Modelling the Impact of Trade Liberalisation

A Critique of Computable General Equilibrium Models

Lance Taylor and Rudiger von Arnim
(New School for Social Research, New York)

Download a PDF file (719K) of the full paper

Francais (91K PDF) Espanol (101K PDF)

Summary

Trade models are of tremendous political importance. By pretending to see into the future and putting numbers on the ‘welfare gains’ accruing from trade liberalisation, such models have provided irresistible ammunition to negotiators, particularly those advocating rapid opening. At the Hong Kong WTO ministerial in December 2005, the press conferences of major players such as Robert Portman (USA) and Peter Mandelson (EU) were sprinkled with the findings of trade models, presented as fact. The incessant repetition of benefit numbers in many billions of dollars inevitably puts those advocating a more cautious approach to liberalisation on the defensive, even when such caution is fully warranted (not only by the mixed experience in recent decades of trade liberalisation in practice, but also by the weaknesses of ‘typical’ CGE models, as this paper clearly demonstrates).

The paper presents a review and critique of the most widely used trade models based on computable general equilibrium (or CGE) models. The emphasis throughout is on methodology. The paper provides concise analytical arguments explaining the fundamental weaknesses of CGE models, paying particular attention to the way that CGE models conceptualise and measure welfare. The authors also show that the manner in which the World Bank uses CGE modelling is highly problematic, making implausible assumptions about elasticities, the exchange rate, and macro causality. World Bank models assume that the most central macro-economic indicators do not change in response to any liberalisation scenario. The authors argue that this is negligent, especially in developing countries with historically large trade deficits, significant debt problems, and a large informal economy with underemployment in modern sectors. The authors also identify a particular inconsistency inherent in the use of ‘Armington’ specifications of elasticities in CGE models. They show that, even if the Bank’s welfare measures and macro causal scheme are accepted, the welfare gains that liberalisation is supposed to induce are estimated incorrectly in LINKAGE, GTAP, and other trade models that adopt the popular Armington specification of imperfect competition between trading partners.

Using a table-top two-region and three-sector CGE model which represents sub-Saharan Africa and the rest of the world, the authors illustrate that the results of modelling are radically different when alternative, and more plausible, assumptions are made. Although the authors’ model still incorporates a great deal of LINKAGE’s lack of realism, it provides unambiguous policy insights:

• If trade elasticities are lower than the World Bank stipulates, sub-Saharan Africa faces welfare losses even in an otherwise optimistic situation which rules out all macro-economic shocks.
• If the current account is allowed to respond to trade liberalisation, and imports to one region grow stronger than that region’s exports, it is Africa – and not the developed world – that faces a deteriorating trade balance.
• If the analysis includes government deficits, the African public balance often deteriorates, whereas the rest of the world’s fiscal position improves.
• If employment and income are variable, they might very well increase in sub-Saharan Africa, but do so in tandem with mounting trade deficits and foreign debt, which renders these advances temporary.

The paper concludes that developing countries would be ill-advised to follow the radical recommendations of the World Bank’s liberalisation strategy insofar as it rests on results drawn from the current trade models. The paper appeals for ‘honest’ simulation strategies showing the variety of outcomes that result from a range of plausible assumptions. These would enable policy makers to assess the different scenarios for themselves.

CGE models can be useful quantitative supplements to experimental thinking about the importance of different potential causal linkages among economic variables at the country or world level. However, mechanically churning out ‘projections’ of welfare gains or any other indicator subject to one single set of causal assumptions and parameter values is a fundamental misuse of a sometimes helpful tool.

Date of original publication: July 2006