The International Consortium on Agricultural Biotechnology Research (ICABR)


Biotechnology and Global Climate Change: High-Yielding Seed Varieties in Brazilian Agriculture:



Christopher Timmins, Yale University, New Haven





This paper models the crop, variety, and seed-source decisions of individual Brazilian farmers and identifies the structural parameters underlying those decisions with readily available aggregate data. Counterfactual simulations are performed with these parameter estimates in order to determine the value to farmers of specific new seed varieties, the value of individual seed attributes, the effects of Aclimate risk@ on seed and crop choices, and the role of new seed technology in offsetting the effects of global climate change on agriculture. These results have implications for policy-making in developing countries with respect to targeting research in agricultural technology in the face of a changing global climate, and in terms of valuing the returns from investments in biotechnology.


In order to measure the ability of biotechnology to offset the potentially detrimental effects of global climate change on agriculture, one must first measure its impact on agricultural productivity in a way that allows an assessment of how that impact varies with changes in climate. We focus on quantifying the value of technological change in agriculture, specifically as it enters through the introduction of new, high-yielding seed varieties.

New seed varieties are bioengineered to embody the favorable characteristics of parent seeds, while leaving-out their undesirable traits. Valuing the introduction of a new variety of seed can therefore be accomplished by treating that new variety as a new bundle of attributes. Valuing bundles of attributes is easily accomplished by hedonic estimation techniques (Rosen, 1974). Our hedonic model, which is based on techniques developed in the industrial organization literature to value the welfare impacts of the introduction of new products into consumer markets (Berry, Levinsohn and Pakes, 1995), uses observed varietal adoption practices of farmers to value the individual attributes of alternative seed varieties, and employs those values in order to assess the introduction of a new seed variety.

The decision of which seed variety to adopt is modeled as the payoff-maximizing decision of an individual farmer (indexed by i in region k). The farmer’s decision is considered to be a discrete choice between J seed options. Given the characteristics of the region k in which he is located and the specific characteristics of his farm, the farmer picks the variety that he expects to yield the greatest return. The return to farmer i from employing variety j is given by:

where Ck is a vector describing various aspects of the climate in region k, Xj is a vector of observable attributes of seed variety j (e.g., disease and insect resistance), x i is an unobservable (to the researcher) characteristic of farmer i’s land that may make variety j more or less appropriate for adoption there, and b represents a vector of parameters to be estimated. f (.) represents a set of linear and quadratic terms as well as interactions of its arguments. e i,j is an idiosyncratic shock to farmer i’s returns from adopting variety j (possibly representing unanticipated weather shocks, disease epidemics, etc...), and is assumed to be independent across both i and j and distributed type-I extreme value. j=0 represents the farmer’s decision to fallow his land; the payoffs from this decision are normalized to zero. The payoffs from adopting any seed variety are then measured relative to the payoffs that would be obtained from fallowing land).

Each farmer is assumed to choose the variety of seed (vj) that maximizes his payoffs that growing season. This implies that farmer i chooses vj if

p i,j p i,m " m

Under the distributional assumption on e i,j, this implies a closed-form solution for the probability that farmer i chooses variety j:

while the probability that farmer i chooses to fallow his land is given by:

Given an assumed distribution for the unobserved farm characteristic, within region k, Fk (x i), which is parameterized by the vector g k, the share of land in region k devoted to variety j is given by:

s j,k ( b , g k ) = P [ vj Ck , Xj, x i; b ] (x i ; g k ) dx i j = 1,2, .... J


Observed shares of land in each region devoted to each variety (sj,k j = 1,2, …,J) can then be used in conjunction with s j,k(b ,g k) to estimate b and g k with the Generalized Method of Moments estimation technique (Greene, 1993).

We can make use of these estimated parameters to determine the value of a new seed variety (i.e., a new bundle of Xj’s), given that farmers will re-optimize in their variety decision after the introduction of that new alternative. First, given the estimated values of b and g k, simulate p i,j for each farmer and determine his optimal choice of variety. The aggregate payoff for region k can then be determined by integrating the maximal payoff for each farmer in the region over x i. Denote the aggregate payoff to farmers in region k when J seed varieties are available as P Jk.

Next, we add a new variety (indexed by J+1), with a new vector of characteristics (XJ+1), and calculate the payoffs from each seed variety that can now be achieved by each farmer in region k. By integrating the payoff from the optimal seed choice of each farmer over x , we can calculate the region-k aggregate payoff when J+1 seed varieties are available, P J+1k. The ratio (P J+1k/P Jk) measures the rate of change in maximal payoffs to farmers in the region as a result of the introduction of variety J+1.

The interaction of climate with the introduction of variety J+1 can be easily quantified in this framework. In particular, (P J+1k/P Jk) can be calculated both under current climate conditions in region k (i.e., Ck) and under predicted climate conditions after significant greenhouse warming has taking place (i.e., Ck). The difference between these two rates describes the effect of climate change on the agricultural productivity effects of the new seed variety.

Another use of this estimation methodology is in the valuation of an input specific seed characteristics (e.g., "How much is disease resistance worth when supplied as to payoff maximizing farmers?"). This value can be determined by simply calculating P Jk as above, and then recalculating it with the trait in question set to zero in all varieties, P - rk. The ratio (P - rk/P Jk) measures the rate at which maximal aggregate payoffs are reduced when the trait in question is removed from the seed pool.



Home Page Program Registration Ravello and surroundings