Using weather index insurance to improve drought response for famine prevention.

There is a strong link between weather and the welfare of poor populations. Low-frequency, short-term, but catastrophic weather shocks can trigger destructive coping responses to disaster--for example, withdrawal of children from school, distress sale of assets, refugee migration, crime, and severe human suffering. Moreover, these adverse impacts often persist as children's physical growth falters, and household productivity, asset accumulation, and income growth are dampened (Dercon and Krishnan 2000; Hoddinott and Kinsey 2001; Hoddinott 2006). The prospect of such shocks may also induce underinvestment in assets at risk, limiting poor households' ability to grow their way out of poverty over time (Carter and Barrett 2006).

The problem originates with the difficulty poor households face in insuring covariate risk. While informal social insurance arrangements and flexible credit contracts often provide the poor with significant insurance against household-specific, idiosyncratic risk, when entire communities or social networks confront the same biophysical shock, their capacity to buffer members' welfare may be insufficient to prevent severe and widespread human suffering. The magnitude and intensity of such suffering sometimes merits the label "famine" (Howe and Devereux 2004). External (domestic and international) relief organizations and governments commonly step in to provide emergency assistance in the wake of catastrophic covariate shocks such as drought, especially when the specter of famine looms. Operational agencies and the donor community are thereby financially exposed to catastrophic weather risks in developing countries via their humanitarian commitment to emergency response.

In addition to their potential for other purposes (Barnett, Barrett and Skees forthcoming; Alderman and Haque 2007), recent innovations in index insurance show promise as a means to facilitate improved emergency response to weather-related catastrophic shocks that threaten famine. Just as improved early warning systems and emergency needs assessment practices have used timely monitoring and analysis of vulnerable areas to significantly improve humanitarian response in recent decades (Barrett and Maxwell 2005), so too can weather index insurance facilitate further improvement by addressing several key remaining weaknesses in global famine prevention efforts. This paper briefly outlines how donors and operational agencies might use weather index insurance for famine prevention, enumerates key prospective benefits from such products, and then illustrates the possibilities with an application to the arid lands of northern Kenya, an area of recurring severe droughts that elicit massive international humanitarian responses.

How to Use Weather Index Insurance for Famine Prevention

Weather index insurance pays claims based on realizations of a weather index that is highly correlated with an outcome variable of interest. The insurance policy specifies an event or threshold at which payments are triggered and a payment schedule as either a lump sum or a function of index values beyond that threshold. The pricing of the product is based on the underlying payment schedule and the probability of realizations of the index that might trigger indemnity payments. Those probabilities are typically derived from historical rainfall records (Turvey 2001).

In slightly more formal terms, the key to designing a weather index insurance product is the existence of some observable relationship, y = f(W, X) + [epsilon], where y is some outcome variable of interest, W represents one or more weather variable of interest (e.g., rainfall), X are other covariates that condition changes in y and that may be correlated with W, f([??]) is a general function, and e is a standard mean zero disturbance term. One will typically use time series observations on the variables to estimate some parametric relation that may involve multiple lags of the independent variables, polynomials in those lags to allow for non-linearities, etc. The key is that the specified relationship explains much of the variation in y and successfully forecasts out-of-sample.

Assuming f([??]) is invertible, and given a threshold level of y at which one wants to trigger a response, [y.bar], and observable X, one can specify and estimate a version of the previous equation and then recover a trigger level for W, [W.sup.*] (Turvey 2001) at which E[f(W, X)] = [y.bar]. Thus [f.sup.-1]([y.bar], X) = [W.sup.*]. It is also possible to estimate the pure reduced form relation y = h(W) + [psi] and similarly derive a threshold value for the weather index 1 if one cannot observe X or if the cost of making such observations exceeds the marginal gains in predictive accuracy. The value of the pure reduced form is that the forecasted human impact conditional on observed weather h(W) depends solely on observed weather, and thus it is objective, verifiable and independent from human manipulation. Therefore, f(W, X) and h(W) offer two alternative forms for a parametric index that proxies the risk associated with observed weather events.

Most commonly, the outcome variable reflects economic losses. In the present case, however, we are interested in measures of severe, widespread human suffering, that is, of famine. The dependent variable we use is the proportion of children aged 6-59 months in a community who suffer a mid-upper arm circumference (MUAC) z-score [less than or equal to] -2. (1) As a measure of wasting, MUAC reflects short-term fluctuations in nutritional stress and is typically easier and less costly to collect than weight-for-height, the most commonly used anthropometric measure of wasting. Furthermore, several studies have found MUAC a far better predictor of child mortality than weight-for-height (Alam, Wojtyniac, and Rahaman 1989; Vella et al. 1994). We follow Howe and Devereux's (2004) definition of famine as a condition where 20% or more of children in a specified area are severely wasted (z [less than or equal to] -2).

