Massive "computable general equilibrium" (CGE) models, estimating the global benefits of trade liberalization, have become increasingly common fixtures in recent trade negotiations. The models routinely find that the world as a whole would benefit from full liberalization, although the numbers have been rapidly shrinking of late, and there are losers as well as winners in many scenarios.

Do these models actually answer the questions that policymakers are asking? Despite the hard work and impressive research that has gone into the models, and many recent improvements, they remain inadequate to the task. Unrealistic simplifying assumptions, required to complete the portrait of the world economy, turn out to determine the results. Changes to these assumptions can change the size of benefits, and can even flip the impacts on developing countries from positive to negative, or vice versa.

At least four problematical assumptions shape CGE model results. First, the so-called "Armington elasticities" determine how fully and rapidly trade flows respond to price changes (such as tariff reductions). There are rival schools of thought about the size of these elasticities; unlike imports, exports, tariffs, and other data, the elasticities are not observable, but must be indirectly inferred - or in some cases, guessed at. Those who believe in bigger elasticities estimate greater gains from trade liberalization. Regardless of the size of the elasticities, the Armington procedure imposes unrealistic conditions on the models, implying, for example, that every country has a differentiated product and some market power, even in bulk commodity trade.

Second, CGE models usually assume no net change in any country's employment occurs as a result of trade policy. In the real world, hopes for employment growth in successful export industries, and fears of job losses in declining industries, are among the major reasons why anyone cares about trade policy. But modelers have traditionally assumed fixed employment in order to simplify their calculations. Models changing this assumption have just begun to appear - and are obtaining qualitatively different results. Compared with the fixed-employment models, the recent Carnegie model, which allows variable levels of unskilled employment in developing countries, finds that more of the gains come from liberalization of manufacturing trade rather than agriculture, and that a much larger share of the global gains go to China. None of the models, Carnegie included, have yet examined what the assumption of variable employment in all sectors and countries would do to the results.

Third, other simplifying assumptions rule out other important economic effects. The World Bank's model, among others, assumes that every country's budget surplus or deficit is unchanged as a result of trade policy; an unspecified tax will be increased to make up for any tariff losses. The basis for this belief, in an era of intense resistance to tax increases, is hard to understand. India, for example, could gain $2.2 billion from one version of a "likely Doha" trade deal. Yet, within that net figure is hidden a reduction in tariff revenues of $7.9 billion, and an assumption that the government will make up those lost revenues through new or increased taxes. Tariff revenues are vitally important to developing country governments, as they were to today's developed countries in the past. More than half of all US federal government revenue came from tariffs in all but four years from independence through 1910. The US then, and India today, could not effortlessly replace those lost revenues.

Fourth, attention has increasingly turned to dynamic effects of liberalization, anticipated over time following a trade agreement. Perhaps this is because the static effects, immediately following the agreement, are now estimated to be quite small. (The static effects have shrunk due to recent model improvements, reducing past exaggerations, and because the world is noticeably closer to free trade than it was 5-10 years ago.) However, arbitrary and often heroic assumptions about future growth are simply tacked on to models in many cases. In the static analysis, the CGE model framework enforces a consistent accounting on calculations of employment and output. In the dynamic variant, where hypotheses about future growth are developed outside the model, the sky is the limit. In a recent literature review, World Bank modeler Kym Anderson assumed on the basis of a few case studies that trade liberalization would lead to a dramatic increase in worldwide growth rates, particularly in developing countries, that would continue undiminished for 45 years. Not surprisingly, he found the "dynamic" benefits of liberalization would be enormous (Click here for the article).

In light of these problems, how should policymakers interpret CGE model results? The numbers may suggest a qualitative story about some possible effects of trade liberalization; they should not be taken as precise, reliable estimates of what is certain to happen. (Interestingly, many modelers personally believe this, but they have failed to communicate it to their public relations departments and to the media.) Model users are warned that there are many unrealistic assumptions lurking inside, which may account for any surprising or counterintuitive results. Simpler, more transparent models of particular commodity markets may provide a helpful alternative for countries concerned primarily with a few exports or imports. And, on the strength (and weaknesses) of the models, it remains an open question whether the actual benefits available from further liberalization outweigh the loss of tariff revenues and policy space.

Frank Ackerman is the director of the Research and Policy Program at the Global Development and the Environment Institute at Tufts University.