatments and findings
- Design highlights:
  - 700 households randomly selected to be visited by trained ICRISAT insurance educators; visits completed for 660 households.
  - Three randomized dimensions for visited households:
    1. Endorsement by local BASIX LSA (two-thirds of villages endorsement-eligible; within these villages the LSA endorsed the educator for half the visits). Instructions followed exactly in 56 percent of cases; 25 percent did not show up or stayed too short a time; 19 percent stayed for the duration.
    2. Cash compensation for time: randomly Rs. 25 or Rs. 100 (half received Rs. 100). Rs. 100 roughly enough cash-on-hand to purchase one policy (premiums range between Rs. 80 and Rs. 125).
    3. Additional education on conversion between soil moisture and millimeters of rainfall given to 350 households.
- Key experimental take-up effects (exact coefficients preserved from Table 5 summary):
  - Visit (1=Yes): 0.172***; 0.128***; 0.115***; 0.117***; 0.114***; 0.118*** across specifications.
  - Endorsed by LSA (1=Yes): coefficients range from 0.064 to 0.194 across columns.
  - Education module (1=Yes): coefficients around 0.003 to 0.007, and -0.003 in one column (economically small, statistically insignificant).
  - High reward (1=Yes): 0.408**; 0.400***; 0.394***; 0.387***; 0.393***; and 1.629*** in reported columns (robust standard errors reported in table).
- Interpretation:
  - Being assigned a household visit increases take-up by 11.5 to 17.2 percentage points.
  - A high reward increases take-up by 39.4 to 40.8 percentage points; estimates statistically significant at the 1 percent level in many specifications.
  - LSA endorsement positively signed and marginally significant (t-stat between 1.5 and 1.7); joint tests of LSA-endorsement and village endorsement significant at 2 percent and 1 percent in some columns, indicating village-level spillovers.
  - Education module effect economically small and statistically insignificant.
- Interaction findings:
  - LSA endorsement increases take-up by 10.1 percentage points for households familiar with BASIX (statistically significant at the 5 percent level); effect absent for households unfamiliar with BASIX.
  - High cash reward effect larger among poor households (significance reported at 10 percent in one column and at 1 percent in another).
- Sample and model stats:
  - Mean of dependent variable: 0.282.
  - R-squared values: 0.279, 0.355, 0.380, 0.384, 0.382, 0.387.
  - Observations: 1047.
  - Significance notation: *, **, and *** indicate significance at the 10, 5 and 1 percent level, respectively.

### Experimental design — Gujarat (2007) treatments and findings
- Design highlights:
  - Field experiments in 50 villages: 30 offered insurance in 2006 and an additional 20 randomly offered insurance in 2007.
  - In 20 new-insurance villages: portable video players delivered a 90-second marketing message; each treated household randomly assigned one of eight different videos.
  - In 30 villages with prior offering: flyers distributed, containing one of six randomly assigned messages.
  - All treated households received a non-transferable coupon; discount coupon randomization in video villages: 40 percent received Rs. 5; 40 percent Rs. 15; 20 percent Rs. 30. In flyer villages discount fixed at Rs. 5.
- Video treatment dimensions:
  - SEWA Brand (Yes/No); Peer vs. Authority Figure; Payout framing (“2/10 yes” or “8/10 no”); Safety vs. Vulnerability framing.
- Flyer treatment dimensions:
  - Religion (Hindu, Muslim, or Neutral) and Individual vs. Group emphasis.
- Key empirical findings — video experiments (exact figures preserved from Table 6 summary):
  - Overall take-up rate: 29.4 percent.
  - Discount (fraction of initial price) coefficients:
    - Column (1): 0.307*** (0.076)
    - Column (2): 0.340*** (0.075)
    - Column (3): 0.372** (0.148)
    - Column (4): 0.405** (0.151)
  - Implied price elasticity of demand reported as 1.04 and 1.16.
  - Interpretation: coefficient 0.307 implies a 10 percent decline in price increases purchase probability by 3.07 percentage points, or 10.4 percent of baseline take-up; implied elasticity 1.04 (or 1.16 in column (2)).
  - Framing and brand effects: Strong SEWA Brand coefficients -0.026, -0.031, -0.081*, -0.082* (not robustly positive); Vulnerability Frame coefficients 0.046, 0.041, 0.131, 0.134 (not statistically significant); Positive Frame coefficients -0.027, -0.035, -0.037, -0.049 (not statistically significant); Peer Endorsed coefficients mixed and not statistically significant.
  - Interactions: Percentage Discount x Strong SEWA Brand: 0.258** and 0.236* in reported columns.
  - Surveyed Household coefficients: 0.159**, 0.179**, 0.207***, 0.210*** — households in 2006 baseline survey significantly more likely to purchase.
- Panel B district-level examples (exact entries):
  - For Rs. 5 discount: Return (gross) 61%; Take-up 25% (and another entry 22%).
  - For Rs. 15 discount: Return (gross) 82%; Take-up 37% (and another entry 22%).
  - For Rs. 30 discount: Return (gross) 169%; Take-up 47% (other entries: 30% and 44%).
  - Text note: for farmers in Ahmedabad receiving the Rs. 30 discount, estimated expected payouts are 169 percent of net premiums; fewer than half chose to buy insurance.
- Flyer experiments (Table 7 summary, exact figures preserved):
  - Overall take-up among flyer-receiving households: 23.8 percent.
  - Baseline flyer treatments (religion cue; group vs. individual) not statistically significant in simple specifications.
  - Interaction effects:
    - Group emphasis combined with neutral religious setting has a significant positive effect.
    - Muslim religious setting on the flyer reduces take-up by 9-10 percentage points (statistically significant at the 5 percent level).
  - Heterogeneous effects by respondent religious identity:
    - Muslim households receiving group-emphasis flyer: inclusion of Hindu symbols reduces take-up by 32.8 or 34.2 percentage points compared to neutral.
    - Hindu households receiving group-emphasis flyer: inclusion of Muslim symbols reduces take-up by 10.1 or 9.6 percentage points compared to neutral.
  - Notes: 219 respondents with coder disagreement omitted from some columns.

### Synthesis of experimental implications
- Marketing and price matter:
  - Direct household visits increase take-up by 11.5 to 17.2 percentage points.
  - Large cash rewards increase take-up by 39.4 to 40.8 percentage points.
  - Price discounts have sizable and statistically significant effects; implied elasticities around 1.04–1.16.
- Information and framing:
  - Short education module ineffective in Andhra Pradesh.
  - Video framing manipulations generally jointly insignificant in Gujarat.
- Trust, endorsement, and familiarity:
  - LSA endorsement increases take-up among households familiar with BASIX; endorsement produces village-level spillovers.
  - SEWA brand prominence in video treatments does not uniformly increase take-up.
- Social identity:
  - Group-emphasis flyers increase take-up when identity cues match household; mismatched religious cues substantially reduce take-up.
- Behavioral and exposure effects:
  - Prior survey exposure strongly associated with higher take-up; being assigned a visit raises take-up even when product available to all, consistent with salience/attention effects.

