This section describes simulation results from the ATPSM model. Moreover, as said at the outset, one important objective in undertaking this study is to clarify results from various model-based assessments of the impact of cotton subsidies. Therefore, while reporting results from the ATPSM model, we also compare results from other studies. This is done realizing fully that it is always difficult to compare results across models in view of differences in the model structure and various assumptions made. One particular difference is assumption made about China's cotton subsidies. All other models, with the exception of Tokarick/IMF, assume that China subsidises cotton, as per the ICAC data set. We do not make this assumption and so simulate the effects of removing only those subsidies that are officially notified to the WTO, which makes comparison with other results somewhat difficult. We, however, present in Annex 2 all corresponding results from a simulation run that assumes that China also subsidises cotton production as per the ICAC data set. We do this merely to facilitate comparison with other models, which is essential given the importance of "clarifying" the impact of subsidies on the world cotton market.
One of the issues discussed in Section 3, in the context of the building blocks of partial-equilibrium models, was the difficulty or uncertainty of pinning down some of the key data and parameters. The three main parameters or issues discussed there were the level of initial distortions (domestic subsidies in the context of cotton, including the issue of coupled versus decoupled effects on production and trade), supply and demand elasticities and world-to-domestic price transmission. All the tables show results from sensitivity tests for alternative values of demand and supply elasticities. A separate sensitivity test, presented at the end of this section, assesses the impact of higher domestic subsidies in the base period,
Four sensitivity scenarios were simulated by either raising or lowering the two sets of elasticities for all important cotton producing and trading nations (those listed in Annex 1 Table 1). For each scenario the relevant supply elasticities (indicated by S in the column headings) and demand elasticities (indicated by D in the column headings) were multiplied by a common factor. So, for instance the column heading S=1, D=1 indicates the base scenario (elasticity values shown in Annex 1 Table 2). All changes are measured with respect to values of the base period (not the results of the base simulation).
Table 7 summarizes estimated impacts of complete liberalization of domestic subsidies and tariffs on world market prices for various combinations of the values of supply and demand elasticities. The impact on world market price of cotton is 3.1 percent for the base scenario with assumed best-estimate values of the elasticities (shown in Annex 1 Table 2). The highest impact was 4.8 percent under the scenario with highly response production and highly inelastic demand. Given the result for the base scenario, these alternative assessments are as expected.
One immediate question that may be raised is why the overall impact on the world cotton price is on the lower side, relative to those from some other studies for cotton as well as for other agricultural commodities in various model-based studies simulating global agricultural reform process. The answer to the question of relatively small impact on the cotton price follows from relatively low level of the initial distortion, which was estimated to be 10 percent in ad valorem equivalent term - almost all due to domestic subsidies as applied tariff rate on most of the traded cotton was negligible (Table 6). Indeed, the ATPSM results show fairly large impact on world market prices for several other agricultural commodities with high levels of distortions in the base period, including tariff distortions.
Based on the level of distortion in the base period, about 66 percent of the overall impact on world market price can be attributed to the removal of the US subsidy, followed by the removal of the EU subsidy (about 33 percent of the total contribution). Although the EU's rate of subsidization is highest, it plays a much smaller role in world cotton production and trade. For the same reason, the domestic subsidies in Brazil and Colombia play a negligible role. The relative importance of subsidies versus border measures has attracted some analyses in the literature in the context of the WTO agricultural negotiations. The purpose of these studies is to see whether developing countries gain more by focussing on one or the other "pillar" of the AoA in these negotiations. It has been established that for most commodities, there is relatively more to gain from reforms in border measures, but cotton is an exception to this pattern. Our results also confirm those findings - when all countries eliminate border restrictions on cotton trade, the impact on the world market price is only 0.6 percent versus 2.6 percent when all domestic subsidies are removed while border restrictions are maintained.
The results of the sensitivity tests are basically consistent with expectations. Thus, the impact on the world price of cotton is higher when demand is assumed to be inelastic. With demand elasticities at one fourth of their base values and base supply elasticities, the increase in the world price from full liberalization is 4.3 percent (compared to 3.1 percent in base scenario) and 4.8 percent when supply elasticities are three times the base values. The latter would amount to much stronger long term response to price increases. Similarly, world price rises by only 2.3 percent when supply elasticities are halved, keeping demand elasticities the same.
