3 MODELLING THE IMPACT OF COTTON SUBSIDIES: THE KEY EMPIRICAL ISSUES

3.1 The level of domestic subsidies on cotton

It is widely held that domestic subsidies are the main source of distortion for cotton, which is unlike the case with many other farm commodities where distortions at the border are also widespread. Therefore, it is important here to review the level and nature of domestic subsidies to cotton. The two main sources of information on domestic subsidies used by various analysts are official figures notified to the WTO and estimates put together by the ICAC. Tables 1 and 2 indicate this data, which show that there are some significant differences in these estimates, which are discussed below.

Note that from the standpoint of both the level of the subsidy and the size of the industry, what matters most for world market effects is subsidies in China and the United States, the two largest producing and trading countries. According to the ICAC data (Table 1), the global total subsidies to cotton averaged US$4.5 billion per year during 1997-02, with a range of US$3.8-5.3 billion. About 75 percent of the total is accounted for by China and the United States - with the share of the United States in the total rising in recent years, from 30 percent during 1997-99 to 47 percent in 2000-02, while China's share fell considerably. The share of the EU subsidies has remained more or less the same, at about 18 percent. The rest, amounting to about 6 percent, is accounted for by Brazil, Egypt, Mexico and Turkey.

The notifications to the WTO are perhaps the most official publicly available sources of information on domestic subsidies. These notifications, however, are not available for most recent years, as shown in Table 2. For example, the latest notification on domestic support measures for both the EU and the United States was for the year 1999, posted in the WTO website only in June 2002 for the EU and February 2003 for the United States. By contrast, more recent information is available from the ICAC, although these are not necessarily official figures. Since this is the information commonly used in debates on the issue and in some models, it is important to note the differences in the level of subsidies from these two sources (Tables 1 and 2).

As regards key differences in the two tables, first concerning countries other than China, EU and the United States, the ICAC data show some subsidies for all four countries reported in Table 1, while these are mostly missing in the WTO data. The differences are striking in some cases. For instance, according to ICAC data, Turkey exhibits considerable subsidies all throughout the period shown, while the WTO data indicate that there are no subsidies. Brazil's notified Aggregate Measure of Support (AMS) levels for cotton vary considerably - US$13 million in 1995/96, US$10.3 million in 1996/97 (both not shown in the table) and US$55.3 million in 1997/98, for an average of about US$32 million for these three years. For 1997/98, about 80 percent of the outlay was for production and marketing credit and the rest was market price support. Egypt has not notified to the WTO any support to cotton, as part of AMS or in other form. Mexico notified about $1.7 million AMS for 1996 and nothing for 1997 and 1998. The ICAC did not show Colombia as having granted cotton subsidies, but WTO notifications show this. The AMS levels for Colombia were US$1.1 million in 1997, less than US$1 million in 1998 and US$3.5 million in 1999, for an average of US$1.8 million. Subsidies were given in the form of both market price support and direct payments. Finally, the WTO notifications do not show any cotton subsidies by both India and Pakistan, two important countries for cotton, which some analysts have reported as having granted cotton subsidies in most recent years.

In the case of the EU, although there are some significant differences in the two sources, the EU cotton subsidies are fairly transparent. The levels reported in the WTO notifications (see Table 3 for details of the computations) are lower than in the ICAC. This could be for the reason that the ICAC estimates also include support outlays other than AMS, which could be part of some Green Box outlays and some non product-specific AMS.

Analysts studying the cotton issue face the greatest difficulty in estimating domestic subsidies for China. The differences between the ICAC estimates and official sources, including the notifications to the WTO are immense. In its accession document, China reported negative domestic subsidies (AMS) on cotton, about negative US$600 million for the base period 1996-98^[2]. China has a de minimis limit of 8.5 percent of the value of cotton output, and so can grant AMS subsidies up to this level (about US$700 million) as and when China desires, without breaching its WTO commitment. So far, there is no information on even this support (China has yet to submit its first WTO notification on support). Experts generally agree that given China's policies, and until substantive changes were made in recent years, the cotton sector most likely received substantive subsidies, but much less so in most recent years. Huang, Rozelle and Chang (2003) estimated that during 2001 the nominal rate of protection (NRP) for cotton averaged 17 percent, while Fang and Beghin (2003) estimated that for the 1997-2000 period the same NRP averaged -20 percent, implying taxation of the sector. But these different estimates could have resulted from differences in assumptions about border measures, and not necessarily through assumptions about domestic subsidies. These conflicting estimates lend some support to the position that there may be little or no effective domestic subsidy for cotton in China, which is also the official position (see also Ke 2003).

