Microinsurance Demand After a Rare Flood Event: Evidence from a Field Experiment in Pakistan The Geneva Papers, 2014, 39 ( )

Size: px

Start display at page:

Download "Microinsurance Demand After a Rare Flood Event: Evidence from a Field Experiment in Pakistan The Geneva Papers, 2014, 39 ( )"

Brenda Henderson
5 years ago
Views:

Swiss Re Farah Said, Lahore School of Economics Uzma Afzal, Lahore School of

1 Microinsurance Demand After a Rare Flood Event: Evidence from a Field Experiment in Pakistan The Geneva Papers, 2014, 39 ( ) Ginger Turner, Swiss Re Farah Said, Lahore School of Economics Uzma Afzal, Lahore School of Economics 10 th International Microinsurance Conference November 13, 2014 Photo: pakistan flooding millions people pictures/#/pakistan flooding aerial view_25078_600x450.jpg

2 Research context Part of a 3 year collaboration ( ) supported by Lahore School of Economics, University of Oxford, The Wharton School's Risk Center, and the British Academy Other papers related to this project flood effects on risk taking, public goods participation early warning systems for natural disasters measurement and determinants of disaster resilience patronage in social networks flood impacts on health care needs, costs, and access (with World Bank and PSLM surveys)

3 Research Motivation Do 'rare' events have a persistent impact on behavior? Can we disentangle the mechanisms underlying natural disaster impacts (losses, experience, perceptions)? Policy & Practice: Better targeting relief and long term risk management Implications for other household risk taking behaviors (investment in education, migration, entrepreneurship)

4 Theoretical background Individuals may update future expectations with new information Bayes theory allows rational agents to update beliefs Observation of others experiences could be as powerful as personal experience of an event But might simplify decision making (using heuristics) Availability: over weighting more recent or salient events (memory or salience could decline over time) Conservatism: under weighting high values Representativeness: over weighting high values Reliance on risk sharing/pooling could change risk attitudes Experience with community risk sharing could reduce risk perception Being burdened by helping neighbors could increase risk perception Expecting external assistance (e.g. from government, NGOs) could reduce risk perception Other characteristics/beliefs could impact risk attitudes Past experience with or (islamic) beliefs about insurance

5 Related empirical literature Learning from experience tests in other contexts: Malmendier and Nagel (2012), Voors et al (2010) Updating climate perceptions, e.g. Deryugina (2010) Flood risk perceptions and insurance demand, e.g. Kunreuther (1978), Botzen (2012), Bubeck (2012) Post disaster experiments come to different conclusions: Cameron and Shah (2010, Indonesia), Eckel et al (2009, Hurricane Katrina); Page, Savage, Torgler (2012) Zylberberg (2010, Philippines), Reynaud (2012, Vietnam), Fleming (2012, Chile) Our contribution: combining post disaster games with detailed survey, to examine underlying mechanisms

6 20% of land underwater Est. $10B damages 1.6 m houses destroyed 20 m people affected ~2,000 deaths ~3,000 injured The Case of Pakistan July Sept 2010 Map: Damage statistics: OPM, NDMA (2011)

7 Why study risk behavior in Pakistan? 1. A rare event: 2010 was the 'worst monsoon in 80 years', yet some areas experience perennial (smaller) floods 2. Heterogeneity across the region in terms of (i) flooding levels, (ii) losses and displacement, (iii) external assistance Islamabad (15.5 inches) Rawalpindi (8.6 inches) Lahore (11.3 inches)

8 Research question Did the severe flood have a persistent impact on risk taking behavior? Photo credit: recovery framework%e2%80%99 for pakistan%e2%80%99s flood victims/

9 Hypotheses to be tested 1. Flood affected vs. non flood affected: household level flood damages are significant determinants of individual risktaking behavior Individual loss measures Observation of others' losses 2. Other flood experience (before or since 2010) reduces or compounds effect of 2010 flood 3. External assistance (government, NGOs, relatives, landlords) reduces insurance demand 4. Experiment history impacts insurance choice differently for those who had real flood experience Controls for other characteristics

10 Sampling strategy 1. Draw from representative Punjab province sampling frame of 30,000 households (2008 UN Multi Indicator Cluster Survey) 2. Compare satellite maps and government flood designations to confirm flood status and select districts with flooded and non flooded areas (Muzafargargh, Layyah, Rahim Yar Khan) 3. Select flooded and non flooded villages with similar characteristics, using propensity score matching based on pre flood characteristics from the 2008 MICS survey (flood propensity, average income, assets, education, occupation) 4. Randomly sample households within selected mauzas 16 villages (8 flood affected, 8 non flood affected) 20 households in each village 1 man and 1 woman per household Total survey sample = 640 individuals (320 in flooded villages, 320 control) Experiment sample = 384 individuals (192 in flooded villages, 192 control)

11 Sample areas

12 Data Community leader survey Check village roster from 2008 for post flood attrition and reasons Household survey modules Flood losses Household characteristics Individual survey modules Perceptions and beliefs Gender, age 3 experiments (real payments, for selected participants only) Risk aversion (lottery) Insurance demand Public goods