Before we proceed with more detailed analysis of the results, it is useful to compare our results with the price impacts from some other recent model-based studies, although as noted earlier direct comparisons are difficult as most other studies use the ICAC data base that show China granting considerable subsidies to cotton. As mentioned earlier, the ICAC (2002) estimates are important as these became the basis for some headline news that started the "cotton issue" in the WTO. The ICAC study had concluded that the average cotton prices during 2000/01 and 2001/02 would have been US$ 0.17 to US$ 0.31 per pound higher, respectively, if all subsidies were eliminated. This implied impacts on world cotton prices of 30 and 72 percent in 2000/01 and 2001/02, respectively, obviously much higher than the results from the ATPSM. A major limitation with this analysis was that the model ignored inter alia supply responses of other countries' to higher world prices following the elimination of the subsidies.
This limitation was dealt with in a subsequent study by Goreux (2003), which otherwise followed similar approach and used the same subsidy levels as the ICAC study. He also made a number of improvements and reported a wide range of results based on sensitivity tests. The impact on world cotton price in his study ranged between 2.9 to 13.4 percent, depending on the combination of price elasticities of supply and demand assumed. He also noted that the impact of subsidies on the world price was markedly sensitive to these elasticities.
A 2002 study from FAPRI, assessing the impact of the Doha Round reforms, is another important study in this area, and is extensively quoted, including in reports such as the cotton section of the World Bank's 2004 Global Economic Prospects report(World Bank, 2004). The FAPRI study finds that the average cotton price during 2001/02 to 2010/11 period would be 13 percent higher with full reform (both subsidies and tariffs). This study also shows that the bulk of the large world price impact is due to domestic production subsidies, as world price would rise by only 4 percent if only trade distortions are eliminated.
The study by Tokarick (2003) was the basis for the cotton section in IMF's 2002 World Economic Outlook report (IMF 2002). Tokarick's partial equilibrium model estimated the impact on the world market price of cotton to be only 2.8 percent, of which 2 percent came from the removal of production subsidies and 0.8 percent from the removal of market price support. In another study, Quirke (2002) estimated that the removal of production and export subsidies by the United States and the EU would raise the world cotton price by 10.7 percent in 2001/02. His estimates were also based on ICAC policy data base.
Finally, the study by Sumner (2003) is interesting because it provides simulated world price impacts separately for the removal of six types of United States domestic subsidies, which were assumed to have different degrees of "coupling". On average, over the nine years modelled, Sumner finds that removing all the subsidies simultaneously would cause the world price to rise by about 12 percent with a range of between 8 and 18 percent.
In summary, first, the range of the impact on world cotton price assessed by various studies seems to lie between 3 percent to 15 percent. Although it is very difficult to reconcile differences among model results for a variety of reasons (see Sharma et al 1996), including the fact that only few studies spell out in detail all the assumptions made, there is some basis to conclude that our own ATPSM results are not out of line with these results. For example, our estimate of 3.1 percent price impact under the base scenario corresponds with Tokaricks' assessment, which is based only on subsidy removal in the United States. By contrast, assessed impacts from other models are fairly comparable with the ATPSM results based on the ICAC subsidy scenario, which we report in several tables in Annex 2 for the sole purpose of demonstrating this similarity. Thus, the 7 percent price impact the ATPSM ICAC subsidy scenario is very close to some of the scenarios simulated by Goreux, where he assumes supply elasticity of 0.9 and demand elasticity of negative 0.6, similar to the average values of parameters used in the ATPSM. Three of the scenarios with the ATPSM model gives the price impact of about 10 percent, in one case 12 percent, which are comparable to those from some of these studies, e.g. 13 percent of the FAPRI assessment. Sumner's results are closer to the FAPRI estimates. While details about parameters and other assumptions used in the FAPRI study are not available, Sumner claims in his paper that he used to the extent possible similar approach and parameters as the FAPRI study. Sumner provides demand and elasticity values used, citing FAPRI sources (see Annex 1 Table 2 for his values and the ATPSM values). One reason for the larger world price impacts in the Sumner and FAPRI studies (relative to ATPSM) could be the assumed values of supply and demand elasticities, in particular demand elasticities which are relatively much lower compared to the ATPSM. Lastly, all studies agree that the bulk of the impact is due to subsidy removal, and only a small part from eliminating border measures.