### Trust, liquidity, and behavioral mechanisms
- Trust:
  - LSA endorsement has no demand effect among farmers unfamiliar with BASIX.
  - LSA endorsement increases take-up by 10.1 percentage points (10.1 percentage points equals 36 percent of average take-up rate in cited context).
  - Endorsement may narrow priors for ambiguity-averse individuals or reduce perceived basis risk.
- Liquidity constraints:
  - Self-reported primary reason for non-purchase: “Insufficient funds to buy insurance” cited by 81 percent of non-purchasers in Andhra Pradesh, and 28 percent in Gujarat.
  - Cross-sectional correlation: demand larger among wealthy households; experimental cash shocks strongly increase take-up.
  - Alternative interpretation: reciprocity may play a role.
- Financial literacy and learning:
  - Probability skill exhibits the most robust positive relationship with purchase.
  - Short education module ineffective; multi-day programs in other studies (Gaurav, Cole and Tobacman (2011)) raised demand by five percentage points.
  - Prior payout experience and vendor familiarity positively correlated with purchase.

### Basis risk, limitations, and measurement notes
- Basis risk and spatial correlation:
  - Andhra Pradesh survey villages on average 4.2 miles (6.8 km) from reference weather station.
  - Payouts triggered by severe rainfall deficits likely spatially correlated with farm-level outcomes, but basis risk remains an acknowledged drawback.
- Data winsorization and sample notes:
  - Data from both states winsorized at 1% from top and bottom tails.
  - Andhra Pradesh stratified random sample from ~7,000 households.
  - Risk aversion lotteries played for real money (payouts between zero and Rs. 110); discounting questions hypothetical.
- Actuarial data windows:
  - Andhra Pradesh contracts: daily rainfall data 1970-2006.
  - Gujarat contracts: 1965-2003.
  - Data not available for three Andhra Pradesh stations using automated gauges, or for Anand in Gujarat.

### Policy implications and tentative conclusions
- Price reductions (lower transactions costs, competition, subsidies) would increase take-up but unlikely alone to generate widespread diffusion in short run.
- Address binding non-price frictions: trust, financial literacy, liquidity constraints, salience/attention.
- Contract and delivery design suggestions:
  - Increase payout frequency or quicker payouts to build track record and learning, noting trade-off with concentration of payouts in worst states.
  - Use automated rain gauges for immediate reporting; consider selling at harvest, combining with short-term loans, or state-contingent loan interest rates.
  - Group-based insurance (village/cooperative) could reduce marketing costs, leverage better management, and relax liquidity constraints.
  - Household visits effective but may not be scalable; selling alongside other financial services or marketing in group settings may be more cost-effective.
  - Technological possibilities: satellite foliage data for area-yield–based policies and mobile payment systems for premium collection.
- Overall conclusion: micro-insurance markets show promise for pooling important household income risks conditional on addressing price sensitivity, trust, liquidity, salience, and contract design.

*Source: _wp12195 - References (IMF working paper PDF content provided).*

### References .............................................................................................................

### _wp12195 - References

### Introduction and research question
- Purpose: study an innovative financial contract (rainfall index insurance) designed to insure rural Indian households against rainfall variation during the monsoon season.
- Context: drought cited by 89 percent of households in the sample as the most important risk they face.
- Product accessibility: policies sold in unit sizes as small as US$1.
- Core research question: What frictions limit adoption of financial products that pool important sources of household income risk?

### Main empirical findings on demand and frictions
- Price elasticity:
  - A ten percent price decline leads to a ten to twelve percent increase in take-up.
  - Point estimates imply demand would increase by 36 to 66 percent if priced at payout ratios similar to US retail insurance contracts.
- Persistent non-price frictions:
  - Even with very large price discounts, less than half of households purchased insurance in a subset of the sample despite an expected return of around 70 percent.
  - Households almost universally purchase only one policy unit, indicating non-price barriers limit scale of uptake per adopter.
- Trust and information:
  - Households do not fully trust or understand the insurance product; trust significantly affects demand.
  - Randomized endorsement: demand is 36 percent higher when an insurance educator is first recommended by a trusted local agent.
  - Households answer simple arithmetic questions correctly 60 percent of the time, indicating limited numeracy/financial literacy.
  - Demand higher in villages that previously experienced a payout and among households with prior insurance experience, higher financial literacy, and greater facility with probability concepts.
- Liquidity constraints and salience:
  - Providing enough cash to buy one policy increases take-up by 140 percent.
  - The cash reward effect is magnified among poor households.
  - Adopters generally buy a single policy; non-purchasers most frequently cite “lack of funds” as reason for not buying.
  - Receiving a product flyer or a visit from an insurance educator increases take-up significantly, consistent with limited attention or salience effects.
- Education and framing interventions:
  - A short insurance education module has no significant effect on demand.
  - Point estimates for the framing effects considered are generally close to zero; standard-error bounds imply smaller effects than those found in Bertrand et al. (2010), where comparable.

### Insurance contract design and delivery (product specifics)
- Contract type: index insurance (payouts linked to publicly observable rainfall index).
- Underwriters:
  - ICICI Lombard in Andhra Pradesh.
  - IFFCO-Tokio in Gujarat.
- Measurement: payoffs calculated based on measured rainfall at a nearby government rainfall station or an automated rain gauge operated by a private third-party vendor.
- ICICI policy structure (Andhra Pradesh):
  - Monsoon divided into three contiguous phases.
  - Phases I and II cover deficit rainfall: pay Rs. 10 for each mm below a pre-specified strike; if rain is below the exit level, a contract pays Rs. 1,000.
  - Phase III covers excess rainfall: pays Rs. 10 for each mm above the strike, with a maximum payout of 1,000 if rainfall meets or exceeds the exit level.
  - Over 60 percent of adopters in the sample purchased Phase I.
- IFFCO-Tokio policy structure (Gujarat):
  - Based on cumulative rainfall over the entire monsoon season (June 1 to August 31).
  - Payout is a non-linear function of the percentage shortfall from a specified "normal" rainfall level.
- Marketing and sales channels:
  - Gujarat: marketed by SEWA (an NGO serving women).
  - Andhra Pradesh: sold by BASIX via local Livelihood Services Agents (LSAs) that also provide loans and other financial services.
- Payout disbursement:
  - Gujarat: SEWA anticipated visiting households individually to make payouts.
  - Andhra Pradesh: payouts disbursed in a public ceremony.

### Actuarial values, observed payouts, and pricing
- Expected payout estimates (historical calculation): point estimates range from 33 to 57 percent of premiums, averaging 46 percent.
- Comparison to U.S. retail insurance:
  - U.S. retail insurance contract payout ratios average 65 to 75 percent (examples cited: private passenger auto liability 76.2 percent, private passenger auto physical damage 68.4 percent, homeowners insurance 64.7 percent; earthquake 20.4 percent; crop insurance aggregate claims-to-premiums ratio 244 percent).
- Observed payouts in study:
  - In Gujarat, sufficient rain in 2006 and 2007 meant no payout was triggered.
  - In Andhra Pradesh, at least one positive payout was observed for each rainfall station between 2004 and 2006, with payouts ranging from Rs. 40 to Rs. 1,796 per policy.

### Sample, setting, and data
- Andhra Pradesh sample:
  - Districts: Mahbubnagar and Anantapur.
  - Summary statistics based on a survey of 1,047 landowner households in 37 villages (survey conducted in 2006; original selection in 2004 from a census of approximately 7,000 landowner households).
  - Same set of households used for field experiments.
- Gujarat sample:
  - Districts: Ahmedabad, Anand, and Patan.
  - Villages selected where SEWA operated and within 30 km of a rainfall station.
  - Summary statistics based on a baseline survey of 1,500 SEWA members in 100 villages (conducted in May).