Overall, far fewer countries seem to grant subsidies on cotton, according to official information provided to the WTO. It should, however, be noted that the WTO notifications are not as updated as the ICAC estimates, and indeed more countries may have granted cotton subsidies (and in larger amounts) in recent years when world cotton prices fell, a claim made by ICAC, as well as other commentators. A second point worth noting here is that some countries may not have reported to the WTO the de minimis levels of subsidies as these are exempted from any discipline, whereas these subsidies could play a significant role in affecting production, and hence should feature in any modelling and economic analysis. In this sense, one would expect that the ICAC figures not only indicate higher levels of subsidies relative to those from the WTO^[3], but also that they are more representative of the true level of overall support, with the exception for China for which there are sharp differences.

3.2 Relationship between subsidies and incentives: The issue of coupled versus decoupled subsidies

Subsidies tend to affect production and hence trade. However, different forms of subsidies have potentially different impact on production (OECD 2000). Whenever a subsidy affects directly the total returns per unit produced, then it acts as if the price received by the producer is increased, and its effect is no different than if the market price was higher. In this sense such forms of subsidy are "coupled", as they affect directly the resources allocated to production. On the other hand there are some types of interventions that affect production cost and returns only indirectly, and some times not at all. For instance programs that directly affect farm income, such as payments for residing in a given locality, without being dependent on product specific production, tend to have lower impact on production of specific products. Such payments are considered "decoupled", and as they do not affect production are not market and trade distorting. The problem in any empirical analysis is that it is quite difficult to determine the degree to which some programs have indirect impacts on production incentives^[4].

This issue applies mainly to the United States since subsidies in most other country cases are acknowledged to be coupled, i.e. similar to market price support. For reasons discussed below, it is not as easy to estimate the level of domestic subsidy that can be used in model-based analysis in the case of the United States (or other countries with a mix of various forms of subsidies). This is mainly because there are several different programmes in place not all of which could have the same impact on production and trade, i.e. programmes could differ in terms of the degree of coupled effects, and so these may not be simply summed to obtain the total subsidy. Table 4 shows United States domestic support to cotton as compiled by Baffes (2003) while Table 5 shows subsidies notified by the United States to the WTO.

Note that the estimates presented in Table 4 and 5 are somewhat different. This is because Table 5 includes only AMS measures while Table 4 seems to have included some other measures in some cases, e.g. this table includes the Production Flexibility Contract or PFC (i.e. direct income payments that are placed under Green Box measures in the WTO notifications) for the United States.

Table 4 shows that during the 1996/97 season -the first year of the 1996 Farm Bill -support to United States cotton growers amounted $858 million, almost $700 million from the PFC contract payments and the rest from insurance subsidy. In 1997/98 the support was $745 million. When prices began declining, the emergency assistance measures were introduced, increasing the support to $1.8 billion in 1998/99, $3.2 billion in 1999/2000, $2.1 billion in 2000/01, and $3.7 billion during the 2001/02 season. According to Baffes, if cotton prices remain at their 2001/02 levels, then United States annual support to its cotton sector is expected to be in the order of $3.5 to $4.0 billion for the next six years, the period of the 2002 Farm Act, implying that the United States cotton producers will be receiving support close to the equivalent of twice the world market price.