One final point worth noting at the end of this sub-section is that the ATPSM model is fairly stable - the world market price did not even double when supply elasticities were tripled and demand elasticities reduced to one-fourth of their original values. Had the results been unstable, this would give less confidence to the results.
As explained earlier, the way most of these models work is that following the removal of subsidies (or tariffs) in country A, production would shrink as per the supply price elasticity used, which in turn reduces net exports. The resulting higher world market price induces production in non-subsidizing countries and their production and trade expands. This process continues until a new equilibrium is reached. Hence, the direction of most simulation results is anticipated - the issue is mainly one of degree, and how the ATPSM results compare with others in the literature. Tables 8 and 9 show estimated impacts on production and exports for selected countries that represent both subsidizing and non-subsidizing major cotton producers and traders.
Before the results are discussed, it is useful to note that, although there is a positive relation between world price impact and the resulting impacts on production and export, the magnitude of this relationship need not be strong. Where supply responses are strong, especially in non-subsidizing countries, the change in the world price could be small but there could be marked impacts on production and trade. By contrast, with inelastic supply responses, the change in the world price could be large, but production and trade effects would be small.
Overall, the simulated effects on production and trade are consistent with expectations. Thus, under full liberalization, production shrinks in countries that reduce subsidies (e.g. EU and US), and in proportion to the base level of subsidy. For example, the 14 and 32 percent reduction in the United States and EU cotton production, respectively, reflect the fact that base subsidy levels were relatively high. At the global level, the 2 percent drop in world production is due to the fact that output in non-subsidizing countries, which rises in response to the increase in the world price, does not rise enough to offset the large declines in production of subsidizing countries. Most countries gain in outputs when subsidizing countries eliminate subsidies, which is expected. Cotton production in the BBCM sub-group (total for Benin, Burkina Faso, Chad and Mali) rises by 2.4 percent.
Other results in the table show that production naturally responds to the size of the supply elasticity assumed, e.g. the impacts are generally 3-4 times higher than in the base case with supply elasticities three times higher. For countries for which initial subsidies were removed, the reduction in production is similarly higher. With higher production where this occurs, exports expand as per the model specification. For instance for the BBCM sub-group, if supply elasticities in all countries are three times the assumed base values (scenario with S=3, D=1), production under full liberalization would expand by 9.9 percent (versus 2.4 percent in the base run), and exports would expand by 14.6 percent versus only 4.1 percent in the base run. But in scenarios with inelastic demand response, consumption would not decline as much when world prices rise as in the base case, and so somewhat reduced cotton would be available for export, relative to production under that scenario.
What do other studies say about production effects? While identical results can not be expected, there are some strong similarities. One of these is that the Goreux model also shows world production of cotton to decline by 4.3 percent. Goreux estimated that, with elasticities higher than his base values and closer to those of the ATPSM, production in the United States would fall by 19 percent, compared with 14 percent in the ATPSM result under the base scenario. While the FAPRI study showed smaller (6.7 percent) negative impact on the United States production, China's production was found to rise by 1.5 percent, closer to ATPSM's 2.8 percent. The production increase for Africa as a whole in the FAPRI study, from elimination of all cotton subsidies, was 6 percent, which is very close to Goreux's 6.7 percent under high elasticity scenario, while the impact was much smaller in the ATPSM results under the base scenario but in the 10-12 percent range under assumptions of higher supply elasticities. In contrast to all these estimates, Sumner finds that the United States production falls by as high as 27 percent on average during 1999/00-2007/08 period (12-47 percent range) under his full subsidy removal scenario.
Reviewing the impact on trade (Table 9) should be relatively simple once the sources of change in production are understood. In the ATPSM, as in real life, trade is strongly influenced by production, as well as by some other factors. Models could differ in terms of this specification but essentially the relationship between production and trade is often strong. This is also evident when Tables 8 and 9 are compared. Overall, global export shrinks by 2 percent.