### Interpretation and implications
- Main summary conclusion: rainfall insurance demand is price-sensitive, but lower prices alone are unlikely to generate widespread index insurance adoption in the short run.
- Important binding non-price frictions: lack of trust and financial literacy, liquidity constraints, and salience/attention.
- Dynamic considerations: as markets mature, non-price barriers may decline (e.g., through increased product familiarity), so lower prices today could have dynamic effects by accelerating learning and diffusion.
- Contract design implications: improvements in insurance contract design could help mitigate non-price frictions; the authors suggest simple design improvements in the paper’s conclusion based on empirical results.

*Source: _wp12195 - References (IMF working paper PDF content provided).*

### 2006. The survey sample is representative of SEWA members in these 100 villages.

### _wp12195 - 2006. The survey sample is representative of SEWA members in these 100 villages.

### Survey sample and field experiments
- The Gujarat household survey: 1,500 households selected from SEWA's membership in 100 villages. For Gujarat, 15 households were selected per village: five randomly selected from the SEWA member list, five randomly selected from the remaining SEWA members with a positive savings account balance, and five selected (non-randomly) based on suggestions from a local SEWA employee that they would be likely to purchase rainfall insurance.
- The entire Gujarat sample of 1,500 households has similar summary statistics to the 500 selected randomly from the SEWA list, implying that the sample is close to representative of SEWA’s overall membership in these 100 villages.
- Field experiments in 2007 were conducted in a randomly selected 50 of these 100 villages but covered a larger set of households within these villages than those included in the 2006 baseline survey.
- Two of the 100 villages were found to be so close that they were grouped together and assigned the same treatment status.

### Basic demographic and wealth characteristics (selected summary statistics)
- Household size: Andhra Pradesh mean 6.26, St. Dev. 2.82; Gujarat mean 5.85, St. Dev. 2.39.
- Scheduled Caste or Scheduled Tribe (1=Yes): Andhra Pradesh 11.60%, St. Dev. 32.04%; Gujarat 43.70%, St. Dev. 49.60%.
- Muslim (1=Yes): Andhra Pradesh 3.90%, St. Dev. 19.37%; Gujarat 8.73%, St. Dev. 28.20%.
- Household head is male (1=Yes): Andhra Pradesh 93.75%, St. Dev. 23.96%; Gujarat 75.70%, St. Dev. 42.90%.
- Household head's age: Andhra Pradesh mean 47.60, St. Dev. 12.13; Gujarat mean 48.93, St. Dev. 12.87.
- Monthly per capita food expenditures: Andhra Pradesh mean 310.53, St. Dev. 126.89; Gujarat mean 555.37, St. Dev. 417.42.
- Total value of all savings deposits: Andhra Pradesh mean 1,030.42, St. Dev. 2,891.43; Gujarat mean 1,060.13, St. Dev. 2,314.97.
- Land holdings (in acres): Andhra Pradesh mean 6.31, St. Dev. 6.17; Gujarat mean 4.11, St. Dev. 5.49.
- Pct. of cultivated land that is irrigated: Andhra Pradesh 43.93%, St. Dev. 43.26%; Gujarat 43.70%, St. Dev. 47.10%.
- Household bought weather insurance in 2004 (1=Yes): Andhra Pradesh 25.31%, St. Dev. 43.50% (Gujarat n.a.).
- Household has some type of insurance (1=Yes): Andhra Pradesh 80.54%, St. Dev. 39.25%; Gujarat 63.78%, St. Dev. 48.08%.
- Belongs to water user group (BUA or WUG) (1=Yes): Andhra Pradesh 1.84%, St. Dev. 13.35% (Gujarat n.a.).
- Number of groups that the household belongs to: Andhra Pradesh mean 0.72, St. Dev. 0.62 (Gujarat n.a.).

Notes from sample context:
- Overall, the state of Gujarat has richer soil and is substantially wealthier than Andhra Pradesh.
- In Gujarat, approximately 52 percent of households report owning no farmland; these households earn their primary income from agricultural labor.
- The value of savings deposits is similar across the two study areas, at around Rs. 1,000 ($21 US).
- A wealth index based on durable goods owned is higher in Andhra Pradesh (index not reported in table).

### Risk attitudes, discounting, and exposure to risk
- Risk aversion index (mapped between 0 and 1 where high values indicate greater risk aversion): Andhra Pradesh mean 0.57, St. Dev. 0.25; Gujarat mean 0.54, St. Dev. 0.32.
- Subjective discount factor (value today of 1 Rs received in future): Andhra Pradesh mean 0.71, St. Dev. 0.29; Gujarat mean 0.75, St. Dev. 0.15.
  - Measured discounting is high: average monthly discount factor is 0.75 in Gujarat and 0.71 in Andhra Pradesh.
- Lotteries used to measure risk aversion had payouts between zero and Rs. 110; lottery choices were mapped into the index reported above.
- Exposure channel: supply and wages for agricultural labor depend importantly on the quality of the monsoon.

### Education, financial literacy, cognitive ability, and insurance comprehension
- Highest level of education (share with Primary school or below): Andhra Pradesh 66.8%; Gujarat 42.0%.
- Secondary school: Andhra Pradesh 7.5%; Gujarat 28.7%.
- High school: Andhra Pradesh 18.2%; Gujarat 11.6%.
- College or above: Andhra Pradesh 7.4%; Gujarat 17.6%.
- Average Score, Math Questions: Gujarat 61.7% (Andhra Pradesh n.a.); another entry shows Gujarat 71.8% in a separate row (table structure preserved as in source).
- Average Score, Financial Literacy: Gujarat 35.8% (Andhra Pradesh n.a.).
- Average Score, Insurance Questions: Andhra Pradesh 79.3%; Gujarat 68.2%.
- Average Score, Probability Questions: Andhra Pradesh n.a.; Gujarat 59.1% (panel data layout as in source).
- Specific insurance comprehension (Andhra Pradesh):
  - a) It rains 120 mm. Will you get an insurance payout? [Ans: No] Correct: 85.8%.
  - b.i) It does not rain at all: Will you get an insurance payout? [Ans: Yes] Correct: 83.0%.
  - b.ii) How much of a payout would you receive? [Ans: Rs. 500] Correct: 80.6%.
  - c.i) It rains 20mm: Will you get an insurance payout? [Ans: Yes] Correct: 81.5%.
  - c.ii) How much of a payout would you receive? [Ans: Rs. 200] Correct: 76.0%.
- Gujarat insurance comprehension entries reported: 63.7%, 58.9%, 79.9% in panel layout corresponding to Andhra Pradesh items (n.a. markers preserved as in source).
- In Gujarat, short tests of math, financial literacy, and understanding of probabilities were administered; the average math score is 62 percent, average financial literacy is 35.8 percent, and average probability question performance is 59.1 percent.
- Understanding of index-based insurance tested: correct answers 79 percent in Andhra Pradesh and 68 percent in Gujarat on hypothetical product comprehension.

### Experimental design — objectives and treatments
- Objective: estimate slope of demand curve for rainfall insurance and determine sensitivity of demand to non-price factors including trust, liquidity constraints, and framing effects.
- Table 4 summarizes share of households receiving different treatments (panels below preserve reported shares).