Historically, "most famines in poor economies are associated with the impact of extreme weather ... [and] the worst famines have been the product of back-to-back shortfalls of the staple crop" (O Grada 2007, p. 7). While weather shocks are neither necessary nor sufficient to induce famine, there is a strong historical correlation that can potentially be exploited. Our preliminary work with detailed data from three districts in northern Kenya finds a strong historical relationship between community-level MUAC indicators--in particular, the proportion of a community's children with MUAC z-score [less than or equal to] -2--and lagged rainfall indicators, with considerable out-of-sample forecast accuracy (Mude et al. 2006). This offers a promising platform on which to build weather insurance for drought response.

The Potential Gains of Weather Index Insurance for Drought Response

There have been a number of recent experiments with weather index insurance programs for protection against disasters. The best-known example is the Mexican public reinsurance program, Agroasemex, which has marketed weather index insurance policies to state governments to insure against drought, and which has links to the National Fund for Natural Disasters, FONDEN (Alderman and Haque 2007).

Weather insurance offers several different, potentially major improvements to the global response to climate-related, slow-onset emergencies such as drought. First, insurance by its nature enables the insured to smooth its stream of payments. Rather than incurring irregular, massive expenses for emergency response, one pays a far smaller amount regularly in the form of insurance premium, but receives large indemnity payments when resources are needed. Given liquidity constraints and the value to expenditure smoothing, such smoothing should be advantageous to operational agencies and donors if such insurance can be reasonably priced in the market.

[FIGURE 1 OMITTED]

Second, the irregularity of need for famine prevention resources underscores the value of insurance for low-probability, high-impact events as part of an effective risk-layering strategy. Communities can easily absorb modest variability in rainfall. In our setting, pastoralists in northern Kenya have developed highly adaptive livelihood strategies over many centuries of coping with volatile rainfall patterns. So a layer of individual and community-level self-insurance is feasible. Bigger covariate shocks commonly demand some outside interventions. Agencies and donors can readily handle small-scale crises within their usual budgets and operational mandates. The problem emerges when rare, widespread and devastating shocks occur and probabilistically threaten famine. With insurance in place to provide resources necessary for such low frequency but potentially catastrophic weather events, other actors can focus better on insuring the range of commonplace risks over which they possess comparative advantage.

Third, index insurance would permit an improved and immediate link between emergency response and recipient need. With most emergency response still based on the provision of food aid that remains tied to procurement, processing, and shipment from donor countries, drought response for famine prevention remains doubly tied: to food as the primary form of response and to donor countries as the primary source of that food. Indeed, a common quip in Ethiopia is that the availability of food aid depends not on whether it rains locally, but on whether it rains in North America. Put differently, the supply of food aid--which is a complex function of donor country harvests and farm support policies, global prices, freight costs, geopolitics, etc.--remains as important a determinant of food aid deliveries as is the need of at-risk populations. This is partly reflected in figure 1, which plots rainfall realizations in the three northern Kenya districts we study (Marsabit, Samburu, Turkana) against the value of World Food Programme (WFP) food aid deliveries into Kenya. (2) Over the period 1991-2006, this relationship was quite weak ([r.sup.2] = 0.067 on the best fit, single log specification), and the difference between annual food aid flows in the wettest and driest years in this period was only $4 million (16% higher in the drier year) even though rainfall volumes in the better year were 334% greater than those in the driest year. Current food aid programs are not responsive enough to drought shocks, at least partly due to supply-side obstacles that could be reduced via the proposed weather index insurance, which links cash payouts entirely to predicted humanitarian need.

Fourth, timely and adequate funding are huge obstacles to effective response to slow-onset disasters, such as drought. The United Nations' Consolidated Appeal Process (CAP) attempts to coordinate global cooperation in emergency response. On average, however, funds raised via CAP amounted to only 56% of requirements by the end of October in 2003-2006 (OCHA). WFP Emergency Operations (EMOP) covers the majority of the humanitarian response, especially related to food security and famine prevention. While WFP's experience is better than that of the CAR it too suffers significant shortfalls and delays. On average, only 70% of EMOP funding needs were provided by donors in 2001-2006, ranging from 57% in 2005 to 79% in 2004, and each year, only an average of 36% of EMOP needs were confirmed for donors' contributions by the beginning of June for early intervention with as low as 22% need fulfillment in mid-2004 (WFP). Moreover, donor contributions take months to arrive. For example, the median response time for U.S. emergency food aid is just under five months from the filing of a formal request (a "call forward") to port delivery (Barrett and Maxwell 2005). Delays are costly, even deadly. As an emergency progresses, unit costs per beneficiary increase sharply as more expensive, processed commodities become increasingly needed for therapeutic feeding, donors pay premium for faster transport (including airlift), and populations migrate to camps where broader support costs (e.g., shelter, water, medical care) become essential. Moreover, late-arriving assistance often fails to protect the livelihoods of affected populations, who often must deplete their productive asset stocks or migrate in response to the shock, which in turn makes them more vulnerable to future shocks.

