Pareto optimal trade in an uncertain world: GMOs and the precautionary principle.

Genetically modified organisms (GMOs) and applications of the Precautionary Principle (PP) to GMOs have recently been the focus of several agricultural trade disputes. Acceptance of GMOs, as well as use of the PE varies. However, the conventional wisdom is that the United States is openly favorable (opposed) to GMOs (the PP), and that the European Union (EU) is openly favorable (opposed) to the PP (GMOs).

Because of its heavy reliance on GMOs, U.S. agriculture seems peculiarly vulnerable to precautionary activities. An example illustrates. Between October 1998 and May 2004, the EU authorized no new GMOs for release (Commission of the European Communities 2000). The U.S. corn industry estimates that it lost more than $1 billion because of this de facto GMO moratorium on approvals. And in May 2003, the United States, together with several other countries, initiated World Trade Organization (WTO) dispute settlement proceedings against the de facto moratorium. In response, the EU relaxed legal restrictions on certain genetically modified foods, and in May 2004, it authorized the import of Syngenta's Bt-11 corn for food and animal use. This was followed by authorization of Monsanto's Roundup-Ready corn for animal use (July 2004) and for human consumption (October 2004). On February 7, 2006, the World Trade Organization issued a preliminary ruling that the EU's moratorium was illegal. The dispute has not yet been completely resolved.

Speaking in broad terms, the U.S. position is that no reputable scientific evidence exists that currently approved GMOs harm human health. Because the United States argues that its position is based on "... scientific risk-based assessment of GMOs" (see, e.g., Sheldon 2004), its stand may seem compelling. But, in stark contrast, the intellectual linchpin of the EU position is the principle of scientific uncertainty. Scientific uncertainty characterizes risks that are so imperfectly known that it is impossible to attach science-based probabilities to them. Put simply, the EU position is that "... scientific risk-based assessment of GMOs" is not currently possible.

Scientific uncertainty characterizes situations where hazards associated with an activity are either imperfectly known or cannot be assessed accurately in a probabilistic framework. There are many reasons to believe that either of these preconditions can be met for many economic activities. Paramount among these is simple ignorance. Human experience is littered with instances where exposure to products (e.g., tobacco, asbestos, DDT, and many currently known carcinogens) entailed hazards that were unanticipated at the time of product introduction.

Even when hazards are properly anticipated, the elicitation of a probability distribution characterizing that hazard is frequently very difficult. A large amount of time and expenditure is typically required to provide exact probability assessments. Moreover, it is routine to encounter "scientific experts" who vehemently disagree with the resulting probability assessments or the statistical evidence (Levi 1980). In the statistics, philosophical, and artificial intelligence literatures, the difficulty in agreeing on single probability measures is manifested in the long-standing and rapidly burgeoning literatures on "imprecise" probabilities (Walley 1991) and robust statistics (Huber 1981). For these and other reasons, distrust of "experts," especially "official experts," is apparently widespread. In fact, empirical evidence suggests that the better educated an individual is, the less likely is he or she to trust information provided by government, private industry, or public interest groups (Huffman et al. 2004).

Scientific uncertainty is undoubtedly present in many economic activities. Take "mad-cow" disease in the United Kingdom. Despite repeated official assurances by the British government that the disease could not be transmitted to humans, highly publicized outbreaks of its human variant (variant Creutzfeldt-Jakob disease) did occur. How cows became infected with the disease remains unknown, as do the exact contamination mechanism for humans and how the probability of being infected relates to past beef consumption (Adda 2003). As a result, current estimates of human victims in the United Kingdom over the next two decades vary from 100 to more than 100,000 (Blakeslee 2001).

This discussion suggests two points: scientific uncertainty is crucial to the debate over GMOs, and scientific uncertainty involves Knightian uncertainty (ambiguity). Therefore, economic analysis of GMOs should recognize the potential presence of (Knightian) uncertainty. And, it should properly account for decision-makers' attitudes toward uncertainty. Put another way, the proper goal in an uncertain setting is not "... scientific risk-based assessment" but scientific uncertainty-based assessment.