Andhra Pradesh (2006) — selected experimental facts:
- In May 2006, 700 households from sample of 1,047 were randomly selected to be visited in their home by trained ICRISAT insurance educators; visits successfully completed for 660 households.
- Households could purchase insurance on-the-spot or later through local BASIX branch or LSA. Educators sometimes revisited if household lacked cash.
- Three randomized dimensions of household visits:
  1. Endorsement by local BASIX LSA:
     - Two-thirds of villages designated endorsement-eligible; within these villages, the LSA endorsed the educator for half the visits.
     - The LSA introduced the educator, declared him/her trustworthy, and left before the educator described the product.
     - Instructions were followed exactly in 56 percent of cases; for the remainder, 25 percent did not show up or stayed too short a time, 19 percent stayed for the duration of the visit.
  2. Cash compensation for time: randomly Rs. 25 or Rs. 100 (half received Rs. 100).
     - Rs. 100 provides roughly enough cash-on-hand to purchase one policy (premiums range between Rs. 80 and Rs. 125).
  3. Additional education on conversion between soil moisture and millimeters of rainfall:
     - For 350 households, showed length of 10mm and 100mm using a ruler and chart mapping 100mm to average soil moisture for their soil type; for other 350 households, this information was not provided.
- Rationale: endorsement tests effect of trust; cash tests liquidity constraints (reference to Rampini and Viswanathan, 2010); education tests comprehension between soil moisture and rainfall millimeters (only 23 percent of households can accurately indicate the length of a fixed number of millimeters).

Gujarat (2007) — selected experimental facts and marketing treatments
- Video marketing treatments were used in villages where rainfall insurance was offered for the first time in 2007. Flyer treatments were used in villages where rainfall insurance was offered in both 2006 and 2007 in Gujarat.
- Panel B summary (selected treatment shares and labels as reported):
  - Strong SEWA Brand: 62% (Total), 100% (Surveyed), 51% (Non-Surveyed).
  - Peer Endorsed: 59% (Total), 100% (Surveyed), 47% (Non-Surveyed).
  - Positive Frame (Pays 2/10 Years): 52% (Total), 50% (Surveyed), 52% (Non-Surveyed).
  - Vulnerability Frame: 11% (Total), 51% (Surveyed), 0% (Non-Surveyed).
  - Discount = Rs. 54: 2% (Total), 48% (Surveyed), 41% (Non-Surveyed).
  - Discount = Rs. 153: 8% (Total), 34% (Surveyed), 40% (Non-Surveyed).
  - Discount = Rs. 301: 9% (Total), 18% (Surveyed), 20% (Non-Surveyed).
- Flyer Treatments (N = 2391) reported shares:
  - Individual Emphasis (not Group): N = 1232, 52% of total.
  - Muslim Emphasis: N = 836, 35% of total.
  - Hindu Emphasis: N = 809, 34% of total.
  - Neutral (Non-religious) Emphasis: N = 746, 31% of total.
- Video treatment descriptions:
  - "Strong SEWA Brand" videos include clear indications that the product is being offered by SEWA.
  - "Peer endorsed" uses farmer-delivered endorsement.
  - "Positive frame" emphasized product would have paid out in 2 of the last 10 years.
  - "Vulnerability frame" warned households of difficulties without insurance.
- Study design notes: the two-tiered endorsement assignment measured possible spillovers of trust within village and helped reduce demands on BASIX staff time.

### Measurement notes and methodological details
- Data from Andhra Pradesh come from surveys conducted in 2006 and BASIX administrative records. Data from Gujarat come from the baseline survey conducted in 2006.
- Data from both Andhra Pradesh and Gujarat have been winsorized at 1% from the top and bottom tails.
- In Andhra Pradesh, a stratified random sample was selected from a census of approximately 7,000 households.
- Andhra Pradesh visit treatments: 700 households visited (67% of sample); Village endorsed 474 (45%); Visit endorsed 238 (23%); Education module given to 350 households (33%); High reward to 302 households (29%).
- Risk aversion lotteries were played for real money; payouts between zero and Rs. 110.
- Discounting question was hypothetical and may reflect present-biasedness, suspicion about default, or measurement error due to hypothetical nature.
- Financial literacy questions adapted from Lusardi and Mitchell (2006). Probability tests involved gauging likelihoods from depictions of bags with black and white balls.

*Source: _wp12195 - 2006. The survey sample is representative of SEWA members in these 100 villages.*

### 2007. SEWA used several techniques to market rainfall insurance, including flyers, videos, and

### _wp12195 - 2007. SEWA used several techniques to market rainfall insurance, including flyers, videos, and

### Experimental design and implementation
- Field experiments conducted in 50 villages in Gujarat where rainfall insurance was offered in 2007; villages were randomly selected for marketing phase-in: 30 offered insurance in 2006 and an additional 20 (randomly selected) offered insurance in 2007.
- Marketing media randomized at the household level:
  - In villages with no prior exposure to insurance (20 villages): portable video players delivered a 90-second marketing message directly to household-decision makers. Each treated household was randomly assigned one of eight different videos.
  - In villages where insurance had been offered in 2006 (30 villages): flyers were distributed, containing one of six randomly assigned messages.
- All treated households received a non-transferable coupon bearing name and address; coupon serial number indicated which marketing message the household received.
- Discount coupon randomization in the 20 video-treated villages:
  - 40 percent of households received Rs. 5
  - 40 percent of households received Rs. 15
  - 20 percent of households received Rs. 30
- In the 30 flyer-treated villages, the discount was fixed at Rs. 5.
- Treatments delivered to a cross-section of households in each village, including all households who participated in the 2006 survey.

### Gujarat: Video and flyer treatments (content dimensions)
- Video treatments randomized along four dimensions:
  - SEWA Brand (Yes or No): “Strong SEWA brand” treatment explicitly indicated product was marketed by SEWA; control did not mention SEWA.
  - Peer vs. Authority Figure: “Peer” treatment endorsement by a local farmer; “Authority” treatment endorsement by a teacher.
  - Payout framing (“2/10 yes” or “8/10 no”): “2/10” statement: “the product would have paid out in approximately 2 of the previous 10 years”; “8/10” statement: “the product would not have paid out in approximately 8 of the previous 10 years”.
  - Safety or Vulnerability: “Safety” described benefits as protecting household and ensuring prosperity; “Vulnerability” warned of difficulties if uninsured.
- Flyer treatments randomized along two dimensions testing group identity interaction:
  - Religion (Hindu, Muslim, or Neutral): photograph and farmer name matching religious cue.
  - Individual or Group emphasis: Individual flyer emphasized benefits to the individual purchaser; Group flyer emphasized value for the purchaser’s family.

### Andhra Pradesh: treatments and estimation approach
- Treatments implemented:
  - Whether the household was visited by an insurance educator.
  - Whether the educator was endorsed by an LSA.
  - Whether the educator presented the education module.
  - Whether the visited household received a high cash reward (Rs. 100 rather than Rs. 25).
- Endorsement occurred in two-thirds of villages; interaction between village-level endorsement and household visit included to identify local spillovers.
- Empirical strategy: linear probability model of household insurance purchase as a function of treatment variables; intent-to-treat interpretation due to imperfect compliance.
- Sample size and baseline:
  - Data from all 1,047 households.
  - Unconditional insurance take-up rate is 28 percent.
  - Columns report models with and without village fixed effects and with household covariates and interactions.