In spite of significant improvements in early warning systems, supply chain management and other key response functions, operational agency interventions continue to lag behind global media reporting on disasters. The 2004-2005 Niger emergency provides a disturbing example. After a November 2004 international appeal by the Government of Niger and the United Nations, WFP's initial food deliveries in February 2005 cost $7 per beneficiary. But global response was anemic. In June 2005, the Niger situation was relabeled an "emergency," and graphic global media coverage in July-August led to a sizable, but slow, global response. The cost per beneficiary for WFP's August deliveries--that is, the same delivery organization, but with badly delayed response--had risen to $23, more than three times the cost six months earlier, due to far greater need for supplemental and therapeutic foods instead of cheaper, bulk commodities, and the need for airlift and other quicker, but more expensive logistics. By enabling rapid payout when the trigger is reached rather than merely starting an appeals process likely to drag on for months and be only partly filled, weather insurance can substantially reduce drought response costs and provide greater asset protection to affected peoples.

Finally, because index insurance is based on the realization of a specific-event outcome that cannot be influenced by insurers or policyholders (e.g., the amount and distribution of rainfall over a season), it has a relatively simple and transparent structure. This makes such products easier to understand and consequently to design, develop, and trade, potentially opening up new sources of finance for emergency drought response and famine prevention. The apparent success of pilot programs conducted in India, Malawi, Mexico, Mongolia, and various other countries has established the feasibility and affordability of such products (World Bank 2005). Weather insurance contracts underwritten by domestic insurers and reinsured or underwritten directly by international investors can provide a new and cost-effective means to transfer low-probability, high-consequence covariate weather risks to global markets where those risks can be easily pooled and diversified as part of global portfolios. If rainfall volumes provide a strong predictive signal of imminent famine, and thus of looming financing needs for emergency drought response, the opportunity exists to design weather insurance to facilitate more effective aid response. This opportunity should be seized.

Rainfall and Famine in Northern Kenya: The Potential of Weather Index Insurance

The arid areas of northern Kenya are largely populated by marginalized pastoral and agro-pastoral populations that traditionally rely on extensive livestock production for their livelihood. We focus on three districts--Turkana, Samburu, and Marsabit--not only because they are the three districts rated most vulnerable to food insecurity, but also because they share similar socioeconomic characteristics, climate patterns, natural resource endowments, and livelihood portfolios which allows us to apply similar concepts and tools to drought response across this vast area.

The unpredictability of rainfall heavily affects livelihood returns and welfare dynamics in pastoral communities. To observe such dynamics, Mude et al. (2006) generated community-level summary statistics of repeated cross-sectional household data collected monthly in 45 communities in these three districts from 2000-2005 by the Government of Kenya's Arid Lands Resources Management Project (ALRMP), which resides within the Office of the President, underscoring the importance of drought response in these regions. The key dependent variable is the proportion of children aged 6-59 months in each community with recorded MUAC z-score [less than or equal to] -2.

Mude et al. (2006) matched the ALRMP data with forage availability data from the USAID Global Livestock CRSP livestock early warning system (LEWS) and livestock information network and knowledge system (LINKS) project, and with METEOSAT-based rainfall series, 1961-2006, from 21 geographically distinct sites in these three districts. While floods occur and cause major losses, the primary weather-related risk in these districts is severe drought. Rainfall is generally bimodal, characterized by long rains that fall from March through May and short rains from October through December. Rainfall is also highly correlated across space in these districts. Table 1 displays the bivariate correlation coefficients of mean district-level cumulative seasonal rainfall, 1961-2006, with the long rains on the lower diagonal and the short rains on the upper diagonal. The high correlations among these series--all are statistically significantly different from zero at the one percent level-signal limited weather risk pooling potential in northern Kenya, hence the need for outside assistance when severe droughts strike.