The economic distinction between risk (known or statistically estimable probabilities) and uncertainty (unknownable or unestimable probabilities) dates to Knight (1921) and Keynes (1921). Nevertheless, the terms are frequently used interchangeably. One reason is that the economic importance of the distinction was largely overlooked until Ellsberg's (1961) classic study. Ellsberg (1961) argued that individuals exhibit behavior sensitive to the weight of evidence about probabilities (the famous "urn" examples). Such behavior directly contradicts both objective and subjective expected utility theory, and, if descriptive of reality, renders expected-utility theory inappropriate for evaluating situations involving Knightian uncertainty? Ellsberg-type behavior has been repeatedly validated in the experimental and empirical literatures. (2)

This article examines optimal trade patterns in an uncertain world, that is, one where objective (science-based) or subjective probabilities may not exist. Attitudes toward uncertainty are represented by the Gilboa-Schmeidler (1989) maximin expected-utility (MMEU) model. The MMEU model has two important advantages: It seems the most analytically convenient model capable of explaining Ellsberg-type behavior, and it has subjective expected utility as a trivial special case. Our central result is that in a two-country, general-equilibrium setting with stochastic production, Pareto optimality can require one trading partner to absorb all uncertainty in the economy if its set of priors is a subset of its trading partner's. An immediate corollary is that autarky is Pareto optimal if the trading partner with the more inclusive set of priors either chooses or is endowed with a certain (nonstochastic) technology. Thus, no trade in uncertain products, can be Pareto optimal, a result that rationalizes the most extreme version of the PP.

In what follows, we first construct a general-equilibrium model of trade between two countries with different MMEU utility structures, but common attitudes toward risk. The countries have access to uncertain technologies, that transform their current period inputs into a product of uncertain quality, and the countries are allowed to freely trade this ex ante uncertain commodity in complete markets. After developing our results for the basic model, we present extensions of the model, and then briefly compare our analysis with other economic analyses. The article then concludes.

The Model

To maintain simplicity, we use a single-product, general-equilibrium model with stochastic production. There are two periods. The first period, 0, is certain, and the second, 1, is uncertain. The uncertainty concerns the quality of a single consumption good of fixed quantity and is represented by a neutral player ("Nature") making a draw from [omega] = {1, 2}. (3) Each element of [omega] is referred to as a state of Nature.

There are two potential trading partners, each representing the representative agents from two countries, which are mnemonically referred as the European Union (EU) and the rest-of-the-world (ROW). Each country is endowed with a potentially stochastic production technology that transforms period 0 applications of a nonstochastic input vector, x [member of] [R.sup.N.sub.+], into uncertain period 1 quality of the good. These uncertain production relations are initially modeled by increasing production functions mapping input committed, x, in period 0 and the realized state of nature, s, into an uncertain period 1 quality according to

[z.sub.s] = [f.sup.i](x, s) s = 1, 2, i = E, R.

Here, for example, [f.sup.i](x, 1) > [f.sup.i](x, 2) implies that quality of the good in state 1 is higher than in state 2. In a later section, this technology is generalized. Input endowments and technologies may vary across countries. These differences are denoted notationally by superscript E for the EU and R for the ROW.

Trading arrangements between the EU and the ROW cover two periods. Contracts are complete and enforceable. In period 0, the trading partners agree on the level of trade for both realizations of [omega]. In period 1, these contracts are executed after Nature has made its choice. There is no trade in period 0 inputs.

Of course, in a truly uncertain world, such complete contracts do not exist. This assumption is motivated not by realism, but by the fact that Pareto optimality typically requires either complete markets, or a market structure that is effectively complete. The assumption's role is to ensure that the theoretical results are not driven by the presence of missing or incomplete markets. By the theory of the second best, it is extremely well known that incomplete markets can justify no-trade results.