### Andhra Pradesh: key empirical findings (Table 5 summary)
- Effect magnitudes (reported as increases in take-up, all exact figures preserved):
  - Visit (1=Yes): 0.172*** (column 1); 0.128*** (column 2); 0.115*** (column 3); 0.117*** (column 4); 0.114*** (column 5); 0.118*** (column 6). Robust standard errors shown in table.
  - Endorsed by LSA (1=Yes): coefficients range from 0.064 to 0.194 across columns (noting some estimates have large standard errors).
  - Education module (1=Yes): coefficients around 0.003 to 0.007, and -0.003 in one column; economically small and statistically insignificant.
  - High reward (1=Yes): 0.408** in one specification; 0.400***, 0.394***, 0.387***, 0.393***, and 1.629*** in other columns (table reports these estimates with robust standard errors).
- Interpretation and statistical significance:
  - Being assigned a household visit alone increases take-up by 11.5 to 17.2 percentage points.
  - A high reward increases take-up by 39.4 to 40.8 percentage points; estimates statistically significant at the 1 percent level.
  - Individual LSA endorsement is positively signed and marginally statistically significant (t-stat between 1.5 and 1.7). LSA-endorsement and village endorsement variables are jointly significant at the 2 percent level (column 2) and the 1 percent level (column 3), implying spillovers to non-endorsed households in endorsed villages.
  - Education module effect is economically small and statistically insignificant.
- Interaction effects:
  - LSA endorsement effect varies sharply by household familiarity with BASIX:
    - For households familiar with BASIX, LSA endorsement increases take-up by 10.1 percentage points (statistically significant at the 5 percent level).
    - For households unfamiliar with BASIX, net effect is 10.1 - 17.1 = -7.0 and statistically insignificant.
  - High cash reward effect is larger among poor households; significance reported at 10 percent in one column and at 1 percent in another.
- Model statistics and sample:
  - Mean of dependent variable: 0.282.
  - R-squared values across specifications: 0.279, 0.355, 0.380, 0.384, 0.382, 0.387.
  - Observations: 1047.
  - *, **, and *** indicate significance at the 10, 5 and 1 percent level, respectively.

### Gujarat: Video experiments — empirical findings (Tables 6 summary)
- Sample and baseline:
  - Overall take-up rate is 29.4 percent.
  - Number of observations in Panel A: 1413 (columns 1–4).
- Price/discount effects:
  - Discount (fraction of initial price) coefficients:
    - Column (1): 0.307*** (0.076)
    - Column (2): 0.340*** (0.075)
    - Column (3): 0.372** (0.148)
    - Column (4): 0.405** (0.151)
  - Implied price elasticity of demand reported as 1.04 and 1.16 (table entries).
  - Interpretation in text: coefficient of 0.307 in column (1) implies that a 10 percent decline in the price of insurance increases the probability of purchase by 3.07 percentage points, or 10.4 percent of the baseline take-up rate; implied elasticity is 1.04 (or 1.16 based on column (2)).
- Framing and brand effects:
  - Strong SEWA Brand coefficients: -0.026, -0.031, -0.081*, -0.082* across columns (not statistically robustly positive).
  - Vulnerability Frame coefficients: 0.046, 0.041, 0.131, 0.134 (not statistically significant).
  - Positive Frame (Pays 2/10 Years) coefficients: -0.027, -0.035, -0.037, -0.049 (not statistically significant).
  - Peer Endorsed coefficients: -0.031, -0.021, 0.022, 0.036 (not statistically significant).
  - Joint tests: framing effects are jointly insignificant.
- Interactions with discount:
  - Percentage Discount x Strong SEWA Brand: 0.258** and 0.236* in columns reported.
  - Other discount interactions not jointly significant (F-test p-values: discount interactions 0.265 and 0.144).
- Survey participation effect:
  - Surveyed Household coefficients: 0.159**, 0.179**, 0.207***, 0.210*** across columns; households in the 2006 baseline survey are significantly more likely to purchase insurance.
- Panel B: district-level take-up and gross return on premium (selected exact entries)
  - For Rs. 5 discount: Return (gross) 61%; Take-up 25% and another entry 22% depending on policy/district.
  - For Rs. 15 discount: Return (gross) 82%; Take-up 37% and another entry 22% depending on policy/district.
  - For Rs. 30 discount: Return (gross) 169%; Take-up 47% and other entries showing take-up 30% and 44% across districts/policies.
  - Text highlights: for farmers in Ahmedabad receiving the Rs. 30 discount, estimated expected payouts are 169 percent of net premiums; fewer than half of eligible farmers receiving this discount chose to buy insurance.
- Interpretation:
  - Size of discount has a large, statistically significant effect on take-up (1 percent significance).
  - Framing effects do not significantly affect overall take-up; interactions between discount and framing mostly statistically indistinguishable from zero.
  - Prior survey exposure (2006 baseline participants) strongly associated with higher take-up.

### Gujarat: Flyer experiments — empirical findings (Table 7 summary)
- Overall take-up:
  - Overall take-up rate among flyer-receiving households is 23.8 percent.
- Baseline flyer treatment effects:
  - None of the baseline treatments (religion cue or group vs. individual emphasis) are statistically significant in simple specifications; coefficients small.
- Interaction effects:
  - Group emphasis combined with neutral religious setting has a significant positive effect on take-up.
  - Use of Muslim religious setting on the flyer (instead of neutral) reduces take-up by 9-10 percentage points, statistically significant at the 5 percent level in specifications reported.
- Heterogeneous effects by respondent religious identity (columns 5–8):
  - For households identified as Muslim receiving a group-emphasis flyer:
    - Inclusion of Hindu symbols on the flyer reduces take-up by 32.8 or 34.2 percentage points compared to the neutral flyer (coefficients reported in table columns).
  - For households identified as Hindu receiving a group-emphasis flyer:
    - Inclusion of Muslim symbols on the flyer reduces take-up by 10.1 or 9.6 percentage points compared to the neutral flyer.
  - These results use religious identity coded from respondent names by Gujarati research assistants; 219 respondents with coder disagreement were omitted from columns (5)-(8).
- Interpretation:
  - Emphasizing communal/group nature of insurance can stimulate demand, but only when cues do not emphasize a different group identity than the household’s; mismatched religious cues substantially reduce take-up for both Hindu and Muslim households, with larger point estimates among the smaller Muslim subsample.

### Synthesis of experimental implications (as presented in the text)
- Marketing and price matter:
  - Direct household visits increase take-up substantially (11.5 to 17.2 percentage points).
  - Large cash rewards markedly increase take-up (39.4 to 40.8 percentage points).
  - Price discounts have a sizable and statistically significant effect on take-up; implied elasticities around 1.04–1.16.
- Information and framing:
  - Short education module had economically small and statistically insignificant effects in Andhra Pradesh.
  - Framing manipulations in Gujarat videos (positive vs. negative payout framing; safety vs. vulnerability; peer vs. authority; SEWA brand prominence) generally did not produce robust increases in take-up; framing effects were jointly insignificant in main specifications.
- Trust, endorsement, and familiarity:
  - LSA endorsement increases take-up for households familiar with BASIX; endorsement effects vary by prior familiarity and produce village-level spillovers.
  - Brand prominence (Strong SEWA Brand) does not uniformly increase take-up in video experiments; price sensitivity may interact with brand emphasis.
- Social identity and group cues:
  - Flyers emphasizing group benefits can increase take-up in neutral settings.
  - Religious cues that signal a different group identity from the household reduce take-up substantially, implying that mismatched identity cues undermine insurance demand.
- Behavioral and exposure effects:
  - Households previously surveyed in 2006 are more likely to purchase insurance, consistent with survey exposure affecting behavior.

*Italic: Content summarized from _wp12195 - 2007. SEWA used several techniques to market rainfall insurance, including flyers, videos, and (source PDF content).*

### 2007. A linear probability model is used, with the dependent variable set to one if the household purchased an insurance

### _wp12195 - 2007. A linear probability model is used, with the dependent variable set to one if the household purchased an insurance policy.

### V. DISCUSSION OF EXPERIMENTAL RESULTS — Overview
- The authors synthesize three sets of field experiments to evaluate barriers to insurance participation.
- Experiments indicate both price and non-price factors matter for take-up of rainfall insurance in the study areas.

