Syndicate content

Some promising-ish news on rainfall insurance

David McKenzie's picture

Index-based rainfall insurance offers the potential to allow farmers to protect themselves against one of the most important risks they face – the risk of drought (or conversely too much rain). However, early experiments offering rainfall insurance proved somewhat disappointing, with take-up rates so low that there was insufficient power to see how such insurance actually changed farmer behavior (see this Finance & PSD Impact note by Xavi Gine for examples).

Two papers presented a week ago at the most recent BREAD conference offer somewhat more encouraging news, providing evidence that this insurance can lead farmers to take on more good risk, but still struggle to get impacts on profits and outputs. This first is a paper by Dean Karlan, Robert Osei, Isaac Osei-Akoto, and Chris Udry, conducted in Northern Ghana, while the second is a paper by Musfiq Mobarak and Mark Rosenzweig conducted in rural India. Both are nicely designed researcher-led experiments aimed at testing particular theories about agricultural production, rather than testing the impact of a particular policy or product per se, but offer insights for policy work in this area.

Key details

Karlan et al:

·         Aims to test whether risk or liquidity is the main barrier to farmers using more inputs with high expected returns such as fertilizer and seeds.

·         Overcome the low take-up problem in part by randomly giving insurance policies for free in the first year (note this doesn’t totally solve the problem if people don’t understand or believe the insurance will work) – and then in years two and three randomize price charged for insurance which ranges between 1/8th of actuarially fair to market price.

·         In the first year a random sample also get cash grants of $85 per acre up to 15 acres.

·         Sample sizes: 502 households allocated to 4 groups: 117 cash only, 135 insurance only, 95 both, 155 control.

·         Key results:

o   Farmers with insurance invest 13% more in cultivation, 24% more chemical fertilizer use, and 13% more land preparation costs (like tractor rental). This investment includes a shift in the mix of crops towards more rain-sensitive maize (the crop the insurance is designed for).

o   Results are noisy for the capital grants and although more fertilizer is used, the point estimate on total preparation costs is small, although again there is a shift towards more maize.

o   Point estimates show about 10% increase in harvest value, but not significant, and once the added costs are taken into account, it doesn’t seem more profitable.

o   The law of demand holds: 11% purchase insurance at market prices, rising to 42% at actuarially fair prices, and 67% when priced at a 75% discount.

Mobarak and Rosenzweig

·         Aims to examine how the demand for formal rainfall insurance varies with the informal insurance available to households (measured by risk-sharing within the jati, or sub-caste).

·         Overcome the low take-up issue by randomizing discounts (10, 50 and 75% discounts), and additionally offering bulk discounts to try and induce farmers to buy more than a trivial amount of coverage if they did buy a policy.

·         Sample sizes: 63 villages – 42 selected for marketing rainfall, 21 control: then within the treatment villages 4667 households, about half of which were farmers, allocated so roughly 7% got offered insurance at market price, 20-30% got a 10% discount, 30-35% got a 50% discount, and 30-35% got a 75% discount.

·         Also randomize which villages get a rainfall station to randomize the amount of basis risk households face (i.e. the amount of mismatch between index-based payouts and actual losses occurred).

·         Key results

o   More informal risk-sharing on average reduces the demand for formal insurance.

o   Take-up is lower the further is distance to the weather station (i.e. the higher is basis risk) – each kilometer of distance lowers take-up by 6%

o    Once both basis risk and informal risk networks are accounted for, a more nuanced finding is obtained – the interaction is positive, so that formal and informal insurance are complements when basis risk is high (as is predicted by the model in the paper)

o   Index insurance causes farmers to shift away from drought-resistant but low-yield rice varieties towards higher-yield varieties that face more weather risk.

o   No results on actual yields or profits are given

o   The Law of Demand applies, with similar slopes to as in the Ghana example

Thoughts

Both papers therefore provide evidence that insurance does encourage farmers to shift towards actions which have higher expected returns but offer more weather risk. Both reinforce the evidence that demand for this insurance is price-sensitive, and also depends on issues such as trust and past experience of payouts.

A common issue for both studies lies in how to interpret the results. The authors write nice, relatively simple models of farmer investment behavior and derive empirical implications which fit their empirical results. But, although insurance is offered randomly, interpreting the results is subject to some issues which aren’t randomized:

·         In Mobarak and Rosenzweig, they have the clever idea to try and randomize basis risk by changing whether a weather station is in the village or not. However, as well as changing basis risk, moving the weather station might also a) affect trust (I might trust the product more if I see there is a weather station in my village, rather than be told there is one in another village that I don’t see each day); b) salience – having the weather station in my village might remind me that I am insured and thus prompt me to change my actions; and c) expectations – I might update my priors of the likelihood of a drought occurring, thinking it is more likely if they go to the trouble of putting a new weather station in my village. The result is that we can’t be clear this is basis risk (I experienced a related comment like this from the editors on the paper I wrote with John Gibson on the use of GPS in household surveys, leading to the published version having to tone down claims about being able to measure the causal impact of distance).

·         Likewise the fact that farmers respond more to insurance than cash grants in Karlan et al. is taken as evidence of risk mattering. However, it could also reflect a) time inconsistency issues – individuals who know they need money in the future for crop production but know they are unable to commit to savings might respond more to money in the future than money today; b) changes in beliefs about the likelihood of bad weather occurring – individuals may think that the fact that someone is coming to give them this free insurance may either mean that it is very unlikely that bad weather will happen (making it cheap for someone to give insurance away), or may now think it is very likely bad weather is coming (which they could think is the reason for someone coming to offer them free insurance).

·         The upshot is that future work needs to try and measure a few more of the key parameters which enter into decision-making under risk: in addition to risk preferences, this should include expectations about the likelihood of risky outcomes occurring, trust, discount rates, etc. Also more work needs to be done to trace ultimate impacts on profits and household well-being, including seeing how the introduction of insurance for agricultural production affects other activities households could be doing, like non-farm businesses and migration.