The impacts on subsidizing and non-subsidizing countries are as expected. Thus, for example, United States exports fall by 14 percent. Note that in each case when a subsidizing country reforms, others benefit in trade terms, which is what is expected. Thus, under full reform, the BBCM subgroup expands cotton exports by 4.1 percent. All other non-subsidising countries also gain.
While the above results are intuitively correct and expected given the nature of the impact on production, the interesting question at this stage is how do these results compare with other studies? The impact on world export of cotton assessed by the FAPRI study (5.8 percent) was closer to the 7 percent gain in the ATPSM model under the ICAC scenario. In general, however, the results from various models on the changes in trade are not as close as those for price and production. For example, the ATPSM showed almost five times larger impact on United States exports from free trade compared with the FAPRI result (3.5 percent fall). The FAPRI analysis showed 13 percent increase in African exports versus only 3-8 percent in the ATPSM for the BBCM subgroup. Goreux does not report changes in export volumes, but he shows markedly positive impact on African export earnings. Finally, Sumner's results on trade for the United States differ the most from those of all the other studies, as is also the case with production. He found that United States exports would fall by 43 percent, with a 28-51 percent range for various years, when all US subsidies were eliminated, and this is 3-4 times the changes predicted by the ATPSM.
Table 10 shows impacts on key welfare indicators for selected countries. The indicators are changes in producer surplus (PS) and total welfare (TS) as the sum of the PS and two other indicators not shown in the table, namely changes in consumer surplus (CS) and changes in government revenue (positive for subsidizing countries denote savings). The results in the table are largely self-explanatory, as these follow from changes in world market prices and production discussed earlier. The numbers in the third column (changes in government revenue) are also easy to explain with reference to Table 6 on policy parameters. For the three countries with tariffs (China, Colombia and the US), a small part of the gain from liberalization is the loss of tariff revenue, but overall for these as well as other two (Brazil and the EU) most of the gains are from the elimination of domestic subsidies. For all other countries (including those not shown in the table), there are no changes in government revenues.
The main focus of this paper is on the impact of cotton subsidies on non-subsidizing countries, and hence the key indicator is producer surplus (and export earnings discussed below). The table shows that under full liberalization, the BBCM subgroup would experience PS gains of $21 million per year and TS gain of $15 million, after deducting the CS loss due to higher price of cotton. Farmers in all non-subsidizing countries gain (the PS gain), and substantial amounts are gained by cotton farmers in China, India and Pakistan, reflecting large production in the base period. However, the overall gains (TS) for these countries are much smaller because of substantive CS losses, which is explained by the fact that, unlike the BBCM case, domestic utilization of cotton is very large in these countries.
As domestic subsidies are eliminated, farmers lose in the two countries that provide high levels of subsidies (i.e. the EU and US). These countries also experience CS losses as cotton price rises. But interestingly, they gain in total welfare terms because the savings in domestic subsidies more than offset the losses in PS and CS. It is for this reason that only 4 of the 11 countries/groups shown in the table lose in total welfare terms. Overall, the gains from the reform process at the global level are substantially positive (about $900 million). By symmetry, these calculations also highlight the global and country specific losses from the current cotton policy regimes. These suggest that the gain in net government revenue could be used, under a liberalizing scenario, to compensate in a decoupled way the loss of farm income. The issue, of course, would be to institute compensatory policies that are indeed decoupled, but as yet economic analysis has not clarified how decoupled the various measures introduced in the past by some subsidizing countries have been. The issue clearly merits further research.
Although export earnings and import expenditures are not separate gains, namely they are not additional to gains in welfare terms, it is useful to report these statistics also in view of the widespread interest on export earnings and trade revenues. Impact on export in volume terms was already discussed earlier, and so this presentation can be short. Table 11 shows changes in export earnings. These are the sum of two components:
Additional export earnings or import expenditures on base period exports or imports, namely base export (or import) volume times (new world price - base world price); and
Total export earnings (or import expenditures) from additional exports (imports) following the reform process, namely (new export (import) volume-base export (import) volume) times new world market price.
The numbers in the table are self-explanatory and require little discussion, especially in view of the discussions earlier. For example, the gain in export earnings for the BBCM subgroup is estimated at $35 million per annum in the base scenario, but much higher - about $100 million - in scenarios with high supply elasticities. These numbers also indicate the loss in export earnings that these countries incur due to the policy regime of the base period.