Pastoralists rely on both rains for water and pasture for their animals, as well as occasional dryland cropping. In a normal year, water availability suffices to ensure adequate yields of milk, meat and blood, most of which is consumed within pastoral households, with the rest sold in order to purchase grains and non-food necessities. Localized rain failures may happen, but migratory herders can commonly adapt to spatiotemporal variability in forage and water availability. But when the rains fail across a wide area, especially if short and long rains both fail in succession, catastrophic herd losses often occur and bring with them severe human deprivation manifest in, among other indicators, more prevalent severe child wasting.

Figure 2 plots mean monthly rainfall volumes across these three districts along with the percentage of the 21 sites in which the short and/or long rains failed, where "failure" reflects cumulative rainfall more than one standard deviation below its long-term, site-specific mean. Three major recent droughts had dire humanitarian consequences: 1997/8, 2000/1, and 2005/6. Aggregate rainfall was low in all of these years, and the drought conditions were spatially widespread and continued across multiple seasons. Mude et al. (2006) show that drought episodes are strongly associated with dramatic herd losses due to mortality, lower livestock lactation rates, and a sharply higher prevalence of severe child wasting. Intriguingly, they also find that forecasts of severe wasting prevalence generated from a relatively simple model based on a small set of variables that ALRMP regularly monitors yields highly accurate out-of-sample forecasts with a lead of three months. Rainfall is the key explanatory variable. It seems that observed rainfall patterns may be useful in predicting and insuring against famine.

In this setting, designing weather index insurance to facilitate financing of drought-related humanitarian response appears attractive. We conceptualize two ways in which weather insurance can be effectively designed to serve this goal. The first is a simple put option based on cumulative long rains (March-May) and/or cumulative short rains (October-December)--appropriately weighted across rainfall sites--as a weather index. This might pay out some pre-determined sum per mm shortfall of seasonal cumulative rainfall relative to a contractually established threshold at the end of the contract term for each season. To take into account the intensity of droughts in cases of severe rainfall deficit, the option payout could be a convex function of the seasonal cumulative rain shortfall. Payout could be even simpler, a lump sum payment at the end of the contract term if seasonal cumulative rainfall fell below the threshold. As historical data show that seasonal rain shortfalls are strongly associated with the emergence of famine conditions, even such simple insurance seems to offer a reasonable hedging tool for organizations committed to humanitarian drought response. The simple nature of such contracts can potentially increase reinsurance opportunities and thus lower the prospective price of such insurance in international markets. As local droughts within districts can effectively be handled by traditional means, it might also be more cost effective to write a single contract for the whole area rather than for each district separately.

[FIGURE 2 OMITTED]

The second weather index insurance design exploits the apparent ability to forecast famine based on rainfall several months ahead. Specifically, one could use a validated forecasting model to establish the rainfall level below which the expected future prevalence of child wasting equals or exceeds 20%, thereby triggering indemnity payments under the insurance contract. The model would be specified in the contract and new forecasts generated in near real-time based on the arrival of weather data. The weather index evolves continuously and can therefore better capture not only the impact of shortfalls in rainfall quantity but also the timing and distribution within a season as well. The forecast model can readily incorporate monthly or seasonal dummy variables and location-specific dummies, in short, any other covariates that affect the dependent variable of interest that can be objectively verified and cannot be manipulated by parties to the contract. The nonstandard nature of this contract might make it somewhat harder to price and sell in financial markets. Weather-based famine index insurance of this sort could complement existing appeals-based systems based more on realizations of human suffering, thereby facilitating faster, lower-cost intervention based more directly on anticipated need and less on supply-side conditions in food aid donor countries.

The famine insurance we envision, especially the second variant, differs in a few key ways from the well-publicized drought insurance contract that WFP took out for Ethiopia with AXA Re in 2006. First, that contract did not use any weather stations from the country's pastoral regions, on which we focus. Second, the weather risks were quantified in terms of expected income loss by at-risk populations based on estimates of the elasticity of crop production to rainfall at different stages of crop growth. Crop- and area-specific estimates were aggregated, mapped to income via price estimates, and then converted into a livelihood loss index. Our design is to tie rainfall directly to a human outcome of interest rather than to indirect measures and to use the commonplace superiority of reduced form forecasting over those based on structural models. Third, the 2006 Ethiopia drought insurance contract covered the entire agricultural season, consisting of two rainy seasons, from March to October, and triggered payment by the end of the contract (in October) when data gathered throughout the contract period indicated that rainfall was significantly below historic averages, pointing to the likelihood of widespread crop failure. The product we envision would pay out at any time within the contract period once the model predicts a prevalence of severe child wasting that meets or exceeds the pre-specified trigger level. Thus, if the seasonal rains failed badly and widely, this might trigger indemnity payments well before the end of the contract so as to allow more effective and lower cost intervention. In parallel work, we explore the framework for pricing such contracts (Chantarat et al. 2007).

