On crop biodiversity, risk exposure, and food security in the highlands of Ethiopia.

Production risk is one of the quintessential features of agriculture. Unpredictable weather can expose farm households to significant production uncertainty and serious hardship.

Under harsh climatic and agroecological conditions, this can result in food insecurity and famine. The highlands of Ethiopia are a prime example of such environment. During the last forty years, Ethiopia has experienced many severe droughts, (1) leading to production levels that fell short of basic subsistence levels for many farm households (Relief Society of Tigray (REST) and NORAGRIC at the Agricultural University of Norway 1995, p. 137). Harvest failure due to drought is the most important cause of risk-related hardship of Ethiopian rural households, with adverse effects on farm household consumption and welfare (Dercon 2004, 2005). When facing prospects of harvest failure, ex-ante farm production decisions, such as crop or varietal choice, remain a part of risk management strategies (Just and Candler 1985; Fafchamps 1992; Chavas and Holt 1996; Dercon 1996; Smale et al. 1998). (2)

We argue that, in dry environments, farmers' reliance on crop biodiversity is an essential part of ex-ante risk management strategies. Diversity in genetic resources embedded in crop seeds can support productivity and help manage risk (Smale et al. 1998). Ethiopia is a recognized global center of genetic diversity for several crops, including barley (Vavilov 1949; Harlan 1992). The majority of varieties grown in Ethiopia are farmers' varieties or "landraces," which exhibit significant genetic heterogeneity.

This paper investigates how crop genetic diversity contributes to farm productivity and affects risk exposure. The analysis relies on a moment-based specification of the stochastic production function (Antle 1983). The approach captures the effects of biodiversity on the mean, variance, and skewness of production.

The evaluation of the mean and variance effects is now standard (e.g., Just and Pope 1979). However, the variance does not distinguish between unexpected bad events and unexpected good ones. On that basis, it seems important to consider skewness in risk analysis. An increase in skewness of yield means a reduction in downside risk exposure (e.g., a decrease in the probability of crop failure). The paper contributes to the existing literature by investigating three questions. First, how does risk exposure affect the incentive to use crop biodiversity as a means of reducing the cost of risk bearing? Second, what is the relative importance of crop failure in the valuation of farmer's welfare under uncertainty? Third, does the role of crop diversity vary with land quality?

The analysis relies on data from a farm survey undertaken in 1999-2000 in the Tigray region of Ethiopia. To our knowledge, this is the only available database recording Ethiopian farm-level information on crop varieties and thus on crop biodiversity. (3) Ethiopian rural households face high weather variability. Significant spatial variations exist in agroecological conditions, including topography, soil type, temperature, and soil fertility (Hagos, Pender, and Gebreselassie 1999). Different landraces of barley can perform differently across agroecological and microclimatic conditions. This poses three specific challenges for our analysis. First, we need to quantify the role of farm-specific agroecological conditions that affect both productivity and risk exposure. Second, we need to control for the effects of unobservable factors (e.g., differences across villages due to location and institutional factors). Third, we need to analyze the interplay between the farm-specific characteristics that are under farmer's control versus those that affect risk exposure.

Our analysis involves a refined econometric estimation of the production process under risk. Special attention is given to the effects of local environmental conditions and managerial decisions. Controlling for such effects is important in order to reduce the potential biases arising from omitted variables (Sherlund, Barrett, and Adesina 2002). This provides a framework to study the influence on productivity of soil quality, crop biodiversity, and their interactions (Bellon and Taylor 1993), with implications for risk management.

The econometric estimates of the stochastic production function are used to assess the welfare effects of biodiversity on production risk. Under risk aversion, risk exposure makes farmers worse off, implying a positive cost of risk (as measured by a risk premium; see Pratt 1964). Most decision makers exhibit both risk aversion and downside risk aversion (e.g., Menezes, Geiss, and Tressler 1980; Binswanger 1981; Antle 1983; Chavas and Holt 1996). Aversion to downside risk indicates that farmers have an incentive to grow crop cultivars or varieties that affect positively the skewness of the distribution of returns (thereby reducing their exposure to crop failure in drought situations). This raises the question: how important is skewness (compared with variance) in the evaluation of diversity effects on the cost of risk?