### A. Price Relative to Actuarial Value — Key findings
- Rural finance is expensive to provide; annual operating costs for non-bank microfinance loans range from 17–26 percent of loan value (Cull, Demirguc-Kunt and Morduch (2009), cited).
- Rainfall insurance demand is significantly price-sensitive.
  - The relevant coefficient in columns (1) and (2) of Table 6 indicates that a price reduction of 10 percent increases demand by 10.4 to 11.6 percent.
  - These point estimates imply that, holding everything else constant, rainfall insurance demand would increase significantly (by approximately 36-66 percent) if insurance could be
- Even with large price reductions, only a small fraction of households would purchase insurance given low baseline take-up rates.
- Ahmedabad-specific result: more than half of households do not purchase rainfall insurance even when the policy price is set significantly below the actuarial value of the insurance policy.
- Conclusion drawn: non-price factors present significant barriers to take-up, at least in the current early stage of the insurance product’s life cycle.

### B. Trust — Key findings
- Purchasing insurance requires paying a known premium today for an uncertain future payout; evaluating benefits may be difficult for households with low financial literacy or little experience with the product.
- Advice from trusted sources or the seller’s reputation is likely to strongly influence household decisions.
- Andhra Pradesh results: a higher level of trust in the otherwise unknown insurance educator, due to an endorsement from the local BASIX LSA, significantly increases insurance take-up for households already familiar with BASIX.
  - The effect holds only amongst households already familiar with BASIX (for whom the word of the LSA is credible).
  - For this subgroup, LSA endorsement increases the insurance purchase probability by [text truncated in source].

### C. Experimental results — Selected regression estimates (Table excerpts)
- Treatments (example coefficients and reported standard errors in parentheses):
  - Muslim emphasis (1=Yes): -0.002 (0.023); -0.004 (0.023); 0.043 (0.034); 0.045 (0.034); 0.134 (0.102); 0.160 (0.113); 0.041 (0.040); 0.041 (0.039)
  - Hindu emphasis (1=Yes): 0.002 (0.019); 0.008 (0.019); 0.012 (0.030); 0.022 (0.030); 0.057 (0.086); 0.121 (0.131); 0.002 (0.034); 0.014 (0.034)
  - Group emphasis (1=Yes): 0.020 (0.018); 0.015 (0.018); 0.060* (0.032); 0.060** (0.028); 0.247** (0.110); 0.239* (0.135); 0.058 (0.037); 0.053 (0.033)
  - Surveyed Household: 0.133*** (0.040); 0.132*** (0.040); 0.134*** (0.040); 0.133*** (0.040); 0.121 (0.136); 0.106 (0.155); 0.107*** (0.039); 0.088** (0.038)
- Religion treatment interactions:
  - Muslim emphasis x group: -0.094** (0.044); -0.101** (0.042); -0.223 (0.219); -0.230 (0.192); -0.101** (0.049); -0.096* (0.048)
  - Hindu emphasis x group: -0.019 (0.047); -0.029 (0.045); -0.328** (0.132); -0.342* (0.171); -0.000 (0.053); -0.015 (0.051)
- Table summary statistics:
  - Village fixed effects: No / Yes (varies by column)
  - Mean of dependent variable: 0.238 0.238 0.238 0.238 0.167 0.167 0.268 0.268
  - R-squared: 0.016 0.120 0.018 0.123 0.085 0.349 0.013 0.134
  - Observations: 2391 2391 2391 2391 1323 1323 2040 2040
  - Significance notation: *, **, and *** indicate significance at the 10, 5 and 1 percent level, respectively.
- Notes on flyer treatments (Gujarat, 2007):
  - A linear probability model is used, with the dependent variable set to one if the household purchased an insurance policy.
  - Robust standard errors reported in parentheses.
  - "Group Emphasis" indicates that the flyer emphasized the benefit of insurance for the family (not the individual).
  - "Muslim, Hindu, and Neutral Emphasis" depict a farmer near a Mosque, Hindu temple, or nondescript building, respectively.
  - Columns (2), (4), (6) and (8) include village fixed effects.
  - Columns (1)-(4) present results for the entire sample; columns (5)-(6) for identifiably Muslim names; columns (7)-(8) for identifiably Hindu names.
  - 219 respondents on which two independent coders disagreed have been omitted from the analysis in columns (5)-(8).

*Source: Excerpt from the referenced IMF working paper (2007) contained in the supplied PDF content.*

### 10.1 percentage points, or 36 percent of the average take-up rate. In contrast, LSA endorsement

### _wp12195 - 10.1 percentage points, or 36 percent of the average take-up rate. In contrast, LSA endorsement

### Trust and Brand Endorsement
- LSA endorsement has no demand effect amongst farmers unfamiliar with BASIX.
- SEWA brand endorsement in the Gujarat video treatments shows no statistically significant effect; possible reasons: differences in reputation or the treatment was too subtle compared to “live” in-person endorsement in Andhra Pradesh.
- The study provides causal experimental evidence that trust matters for demand for microinsurance and financial products in environments with weak formal legal protections and low household financial literacy.
- Interpretations of the “trust” effect include:
  - Endorsement may narrow priors for ambiguity-averse individuals (Bryan (2010)-type mechanism).
  - Endorsement may reduce perceived basis risk.
- Measured household risk aversion is negatively correlated with insurance demand in both Andhra Pradesh and Gujarat.

### Price Sensitivity and Elasticities
- Insurance demand is significantly price sensitive, with an elasticity of around unity.
- Numerical comparisons cited:
  - U.S. contracts provide an average payout-to-premia ratio of 70percent, compared to 46percent for the Indian rainfall insurance contracts.
  - This implies the price per unit of payout is 34percent lower for the US contracts.
  - Point estimates of 0.3 to 0.34 suggest cutting the price of the Indian contracts by 34percent would increase demand by 35percent to 40percent.
  - The upper bound of 65percent is calculated by comparing the price of the lowest-value Indian insurance contract to the highest-value U.S. contract.
- A two standard deviation confidence interval for individual framing treatments is generally no larger than ± 6 percentage points; in nearly every case the null that frame shifts demand by more than 10 percentage points can be rejected.

### Liquidity Constraints
- Experimental results from Andhra Pradesh: a positive liquidity shock at the time of the household visit has a large positive effect on household insurance demand, magnified among less wealthy households.
- Non-experimental suggestive evidence:
  - Cross-sectionally, insurance demand is larger among wealthy households.
  - Non-participating households self-report “lack of funds to buy” as the most common reason for not purchasing insurance.
- Self-reported primary reason for non-purchase:
  - “Insufficient funds to buy insurance” cited by 81 percent of non-purchasers in Andhra Pradesh, and 28 percent in Gujarat.
  - “Do not need insurance” cited by 3 percent in Andhra Pradesh and 25 percent in Gujarat.
- Alternative interpretation: reciprocity—cash given by ICRISAT educator may create obligation to use funds to purchase insurance.
- Recommendation: further research to confirm the role of liquidity constraints; if confirmed, improvements in credit access could increase insurance demand.

### Financial Literacy, Education, and Learning
- Table 3 statistics show a significant fraction of households are unable to answer simple mathematics or financial questions; a smaller fraction do not understand basic features of the rainfall insurance contracts.
- Short rainfall insurance education module in Andhra Pradesh has no significant effect on insurance demand.
- Gaurav, Cole and Tobacman (2011) confirm the inefficacy of a similar short module, but find a two-day educational program with simulation games increased demand by five percentage points.
- Prior experience and familiarity with vendor are positively correlated with purchase; a history of positive past payouts increases demand, suggesting diffusion may be slow until a track record develops.