13 Empirical methodology Estimate the following equation by OLS and probit: G ij = binary insurance choice for individual iin round j Explanatory variables from the survey and lottery game: Lottery choice (risk aversion parameter) Demographics, income, assets Losses due to flooding, if any (2010 and others) External assistance: government, NGOs, relatives (flood and general) Psychological attributes: optimism, determinism, religious beliefs Patron / landlord relationships Community participation (informal savings/other groups) Information sources (flood and general) Financing sources (flood and general)

Experiment recruitment and procedure Post survey invitation for head of household and adult member of opposite gender closest in age Escort available for women from home to experiment location

14 Experiment recruitment and procedure Post survey invitation for head of household and adult member of opposite gender closest in age Escort available for women from home to experiment location Experiments at community building or tent 20% over recruitment in case of no shows (those who showed up but were not needed were paid Rs. 100 compensation) Experiment conducted in Urdu (or local dialect) by trained local enumerators (observed by survey firm manager and Pakistani co author) Men and women in separate rooms 1 demonstration, 1 pilot, followed by Q&A, to confirm participant comprehension Real payoffs given for a random round, unknown to participants until conclusion of all 3 experiments Participants mark choice silently on a card (limited influence from other participants) Photos: pakistan flooding millions people pictures/#/pakistan flooding food handout_25080_600x450.jpg

round offers higher payoffs and spreads Photos: http://news.nationalgeographic.

15 Experiment design: lottery Choices shown on card Probabilities known by all Each participant selects a lottery choice #1 #4 with increasing risk Random ball draw (50% blue, 50% red) 3 rounds of play Each round offers higher payoffs and spreads Photos: pakistan flooding millions people pictures/#/pakistan flooding food handout_25080_600x450.jpg

Experiment design: insurance demand Endowment of Rs. 100 ($1, or 1 day's wage) Each participant can choose whether or not to pay an actuarially fair price of Rs.

16 Experiment design: insurance demand Endowment of Rs. 100 ($1, or 1 day's wage) Each participant can choose whether or not to pay an actuarially fair price of Rs. 5 for flood insurance in each round Random draw of 1 in 5 event costs Rs. 25 for uninsured Balls drawn (with replacement) from bag that is visible to participants Experiment repeated for 20 rounds 'flood protection payment' instead of 'insurance' Photos: pakistan flooding millions people pictures/#/pakistan flooding food handout_25080_600x450.jpg

17 Insurance choice: flood experience *controlling for age, gender, flood propensity score, income

18 Insurance choice: flood recovery assistance *controlling for age, gender, flood propensity score, income

19 Insurance choice: experiment history Overall takeup is high among both flooded and non flooded Those in flooded clusters are systematically higher 'Switching' in and out of insurance is highest in early rounds but still continues for the majority of rounds

20 Insurance choice: experiment history *controlling for age, gender, flood propensity score, income

21 Insurance choice: individual characteristics and beliefs

22 Conclusions Individuals personally affected by the 2010 flood, as well as others in flood affected villages, demand significantly more experiment 'insurance', compared to non flood affected individuals Having received assistance with recovering major loss types reduces demand, controlling for loss severity (though post flood government assistance does not have a significant effect Across multiple experiment rounds, individuals who receive protection from the insurance product are more likely to continue choosing insurance in later rounds (but not significantly different for 2010 flood affected individuals) Religious views of insurance were consistent with takeup; however, contrary to popular belief, religious views of insurance are not a major reason reported in the survey for not holding actual insurance

23 Practical Applications Plans to deliver the results to government, microfinance organization (MFIs), and local insurance companies Three types of micro finance organizations in Pakistan Some micro savings groups, but little hedging savings across geographies and uncorrelated event types Insurance with Islamic principles seems to be accepted Started in 1990s with savings instruments in rural support programs Health insurance and life insurance most common in rural areas Benazir Income Support Program (BISP) is now also piloting a micro health insurance program in 15 districts by launching the Benazir Health Card through National/International Health Insurance Companies 4 regions in Punjab have village backing by MFIs MFIs give money to village to decide distribution internally Islamic saving instrument not based on interest Payouts are proportional to donations, but some weighting for need allowed These funds are just used for credit at the moment, but same principle could apply to insurance

24 Thanks The British Academy Lahore School of Economics Center for Study of African Economies, University of Oxford Risk Management & Decision Processes Center, The Wharton School, U. of Pennsylvania Travelers Foundation Sohaib Khan, Lahore University of Management Science Punjab Bureau of Statistics Comments from Shahid Amjad, Salman Asim, Wouter Botzen, Daniel Clarke, Howard Kunreuther, Mahreen Mahmud, Robert Meyer, Albert Park, Najam Shah, Jeremy Tobacman, Kate Vyborny, seminar participants at The Wharton School, Georgia State CEAR, Lahore School of Economics, and University of Oxford

27 Flood maps: satellite data Example: Jhelum district (Punjab) Flood maps combined with Google Earth Zoom to flood lines at village level Rough identification at plot level

Contract Nonperformance Risk and Ambiguity in Insurance Markets

Contract Nonperformance Risk and in Insurance Markets Christian Biener, Martin Eling (University of St. Gallen) Andreas Landmann, Maria Isabel Santana (University of Mannheim) 11 th Microinsurance Conference