Applied to the Tigray region in Ethiopia, our analysis provides evidence on how crop biodiversity affects farm productivity, risk exposure, and household welfare and how such effects vary with soil fertility. We find several important results. First, our estimates show that biodiversity increases farm productivity in Ethiopia. Second, we uncover evidence that higher biodiversity increases variance but decreases downside risk exposure (by increasing skewness). Third, our analysis shows that the skewness effect dominates the variance effect so that higher biodiversity tends to reduce the cost of risk (as measured by the risk premium). Finally, we find that the risk benefit of biodiversity becomes larger under less fertile soils. This provides empirical evidence that biodiversity can help farmers deal with harsh climatic conditions, especially in degraded lands.

Conceptual Framework

Consider a farm producing output y using inputs x under risk. The production technology is represented by the stochastic production function y = g(x, v), where v is a vector of random variables reflecting uncontrollable factors affecting output (e.g., rainfall). The farm output y can either be consumed by the household or be marketed: y = [c.sub.1] + m, where [c.sub.1] is the part of farm output consumed by the household, and m is the marketed surplus that can be marketed at price [p.sub.1]. In general, m is unrestricted in sign. The marketed surplus can be positive (m > 0) when the farm household produces more than it consumes, or negative (m < 0) when the household produces less than it consumes. The household also consumes another good [c.sub.2] that it can purchase at price [p.sub.2]. For simplicity, assume that all prices are normalized such that [p.sub.2] = 1. The household income is: [p.sub.1]m + N(x), where [p.sub.1]m is the income generated from the marketed surplus, and N(x) denotes the net income from other activities (net of the cost of inputs x). Given [p.sub.2] = 1, the household budget constraint is: [c.sub.2] [less than or equal to] [p.sub.1]m + N(x), where m = y - [c.sub.1] = g(x, v) - [c.sub.1]. Assuming that it is binding, the budget constraint becomes: [c.sub.2] = N(x) + [p.sub.1] [g(x, v) - [c.sub.1]]. Let U([c.sub.1], [c.sub.2]) be avon Neumann-Morgenstern utility function representing household preferences under risk. Under the expected utility model, the household makes decisions so as to solve the optimization problem

(1) Max{EU([c.sub.1], N(x) + [p.sub.1][g(x, v) - [c.sub.1]])}

where E is the expectation operator based on the subjective probability distribution of the uncertain variables facing the decision maker. Under nonsatiation in [c.sub.2] (where [partial derivative]U/[partial derivative][c.sub.2] > 0), the choice of x in (1) can be written in terms of the "certainty equivalent" (CE), which satisfies

(2) U([c.sub.1], CE - [p.sub.1][c.sub.1]) = EU([c.sub.1], N(x) + [p.sub.1][g(x, v) - [c.sub.1]]).

Letting [pi] = N(x) +[p.sub.1] [g(x, v)] and following Pratt (1964), equation (2) can be alternatively expressed as

(2') U([c.sub.1], E([pi]) - R - [p.sub.1][c.sub.1]) = EU([c.sub.1], [pi] - [p.sub.1][c.sub.1])

where E([pi]) is expected income, and R is a risk premium measuring the cost of private risk bearing. The risk premium R in (2') measures the decision maker's willingness to pay for an insurance scheme that would replace the random variable [pi] by its mean. Combining equations (2) and (2') implies that the CE can be decomposed into two additive parts:

(3) CE = E([pi]) - R.

By definition, risk aversion corresponds to a positive risk premium, with R > 0. Then, equation (3) implies that risk preferences affect behavior and welfare. In general, under risk aversion, risk exposure tends to lower the CE and make the decision maker worse off. As shown by Pratt (1964), risk aversion can be assessed "locally" using the Arrow-Pratt risk aversion coefficient [r.sub.2] [equivalent to] -([[partial derivative].sup.2]U/[partial derivative][[pi].sup.2])/ ([partial derivative]U/[partial derivative][pi]). With ([partial derivative]U/[partial derivative][pi]) > 0, risk aversion corresponds to R > 0, [[partial derivative].sup.2]U/[partial derivative][[pi].sup.2] < 0, and [r.sub.2] > 0. Pratt (1964) also defined decreasing absolute risk aversion (DARA) as situations in which increasing mean income tends to reduce the risk premium R. Thus, DARA implies that increasing expected income behaves as a substitute for "insurance motives." Pratt (1964) showed that [partial derivative][r.sub.2]/[partial derivative][pi] < 0 under DARA. The empirical evidence shows that most decision makers exhibit risk aversion and DARA risk preferences (e.g., Binswanger 1981; Chavas and Holt 1996).

Print Get the Mag Weekly Updates