Concerning the nature of support as far as coupling is concerned, there has been some experimentation underway in the University of Missouri's FAPRI model, towards developing supply response systems that take into account the degree of decoupling of various subsidy programmes. The approach tried is roughly as follows^[5]. For each region of the United States, mixed estimation methods were used to come up with total area elasticities, where total area devoted to major crops (area planted for grains, oilseeds, and cotton) is a function of a weighted average of expected net returns from the market and the loan program plus 25 percent of "less coupled" payments (production flexibility contract payments and market loss assistance payments for the 1997-2001 period, and counter-cyclical payments (CCPs) and direct payments (DPs) for the projection period). These total area elasticities with respect to expected net returns are generally small, and the weighted average for the United States is only 0.06. A matrix of own- and cross-effects is also constructed consistent with the estimated total area elasticity. While the parameters are synthetic, the estimations are done systematically, imposing symmetry, etc.

Less coupled payments (LCPs) come into play in two ways. First, they have a non-commodity specific effect on total area. Since the total area elasticities are small and the direct and CCP payments are multiplied by 0.25, this is a very small effect. Second, 25 percent of the CCP is also included in the expected net return for individual commodities. The logic is that CCPs have a commodity-specific price risk reduction effect, and the 2002 Farm Act updating of programme bases and yields for CCPs may mean they have been more effective than a more purely decoupled payment. In total, then, $1 of DPs has 25 percent of the effect on production as would a $1 of market returns, and $1 of CCPs has 50 percent of the effect (25 percent crop-specific, 25 percent non-crop specific). The contribution of various payments to total net returns obviously depends on market prices. For example, with a market price of 60 cents/pound, the LCPs are likely to be near zero and CCPs will be less than their maximum levels^[6]. Notice that the degree of coupling of the various programs is in essence assumed, and depends on the interpretation of the various programs by the analysts.

Sumner (2003) is another study that treats various United States cotton subsidies differently and assesses their separate impacts on price, production etc. In his model, planted area is determined by expected net revenue times the linear supply coefficient. The expected net revenue per acre is defined as follows:

Expected Net Revenue = Expected [(Market Price*Yield) + (MLB*Yield) + (b_pfcPFC+ b_dpDP)) + (b_mlaMLA+ b_ccpCCP) + CIS - Cost per acre]

where, besides market price and yield, MLB is marketing loan benefits (which includes both loan deficiency payments (LDP) and marketing loan gains); PFC is production flexibility contract payments (which applied during the period 1999-2001) and direct payments (DP) which apply during 2002 to 2007; MLA is market loss assistance payments (which applied during the period 1999-2001) and counter-cyclical payments (CCP) (which apply during 2002 to 2007); and CIS is the crop insurance subsidy.

A reduction in the expected amount of any of the four production subsidies affects planted acres and hence United States cotton production through the impact on expected net revenue per acre^[7]. The various b_i coefficients are intended to measure the impact on cotton net returns per acre of a given form of subsidy, relative to the impact of a simple market price change, and as such they measure the degree of coupling of the various payment types. A value b=0 would imply that a particular payment has no impact on market returns, and hence is fully decoupled, while a value of b equal to 1 signifies that this type of payment is fully reflected in producer per acre returns, and hence is fully coupled. For instance, b_pfc and b_dp measure the impacts on net cotton return revenue per acre of PFC payments and DP relative to the impacts of market price changes. The same holds for the other forms of subsidies, namely marketing loan benefit and crop insurance subsidy. The author makes several arguments and concludes that 0< b_pfc<b_dp<1.0. He further notes, and rightly so, that there is no conclusive evidence for specifying the magnitudes of b_pfc and b_dp precisely, as no comprehensive statistical evidence has been produced - part of the problem being that there is little time-series data available for an econometric analysis. In other words one has to essentially assume the values of the b's. The paper discusses in detail the contributions of these payments to the per-acre net revenue and the magnitude of these coefficients.^[8]

Sumner states that for the PFC impacts, a value of b_pfc between 0.15 and 0.4 seems appropriate, considering various channels of influence discussed in his paper, but uses the lower value of b_pfc = 0.15 for his simulations. For reasons discussed in the paper, the impact of direct payments on expected net revenue is assumed to be larger than that on the PFC payments. A range of 0.25 to 0.5 was considered appropriate, but he uses the lower bound value of b_dp = 0.25. The MLA payments are assumed to have larger production incentive than PFC payments and DPs (the MLA payments were notified to the WTO as Amber Box payments), but Sumner - to be on the conservative side - assumes a value of b_mla = 0.25. He also assumed a value of b_ccp = 0.40 for CCP payments although these were deemed to be almost as trade-distorting as loan payments. No adjustment was made for crop insurance (CIS), thus assuming that these payments are fully coupled.