The results from sensitivity tests show that if elasticities of supply are three times larger than in the base, the BBCM sub-group would obtain, under full liberalization, export earnings that are larger by about $100 million or 20 percent higher than in the base (or in other words the loss of export earnings they incur because of the subsidies is $100 million yearly rather than the $35 million in the base scenario). It can be seen that the changes in the export earnings under various elasticity assumptions are considerably different than those in the base run.
Although the levels of subsidies used for the base period in this paper are all official as notified to the WTO, an additional sensitivity test is conducted assuming a higher level of US subsidies on cotton. This test could have been done for any other combination of countries, but the US application has an added value in that recent cotton studies seem to focus on this aspect. Also the scenarios are useful in that the level of US subsidies that is distorting is a matter of contention, and as we saw earlier, different analysts consider various parts of the subsidies to have different degrees of coupling. Table 12 shows the results for full liberalization by all countries with the US subsidy assumed to be $2.5 billion, instead of $1.6 billion in the base scenario, keeping all other parameters and assumptions the same.
First, note the impact on world prices. If the US subsidy were $2.5 billion initially, the world cotton price would rise by 3.7 percent instead of 3.1 percent in the base scenario. This follows from an increase in the overall total distortion in the world cotton market for the base period to 14 percent in ad valorem equivalent term, compared with 10 percent in the base scenario. Looking at this level of overall distortion in relation to those for many other commodities in the ATPSM, it appears low, and hence the 3.7 percent impact is relatively low. The reason why the overall level of world distortion does not increase significantly when US subsidies are doubled or trebled is due to the fact that the United States accounts for only 20 percent of world cotton output. This is also the reason that world price is not expected to rise too much if the United States eliminates its subsidies.
Other results in the table follow mostly from this change in the world market price since other parameters are not changed. Thus, for example, United States cotton production is reduced by 730 000 tonnes (20 percent less). The reductions in export volumes and export earnings follow from this. Interestingly, while other countries increase production in response to higher prices, these are not adequate enough to offset the sharp decline in United States production, and so world output is lower. Clearly, as stressed throughout this section, the results depend a lot on the values of the elasticities assumed. For instance if all elasticities were increased by a factor of 3, it is clear that the reduction in US production and exports would be much larger under the new subsidy scenario, and the exports of all other competing countries would correspondingly be much larger.
 For example, the ATPSM
results show the following impacts on world market prices under full
liberalization: 16 percent for wheat; 14 percent for sugar; 30 percent for
butter; 21 percent for cheese and 23 percent for milk powder, etc.|
 See for example Hoeckman, Ng and Olarreaga (2002) and Rae and Strut (2003).
 Mr Goreux was indeed hired as a consultant following a decision of the Agricultural Ministerial Meeting of Western and Central Africa (AOC) in June 2002 and in the context of the submission of the cotton issue to the WTO by Benin, Burkina Faso, Chad and Mali.
 For countries without domestic subsidies, e.g. Turkey, the small negative effect is the result of the removal of border restrictions.
 These similarities become very apparent when Goreux's and other estimates are compared with the ICAC scenario of the ATPSM run (Annex 2). That the assumption about Chinese cotton subsidies in these other studies explains bulk of the differences is also obvious.
 By contrast, global trade expands by 6.7 percent under the ICAC subsidy scenario. The main reason seems to be that in this scenario China's production is reduced considerably, and this requires large imports to satisfy China's large domestic demand. This is what leads to the expansion of world exports. This, on the other hand, is not the case under the WTO scenario with no cotton subsidies in China.
 See section 4 for definitions of these measures.
 For exporting countries that do not have domestic support, their excess supply function does not change after liberalization, and hence the sum of the above two components is equal to the difference between the simulated and the base values of export earnings. For countries that experience a leftward shift in their excess supply functions, the same holds only if exports expand post-liberalization. Otherwise there is a small second order term difference between the sum of the two components and the difference between the simulated value of export earnings value and the base value. By symmetry the same holds for imports. In any case, since tariffs are not very important in cotton trade the import demand functions do not shift much post-liberalization.