We gratefully acknowledge financial support from the Global Livestock Collaborative Research Support Program (GL CRSP), funded by the Office of Agriculture and Food Security, Global Bureau, USAID, under grant number DAN-1328-G-00-0046-00. The views expressed here are those of the authors and do not represent any official agency. Any remaining errors are our own.

References

Alam, A., B. Wojtyniac, and M. Rahaman. 1989. "Anthropometric Indicators and Risk of Death." American Journal of Clinical Nutrition 49:884-88.

Alderman, H., and T. Haque. 2007. "Insurance Against Covariate Shocks: The Role of Index-Based Insurance in Social Protection in Low-Income Countries of Africa." Working paper 95, The World Bank.

Barnett, B.J., C.B. Barrett, and J.R. Skees. Forthcoming. "Poverty Traps and Index-Based Risk Transfer Products." Worm Development, in press.

Barrett, C.B., and D.G. Maxwell. 2005. Food Aid after Fifty Years: Recasting Its Role. New York: Routledge.

Carter, M.R., and C.B. Barrett. 2006. "The Economics of Poverty Traps and Persistent Poverty: An Asset-Based Approach." Journal of Development Studies 42(2):178-99.

Chantarat, S., C.G. Turvey, A.G. Mude, and C.B. Barrett. 2007. "Improving Humanitarian Response to Slow-Onset Disasters Using Famine Indexed Weather Derivatives." Paper presented at the EAAE annual meeting, Berlin, 5-7 July.

Dercon, S., and R Krishnan. 2000. "In Sickness and in Health: Risk-Sharing within Households in Rural Ethiopia." Journal of Political Economy 108:688-727.

Hoddinott, J. 2006. "Shocks and Their Consequences across and within Households in Rural Zimbabwe." Journal of Development Studies 42(2):301-21.

Hoddinott, J., and B. Kinsey. 2001. "Child Growth in the Time of Drought." Oxford Bulletin of Economics and Statistics 63(4):409-36.

Howe, R, and S. Devereux. 2004. "Famine Intensity and Magnitude Scales: A Proposal for an Instrumental Definition of Famine." Disasters 28(4):353-72.

Mude, A.G., C.B. Barrett, J. McPeak, R. Kaitho, and P. Kristjanson. 2006. "Empirical Forecasting of Slow-Onset Disasters for Improved Emergency Response: An Application to Kenya's Arid and Semi-Arid Lands." Working paper, Cornell University.

O Grada, C. 2007. "Making Famine History." Journal of Economic Literature 95:5-38.

Turvey, C.G. 2001. "Weather Derivatives for Specific Event Risks in Agriculture." Review of Agricultural Economics 23(2):333-51.

United Nations Office for the Coordination of Humanitarian Affairs (OCHA). 2007. Humanitarian Appeals. Available at: http://ochaonline.un.org/humanitarianappeal/

Vella, V., A. Tomkins, J. Ndiku, T. Marchal, and I. Cortinovis. 1994. "Anthropometry as a Predictor for Mortality among Uganda Children, Allowing for Socio-economic Variables." European Journal of Clinical Nutrition 48:189-97.

World Bank. 2005. Managing Agricultural Production Risk: Innovations in Developing Countries. Washington D.C.

World Food Programme. 2000. Food and Nutrition Handbook. WFP, Rome.

--. 2001-2006. Estimated Needs and Shortfalls for WFP Operational Activities. Available at: http://www.wfp.org/YellowPages.

(1) MUAC data are standardized using the international standard 1978 CDC/WHO growth chart. The threshold z [less than or equal to] 2 is consistent with the famine benchmark often employed by emergency relief agencies (Howe and Devereux 2004; World Food Programme 2000).

(2) The food aid figures, obtained from WFP annual reports, reflect deliveries into the whole of Kenya, not just the northern three districts we study. Unfortunately, we could not obtain district-level disaggregated figures. However, these three districts were among the leading recipients of food aid within the country over this period, thus we are confident that the basic patterns are satisfactorily reflected in these data.

Chantarat is Ph.D. candidate, Barrett is International Professor and Turvey is W.I. Myers Professor of Agricultural Finance, all at Cornell University: Mude is Research Scientist, International Livestock Research Institute, Nairobi, Kenya.

This article was presented in a principal paper session at the AAEA annual meeting (Portland, OR, July 2007). The articles in these sessions are not subjected to the journal's standard refereeing process. Table 1. District-Level Seasonal Rainfall Correlations, 1961-2006 District Turkana Marsabit Samburu Turkana 0.60 0.90 Marsabit 0.71 0.72 Samburu 0.86 0.87