In summary, while a considerable number of theoretical and analytical studies are available on the extent of production and trade distortive effects of various forms of subsidies, there are very few studies that have actually measured the coefficients in a manner that global trade models can use. In large part this is because there is not enough time-series data to measure the coefficients econometrically since such programmes are of a fairly recent origin (e.g. only since 1996 in the United States, and more recently in the EU). The FAPRI and Sumner reviews show that modellers would need to essentially assume certain values for the degree of coupling for some time to come. What is important to note is that this matter cannot be ignored, namely if there are some payments that are only partially decoupled, then it would not be accurate to assume that the entire $6 billion or so of cotton subsidies as estimated by the ICAC can be treated as fully trade-distorting. In models like the ATPSM utilized here, that do not allow the treatment of various forms of subsidies differently, it would seem preferable to use somewhat lower levels than the total subsidy to reflect the fact that some of the payments are less than fully coupled^[9]. The way this ambiguity about the degree of coupling can be resolved in empirical analyses, is via sensitivity analysis of the results, and in the ATPSM application in this paper (next section), sensitivity tests have been done to reflect this aspect.

3.3 Magnitudes of price elasticities of supply and demand

In the comparative statics framework presented above, the extent of the decline in cotton production in a subsidizing country and increased production in non-subsidizing countries depends on the price elasticity of supply. Similarly the extent to which prices are changed in response to policy changes depend on the assumed price elasticities of demand. The problem at hand is one of accurately pinning down these values, as empirical studies often utilize different values and various models assume different parameters. What follows is a brief commentary on these parameters, and the values assumed for the ATPSM model used for simulations in this paper.

Before we review the elasticity values utilized in previous analyses it is helpful to discuss the nature of these elasticities. In other words should one think of them as short or long run elasticities? As the objective of most analyses is to estimate the longer term impacts of any policy reforms, the correct way to think of any elasticities is in long run form, namely after producers and consumers have adjusted fully to a given policy change. Normally such adjustments take some time, and in fact it is often the case that several years are required before adjustments to any permanent price signals are fully reflected in production decisions. The exact dynamic pattern of adjustment is not the concern in this or other similar studies, but rather the overall longer term impact. This aspect of the analysis must be emphasized, and must be distinguished from the shorter term responses to any particular yearly price shock. As the world and domestic prices for cotton (and many other agricultural commodities) are highly variable from year to year, it is not always easy to discern the longer term, or more "permanent" patterns of price changes. In fact, the longer term trends, for instance in world prices, are invariably much smaller than the short term price changes, but it is those that are important from an adjustment perspective.

The ICAC study assumed - for all countries - a price elasticity of cotton supply of 0.47 and a price elasticity of demand equal to -0.1. There are two fundamental problems with these assumptions. First, their model ignores supply responses of other countries and thus overestimates the impact on the world price of the subsidy removal. Second, the demand elasticity used seems to be too low (see below) which also overestimates the price impact.

One argument often made in favour of low price elasticity of demand for cotton is that raw materials accounted for less than 10 percent of cotton clothing. However, the demand for cotton is from cotton mills, where cotton accounts for nearly 70 percent of total production cost, which means that changes in cotton prices should have considerable effect on the mill consumption decision. In fact it is the mills that decide the mix of fibres to use for producing yarn, and this mix is affected by the relative prices of cotton and other fibres. Given the high degree of substitution between cotton and other fibres, demand elasticity must be larger.

In the model used by Goreux (2003), similar elasticities as the ICAC study (i.e. 0.5 and -0.1) were used for the base case but the importance of this assumption was very much noted and therefore the author presents results for alternative values of the parameters (i.e. sensitivity analyses), with supply elasticities in the range of 0.15 to 0.9 and demand elasticities in the -0.05 to -0.6 range. In view of the importance of these parameters, Goreux reports the full range of results corresponding to these assumptions. We also discuss these results later for comparative purposes.