### Framing, Salience, and Behavioral Factors
- Limited evidence that pure framing effects significantly affect take-up in Gujarat video experiments (no significant differences among eight frames).
- Flyer experiments show framing interacts with group and religion:
  - Among households receiving a group-emphasis flyer, Muslim households have significantly lower take-up when the flyer includes Hindu symbols; symmetrically, Hindu households have significantly lower take-up when the flyer includes Muslim symbols.
  - Emphasizing communal nature of insurance stimulates demand only when cues match the household’s group identity.
- Being assigned a household visit in Andhra Pradesh significantly increases insurance take-up even when product is available to all—possible mechanisms: added convenience, baseline information from the educator, or increased salience consistent with limited attention models.

### Non-Experimental Correlates of Purchase (Regression Results Summary)
- Wealth and liquidity proxies:
  - Wealth index and log of monthly per capita food expenditures positively correlated with insurance purchase (stronger for Gujarat).
- Financial and probability skills:
  - “Probability skill” exhibits the most robust positive relationship with purchase.
  - Financial literacy, math skills, and insurance skills generally positively correlated with purchase; “probability skill” particularly important for index insurance.
- Prior experience and familiarity:
  - Average insurance payouts in the village (2004 and 2005), having bought weather insurance in 2004, and having other insurance policies are positively correlated with purchase.
- Demographics and other correlates:
  - Higher measured risk aversion is negatively correlated with purchase in both samples.
  - Fraction of irrigated land (proxy for income variance) statistically insignificant in multivariate specifications.
- Table 8 notes:
  - Linear probability model with robust standard errors.
  - Columns include univariate and multivariate regressions; Columns (5) and (6) include village fixed effects.
  - Observations reported: 10477 72 10477 77 10477 72 (as shown in the table).

### Basis Risk and Limitations
- The study does not directly analyze how basis risk affects demand; randomized treatments were not designed to measure basis risk.
- Survey villages in Andhra Pradesh are on average 4.2 miles, or 6.8 km, from the reference weather station; payouts triggered by severe rainfall deficits are likely spatially correlated, making payouts strongly indicative of low rainfall and yields at the farm level.
- Acknowledged that policies do not cover other risks (e.g., pestilence, health), so insurance payouts will be imperfectly correlated with household income and consumption.

### Policy Implications and Tentative Conclusions
- Price reductions via lower transactions costs, greater competition, or subsidies would significantly increase take-up but are unlikely alone to generate widespread diffusion in the short run.
- Trust and a track record of payouts are important; design changes that increase payout frequency could facilitate learning but may reduce concentration of payouts in worst states.
- Group-based insurance (village, producer group, cooperative) could reduce marketing costs, leverage more educated group management, and relax liquidity constraints.
- Policies should aim to provide payouts quickly, especially during the monsoon when discounting of future cashflow is high:
  - Automated rain gauges (e.g., ICICI Lombard’s use) can report rainfall immediately, allowing faster payouts and reducing basis risk.
  - Other options: sell policies at harvest, combine product with short-term loans, or originate loans with state-contingent interest rates based on rainfall outcomes.
- Delivery considerations:
  - Household visits are effective in raising take-up but may not be scalable or cost-effective relative to commission structures.
  - Alternative delivery: selling insurance alongside other financial services or marketing in group settings may be more cost-effective.
- Technological possibilities include satellite foliage data for area-yield–based policies and mobile payment systems for premium collection.
- Overall: micro-insurance markets show promise as a channel for pooling important household income risks, conditional on addressing price sensitivity, trust, liquidity, salience, and contract design improvements.

*Source: Excerpt from the provided IMF working paper content.*

### REFERENCES

### _wp12195 - REFERENCES

### Key literature cited
- Extensive bibliography covering financial innovation, insurance theory, index insurance, household finance, behavioral economics, microinsurance field experiments, and weather/rainfall insurance. Representative authors and works include:
  - Allen, Franklin, and Douglas Gale. 1994. Financial Innovation and Risk Sharing.
  - Athanasoulis, Stefano, and Robert Shiller. 2000; 2001.
  - Binswanger, Hans P. 1980. “Attitudes toward Risk: Experimental Measurement in Rural India.”
  - Giné, Xavier, Robert Townsend and James Vickery. 2008; Giné et al. 2012.
  - Kahneman and Tversky (1979) and related behavioral probability misjudgment literature.
  - World Bank (2005, 2011) and Skees (2008) on index insurance; ICICI Lombard product history (Hess, 2003; Bryla and Syroka, 2007).

### Appendix: Definition of variables (selected highlights)
- Demographic characteristics
  - Household Size — Number of individuals (of any age) in the household.
  - Scheduled Caste / Scheduled Tribe (Both) — Dummy variable equal to 1 if household belongs to a scheduled caste or tribe.
  - Muslim (Both) — Dummy variable equal to 1 if household's religion is Muslim.
  - Household head is male (Both) — Dummy variable equal to 1 if the household head is male.
  - Household head's age (Both) — Age of household head in years.
- Utility and risk preferences
  - Risk aversion (Both) — Constructed from the choice over several lotteries as in Binswanger (1980). Assigns value 1 to individuals that choose the safe lottery; for those who choose riskier lotteries, indicates the maximum rate at which they are revealed to accept additional risk (standard deviation) in return for higher expected return (ΔE / Δrisk).
  - Subjective discount rate (Both) — Defined as (X- Xnow)/Xnow where X is the amount that leaves respondent indifferent between Xnow now and X in one month. In AP Xnow is Rs 200 and X can take values: Rs 201, Rs 205, Rs 210, Rs 220, Rs 240, Rs 260, Rs 300, Rs 400 or Rs 1000. In Gujarat, Xnow is Rs 8 and X can take values: Rs 7, 8, 9, 10, 11, 12.
- Expectations and exposure
  - Above average expected monsoon rain (1=Yes) (Both) — Dummy variable equal to 1 if household's expectation for monsoon rainfall (relative to past years) is higher than the average value of the sample as a whole. Elicited before the monsoon.
  - % cultivated land that is irrigated (Both) — Acres of cultivated land that is irrigated over total owned land. 1% winsorization of each tail.
- Wealth, consumption, and financial literacy
  - Wealth Index (Both) — First component of PCA score for dummy variables for items: tractor, thresher, bullock cart, furniture, bicycle, motorcycle, sewmach, electricity, telephone.
  - Monthly Per Capita Food Expenditures (Both) — Total monthly consumption expenditures on food divided by household size. 1% winsorization of left and right tail.
  - Total value of all savings deposits (Both) — Value of all deposits with any bank, post office or financial institution. 1% winsorization of left and right tail.
  - Insurance Questions (Both) — Number of correct answers to hypothetical questions in Table 3, Panel C.
  - Math Questions (Gujarat) — Number of correct answers to 8 arithmetic questions (list provided).
  - Probability Questions (Gujarat) — Number of correct answers to simple probability problems.
  - Financial Literacy (Gujarat) — Number of correct answers to hypothetical questions in Table 3, Panel B.
- Institutional and contextual variables (AP-specific examples)
  - Frequency of insurance payouts in the village (2004 and 2005) (AP) — Equal to 1 if there were insurance payouts in the village in both 2004 and 2005; equal to 0.5 if there was a payout in either 2004 or 2005, but not both; equals 0 if no payouts in either year.
  - HH bought rainfall insurance in 2004 (1=Yes) (AP) — Dummy = 1 if household bought weather insurance in 2004.
  - Does not know BASIX (1=Yes) (AP) — Dummy = 1 if respondent does not know BASIX, the insurance provider.
  - Household has other insurance (1=Yes) (Both) — Dummy = 1 if household has other insurances of any type besides rainfall insurance sold by BASIX (AP) or SEWA (Gujurat).
  - Understanding of millimeters (1=Yes) (AP) — Dummy = 1 if respondent correctly measured the distance between two points: asked to report the letter located 60mm from the starting point on a printed ruler.

