Does the design matter? Comparing satellite-based indices for insuring pastoralists against drought

Abstract

For those developing satellite-based insurance products, there is no consensus in the scientific community on which of the many available indices most accurately track agro-ecological shocks as experienced by farmers and pastoralists. Furthermore, metrics commonly used by the remote sensing community for assessing the accuracy of indices in retrieving biophysical variables are insufficient in the case of insurance, because they do not consider the value of insurance coverage in terms of household welfare. This study begins to fill this knowledge gap by bridging two index insurance literatures: the remote sensing science literature that focuses on the predictive power of indices, and the economic literature that focuses on welfare outcomes. The article uses a longitudinal panel of household survey data from Kenya to compare the quality of existing and potential insurance products. These products are developed from different processing chains applied to time series of the satellite-based Normalized Difference Vegetation Index (NDVI). Although the indices are highly correlated to each other (ρ > 0.95), a utility analysis provides insight into how small differences can lead to larger differences in product value. Our results highlight that index accuracy, cost, and timeliness of payments must be considered jointly when assessing insurance quality for clients.

Introduction

Index insurance—whereby insurance payments are triggered by an index that correlates with the insured risk—is promoted as a low-cost approach to providing formal insurance, offering the possibility of opening insurance markets to households that were previously excluded from conventional insurance products (Barnett and Mahul, 2007; Jensen and Barrett, 2017). Index insurance is now used in multiple contexts and by various government and development organizations to protect the vulnerable from the sometimes devastating impacts of weather-related shocks (Greatrex et al., 2015). However, index insurance is, by definition, an imperfect type of insurance, as it relies on a single index (e.g., rainfall measurements, average yields of a unit) to determine individual payments, rather than an individual's losses. The challenge for index insurance design is to identify indices that closely track outcomes and to develop policies that offer valuable risk protection to clients.

Generating valuable index insurance policies requires several decisions regarding data selection, processing, and contract design to maximize the protection provided by the product (de Leeuw et al., 2014), while also considering operational features including the costs associated with accessing and processing indices. Protecting households from shocks effectively requires a low basis risk, i.e. the gap between actual losses and the insurance payments they receive. Basis risk is inherent in all index products and is often found to play an important role in the (low) demand for index-insurance (Giné et al., 2007; Mobarak and Rosenzweig, 2013; Hill et al., 2013; Jensen et al., 2018). Even more troubling are the implications of basis risk for those promoting index insurance as a tool to fight poverty and spending resources to increase uptake of index insurance among the poor. Low quality index insurance products can fail to protect households against adverse shocks and could even increase their vulnerability to shocks because insured households have fewer resources on hand due to premium payments (Clarke, 2016).

For this reason, there is a growing effort to analyze and improve the quality of the protection offered by index insurance (e.g., Barré et al., 2016; Clarke et al., 2012; Elabed et al., 2013; Jensen et al., 2016). To maximize value, contract design needs to address critical aspects related to the relationship between the index and losses, preferences and constraints faced by prospective of clients, frequency and timeliness of payouts, the simplicity and flexibility of the contract, as well as the modalities of payouts. Despite the growing efforts, there is no uniformity in the way these important quality aspects of index-insurance products have been evaluated.¹

In this paper, we focus on factors of index insurance quality that are rarely assessed using micro-level data and economic foundations: the processes used to develop indices from raw satellite data and their implications for contract design and therefore product value. Until now, alternative choices for these factors have not been evaluated against households' actual losses. Our study aims to bridge the two literatures from economics and remote sensing to assess the impact of the processes used to develop indices on index insurance quality, measured in terms of household economic wellbeing. In doing so, it assesses the potential of alternative index processing solutions to improve the performance of index insurance products with the intention of guiding the growing number of cash transfer and index insurance products that rely on remote sensing technologies to protect households from covariate shocks.²We use the case of the Index-Based Livestock Insurance (IBLI) project in Kenya to perform this analysis. IBLI insures clients against the risk of livestock forage scarcity and is based on satellite-derived time series of the Normalized Difference Vegetation Index (NDVI), a radiometric index sensitive to vegetation health and status. The insurance is commercially sold to herders in arid and semi-arid land areas (ASALs) of Northern Kenya and Southern Ethiopia. Originally designed by Chantarat et al. (2013), the IBLI contract has evolved in a constant effort to understand the factors limiting its quality and to improve its design (Jensen et al., 2016). For example, the index-units—the geographic space aggregated to calculate a common index—have recently been updated to reflect input from prospective clients on the rangelands that they commonly access. Another manifestation of this effort, which is examined in this manuscript, is the shift from an initial asset replacement program, in which estimated area-average livestock mortality triggered payments after a drought seasons, to an asset protection logic, whereby a forage scarcity index allows for payments prior to livestock losses. Early payments can help herders take measures to avoid livestock death, such as buying forage, water or medicine. However, providing earlier payments may lead to larger basis risk because it draws on data from a shorter period of environmental conditions.

Given these considerations, we investigate two important aspects of index insurance connected with the accuracy of the index and the timing of payouts to estimate how the generated indices correlate with livestock mortality rate and affect expected utility.

• The NDVI data product: There are different methods available for processing raw satellite data to produce a clean NDVI signal. The type of processing has implications for both accuracy and the timing at which the index become available. We compared the eMODIS NDVI product provided by the United States Geological Survey and currently used for IBLI with BOKU's MODIS NDVI product provided by BOKU University and used by Kenya's National Drought Management Authority (NDMA) to monitor drought conditions and to trigger transfers of disaster contingency funds (Klisch and Atzberger, 2016). The products differ in (1) their level of pre-processing, and (2) when they become available, as explained in Section 4.2.

• The NDVI temporal aggregation: We compared indices that combine NDVI data from a rainy and dry season (as initially used in IBLI) and are generated at the end of the dry season, with the recently proposed indices by Vrieling et al. (2016) that rely only on information from the vegetation growing (rainy) period, which take place during the first few months of the season and allows indemnity payments to be made 1–3 months earlier than for the initial IBLI design.³

For the analysis, we first compared the indices using methods that are commonly used to examine index accuracy. We then analyzed how meaningful the between index formulations are for household welfare. Our results relate to a growing number of studies looking at policy design options to best allocate limited public resources and maximize impacts on beneficiaries (Dhaliwal et al., 2013; Jensen et al., 2017), and by focusing on household welfare in particular, it also relates to the literature on targeting poor and vulnerable households (Stoeffler et al., 2016a, Stoeffler et al., 2016b).