Goreux rightly complains about the lack of reliable estimated elasticities of supply response to world prices for African countries, his main focus of the study. Estimation of these parameters is difficult because the degree of transmission of world to domestic producer prices in Africa varies from year to year. Goreux notes in a footnote that in the CFA countries production was not very correlated to the main world cotton reference price, namely the Cotlook Index A. He also makes the important point that supply price elasticity in general would not remain constant for large price changes, as, for instance, production would respond much more strongly when prices fall substantially below production cost.^[10] For lack of reliable estimates, he assumed for the model runs the same elasticity values for all countries, but reported a range of sensitivity tests to offset this limitation.

The assumption that all countries have the same supply and demand elasticities, as made by several models, is obviously questionable, in view of observed price-production trends. For example, the 2002/03 cotton plantation area in China surged 26 percent in response to a 20 percent increase in domestic and world cotton prices during the planting season, while many other cotton producing countries expanded their cotton areas by very little^[11]. In some major producing countries, notably China, India and Pakistan, cotton planting area accounts for only 2-4 percent of total agricultural land, which permits production expansion significantly with small changes in relative prices (high supply response). Also, the domination of small size farms in cotton production in these countries allows them to respond to any price movement in a massive way. In particular, if prices of other agricultural crops remain unchanged, a significant increase in cotton price would induce significant shift of land from other crops to cotton, and swiftly so (i.e. even in the short-run).

In summary, it seems that values for long term supply and demand price elasticities are larger than assumed in the above reviewed, and also some other models. Some studies also attest to this. For instance, price elasticities of demand in the United States estimated by Shui et al. (1993) were -0.64 in the short-run and -1.27 in the long-run. Monke and Taylor (1985) found the long-run own price supply elasticity for cotton in the United States to be as high as 2.36. The model used by Tokarick (2003) for the IMF study used trade elasticities rather than domestic supply and demand elasticities, with import elasticity of demand being -0.75 and export elasticity of 1.5, for all countries.

For the ATPSM model used in this paper, the values of the price elasticities of supply and demand for all major countries were thoroughly rechecked for correctness, using literature as the basis, as well as recent trends in price-production relationships and policy regimes.

3.4 The world to farm transmission of cotton prices

All global trade models have specifications to transmit or pass-through changes in world market prices to domestic farm and consumer levels. Where there is no transmission, there is no impact of the simulated change in world price on domestic markets, while full pass-through means that domestic prices change as much as the change in the world prices. Many studies have tried to quantify the extent of the domestic price changes in response to international price changes. The normal way of thinking about this matter is in terms of price transmission elasticity, defined as the percentage by which the domestic price changes in response to a one percent world price change. The estimates vary widely, from full transmission (elasticity equal to one) to no or little transmission (elasticity close to or equal to zero). Transmissions all the way to the farm level are often found to be less than complete, and for a number of valid reasons, notably trade and domestic policies, market structure and measurement errors. For reasons of trade and domestic policies, transmissions have often been found to be fairly weak for basic food commodities where government intervention is high and rather strong for other commodities like tropical beverages and agricultural raw materials (Sharma 2002, Rapsomanikis et al. 2003).

The ATPSM model used for this study assumes full transmission for cotton. This also has been the assumption made in the majority of cotton studies using global trade models. The assumption of full transmission for cotton in this study appears reasonable, at least for most countries, as review of policy parameters shows that there is little border distortion, that could account for imperfect transmission. Moreover, this assumption is reasonable for a study that models long-term outcomes (static equilibrium). Even in countries where there may be some blockages of transmission in the short run, the trend is towards less and less price interventions, including the move towards decoupled forms of support. At least this is the declared policy of all countries, as well as in the countries currently granting high levels of domestic subsidies. Where price transmissions are not complete, the implication is that models that assume full transmission overestimate the impacts. In the case of cotton, the chance that this is the case is small.