### Institutional details regarding rainfall index insurance (concise)
- Nature of product
  - Rainfall insurance is a weather-based index product: payout determined by a publicly observable and contractible index such as rainfall or temperature.
  - Advantages: payouts can be calculated and disbursed quickly and automatically without formal claim filing; index products are free of adverse selection and moral hazard because payouts are based only on publicly observed data.
  - Main drawback: basis risk — the index realization may not be perfectly correlated with an individual policyholder's loss.
- Historical rollout and providers
  - First Indian rainfall insurance policies developed by ICICI Lombard with technical support from the World Bank (Hess, 2003; Bryla and Syroka, 2007).
  - Over time, rainfall insurance became more available across many parts of India; policies also underwritten by Agricultural Insurance Company of India (AIC) and others.
- Contract design specifics
  - ICICI Lombard (Andhra Pradesh) divides the monsoon “Kharif” into three contiguous phases of 35-45 days corresponding to sowing, flowering, and harvest. Note: monsoon onset varies; the start of the first phase is defined as the day in June when accumulated rainfall since June 1 exceeds 50mm. If <50mm of rain falls in June, the first phase begins automatically on July 1.
  - Policies sold for each phase at a premium between Rs. 80 and Rs. 120 ($2-3 US).
  - A policy covering all three phases (Combined Premium) costs Rs. 260 to Rs. 340 ($6-8 US), including a Rs. 10 discount.
  - IFFCO-Tokio (Gujarat) policies based on cumulative rainfall over the entire monsoon season (defined as June 1 to August 31) at government rainfall stations; premiums between Rs. 44 and Rs. 86 ($1-2 US).
  - Households free to purchase any whole number of policies.
- Payoff structure examples
  - Each contract specifies a threshold amount of rainfall approximating minimum required for successful crop growth.
  - Example: Phase I ICICI Lombard policy in Mahbubnagar pays zero when cumulative rainfall during 35-day coverage exceeds the strike of 70mm. Payouts are linear in the rainfall deficit; payout jumps to Rs. 1000 when cumulative rainfall is below the exit of 10mm. Example calculation: an Anantapur Phase I policy would pay Rs. 30 for a realized rainfall of 27 mm.
  - IFFCO-Tokio policies pay out whenever rainfall during the entire monsoon season is at least 40 percent below a specified average level for that district (normal rain).
  - Exception: Phase III ICICI Lombard contracts (harvest) pay off when rainfall is excessively high rather than low (to insure against flood or excess rain).
- Marketing and distribution
  - Underwriters typically do not sell directly; use brokers or partner with local financial institutions and NGOs (BASIX in AP, SEWA in Gujarat) with established rural networks.
  - Payouts are calculated automatically; distributor announces village visit and pays all claimants in a single day. Policy-members unable to collect that day may collect from the NGO subsequently.
- Actuarial values, observed payouts and pricing notes
  - Historical daily rainfall data availability:
    - Andhra Pradesh contracts: 1970-2006.
    - Gujarat contracts: 1965-2003.
    - Data not available for three Andhra Pradesh stations where payouts based on automated gauges, or for Anand in Gujarat.
  - Observed behavior of returns:
    - Giné et al. (2007): policy-holders obtain a positive return in only 11 percent of phases; maximum return observed in about 1 percent of phases is 900 percent.
    - Using administrative data for all policies sold by BASIX in Andhra Pradesh from 2003 to 2009, Giné et al. (2012) find an average ratio of total insurance payouts to total premiums of 138 percent.
  - Notes on estimation and perception:
    - Simple historical expected payout measures may misstate true policy value due to difficulty measuring frequency of low-probability events and highly skewed return distributions.
    - Individuals may have difficulty assessing policy value because probabilities are often misjudged and because information about expected payouts and standard errors was not provided during marketing.
  - Contextual reference: average daily wage for agricultural laborers in survey areas at time of study is around Rs. 50.

### Study design & sample size highlights (selected figures)
- Andhra Pradesh experimental design (2006)
  - Total sample: 1,047 (as indicated in Appendix Table 1 header).
  - Design: villages randomly assigned to two groups — no endorsement visits and endorsement villages (half of visits endorsed). Within groups, households randomly assigned to combinations of marketing treatments (education module × high reward) and endorsement treatments.
- Other study group sample counts (selected)
  - Group 1 Flyer Treatments — Total sample 2,391 across six subgroups (1A–1F).
  - Group 2 Video — Surveyed house holds in new treatment villages — Total sample 315 (groups 2A–2D).
  - Group 3 Video — Non-surveyed respondents in new treatment villages — Total sample 1,100 (groups 3A–3H).
  - Discounts (All Video Households) — Total sample 1,415 with discount levels listed as:
    - D1 Rs. 5 5 66
    - D2 Rs. 10 5 66
    - D3 Rs. 20 2 83
  - (Note: table entries replicate formatting as presented in source.)

*Source: _wp12195 - REFERENCES (PDF chapter/section).*

### Appendix Table 2. Study Design, G ujarat

### Appendix Table 2. Study Design, Gujarat

### Study design overview
- The table describes the experimental design for Gujarat in 2007.
- Households in the 21 villages which were offered insurance for the first time in 2007 received video treatments.
- Households receiving video treatments that were in the original survey sample were shown one of four videos; other households were shown one of eight different videos.
- All households observing videos were offered a discount of either Rs. 5, 10, or 20 on their first policy.
- Households in the 30 villages where insurance was offered in both 2006 and 2007 were given one of six flyers.

### Binswanger Lotteries — description
- The Binswanger Lotteries were used to measure risk aversion amongst sample groups in Andhra Pradesh and Gujarat.
- Each respondent chose one of the listed lotteries, which increased in risk and expected value.
- The measure of risk aversion assigns a value of 1 to those who choose the safe lottery and, for those who choose riskier lotteries, indicates the maximum rate at which they are revealed to accept additional risk (standard deviation) in return for higher expected return.

### Binswanger Lotteries — Gujarat (as presented)
- Heads Tails ΔE / Δrisk — Percent choosing this lottery 2006 (first block)
  - 25 25 1.00 10.3%
  - 20 60 0.75 25.6%
  - 15 80 0.60 18.0%
  - 10 95 0.50 25.3%
  - 5 105 0.33 11.0%
  - 0 110 0.00 9.9%
  - Average ΔE / Δrisk 0.57
- Heads Tails ΔE / Δrisk — Main Sample (N=1500) (second block)
  - 25 25 1.00 14.0%
  - 22 47 0.76 12.3%
  - 20 60 0.73 15.4%
  - 17 63 0.72 15.6%
  - 15 75 0.71 9.3%
  - 10 80 0.58 15.6%
  - 5 95 0.45 7.9%
  - 0 100    9.9%
  - Average ΔE / Δrisk 0.42

*Source: Appendix Table 2 and Appendix Table 3 content from the supplied PDF content unit.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12195.pdf_