Primary Outcomes (end points)
The primary outcome of interest is index insurance demand conditional on contract terms. This is elicited through a binary choice between $20.000 COP close to harvest or $500.000 COP conditional on rainfall, measured in location x, exceeds a given threshold y. Locations vary between Tecnicafé (our local implementing partner), Bogotá Airport, and Santiago, Chile (where we have access to live weather station data). The data generated are thus 45 binary buy/no-buy choices, for different contract terms, for each participant. These are elicited at baseline and endline.
• Rainfall thresholds are drawn randomly from a skewed triangular distribution, ranging from 0 to 313 mm.
• Locations are randomly drawn from all orders that ensure no repeated locations.
Sufficient variation in rainfall payment thresholds y, results in some products never paying, while others will practically always pay. Conditional demand means we estimate responsiveness to product characteristics, both at the participant level and in the aggregate. We also estimate the contract characteristics at which demand probability is 50%, or their indifference point.
• These response curves are to be estimated, with and without participant fixed effects, and with uniform vs. non-uniform noise assumptions, and normal and extreme-valued noise distributions. Each of these methods jointly identifies indifference points (preferences) and responsiveness:
Every four questions participants also reflect on their last binary choice, by answering their level of cognitive uncertainty, as well as their subjective probability of the insurance product paying out next year. We can thus also directly compare self-reported levels of perceived precision.
We also elicit willingness to pay for a subset of the shown insurance products in the binary choice module. This is elicited via a Becker-Degroot-Marshak mechanism, with participants answering their willingness to pay for a random subset of 18 products already presented as binary options.
We estimate welfare first by comparing the proportion of unfavorable products purchased/unfavorable products not purchased, for a range of risk-aversion parameters. We also estimate risk-aversion directly via a real effort incentivized task, in which
Main treatment effects are sensitivity to product characteristics, as measured in the slope of the response curve and indifference points, with heterogeneity analysis based on education level and experienced climate. Changes in self-reported cognitive uncertainty is also a main treatment effect, as it shows participants’ self-awareness of the degree of choice uncertainty. Pre and post measurements are used to increase statistical power in our estimates.