2.1 The United States COOL regulation and earlier phases of the WTO dispute

The US country of origin labeling legislation was introduced via the 2002 Farm Bill, which required that several agriculture goods, including beef, pork, nuts, and shellfish, be labeled with their source country. Inclusion of the COOL requirement in the 2002 Farm Bill was a legislative victory for the segment of US farming groups – as well as consumer rights groups – that had advocated for such labeling. The statute established a three-tier ‘born, raised, and processed’ system of identifying the source of beef and pork. Only animals that were born, exclusively raised, and slaughtered in the United States were eligible for the US origin label.

However, the 2002 COOL legislative victory was undermined when opponents of the legislation were able to defund the implementation process, delaying implementation for six years. The statute did not specify who would collect the country-of-origin information, how the label would be structured, what types of goods would be subject to the labeling, and how the system would be monitored and enforced. These issues were left to the US Department of Agriculture's (USDA) rulemaking process. At the very end of the Bush administration's term in 2008, the USDA published the first version of the COOL regulations. They permitted a mixed-origin label and disappointed COOL proponents. Even COOL opponents concurred that the rule was the most relaxed regulation possible.

With the arrival of the Obama administration in 2009, regulators became more receptive in their attempts to implement the spirit of the original COOL legislation. The new head of the USDA, Tom Vilsack, on implementing the rule, issued a non-binding letter stating that meatpackers should not use the mixed-origin label if the slaughterhouse processed only US-born, -raised, and -slaughtered meat in one 24-hour period. The letter also noted that the mixed-origin label was not intended to apply to the majority of products eligible for the US-born, -raised, and -slaughtered label (see Greene, Reference Greene2015: 5).

Shortly before the final COOL rule was implemented, the Canadian and Mexican governments filed a request for consultations at the WTO in December 2008 and June 2009 respectively, and both countries requested a panel in October 2009. The subsequent Panel Report and Appellate Body Report found that the COOL regulation breached the TBT agreement by offering less favorable treatment to foreign goods (Howse and Levy, Reference Howse and Levy2013; Mavroidis and Saggi, Reference Mavroidis and Saggi2014). In July 2012, the DSB adopted the Appellate Body's Report.

The WTO decisions required the US government to revisit the COOL rule. Instead of simply repealing it, and to address the concerns raised by the decisions, the United States issued a new and stricter COOL regulation and notified the DSB in May 2013 that it was ‘in compliance’. The option of the mixed-origin label was eliminated, thus providing more information to consumers as an attempt to address the appellate body's concern that information gathered upstream was not being communicated to consumers, and thus that the costs were not justified in terms of greater consumer information. The new labeling scheme identified where the animal had been born, raised, and slaughtered.

Canada and Mexico disagreed that the US's new COOL regulation was consistent with WTO law and requested a compliance panel in August 2013. In October 2014, the compliance Panel found that the US policy was still WTO-inconsistent. The United States again appealed, and lost in an Appellate Body decision issued in May 2015. Canada and Mexico announced their intention to suspend concessions for the United States and, in June 2015, the latter requested an Arbitration Panel to adjudicate the appropriate level of retaliation.

The Arbitration Panel issued its decision on 7 December 2015, and the DSB adopted the report on 21 December. The United States ultimately repealed the COOL regulation through an omnibus bill passed by Congress on 18 December and signed into law by President Obama the same day. Thus, the US government repealed COOL before the DSB officially authorized retaliation on 21 December 2015.

2.2 The United States’ import market for livestock

The North American livestock market has become increasingly integrated since the implementation of first the Canada–US Free Trade Agreement (CUSFTA) in 1988 and the North American Free Trade Agreement (NAFTA) in 1994. As bilaterally applied tariffs on livestock (cattle and hogs) and other non-tariff barriers have fallen, there have been changes to the patterns in live animal shipments across borders in the more integrated North American beef and pork supply chains.Footnote ³ Firms and industries have reorganized regionally across North America, and trade in ‘new’ animal products – e.g., livestock of different ages – has emerged to exploit differences in comparative advantage, economies of scale, and tastes and preferences. Overwhelmingly, trade in live animals has tended to move cattle and hogs from Canada and cattle from Mexico to the US market for processing into beef and pork products for consumption.

Figure 1 illustrates the US import market for cattle and hogs from 1989 to 2015, based on public data provided by the US government. Note that, in nominal terms, the combined value of these US imports rose from $800 million in 1989 to $2.5 billion by 2007. In August 2008, the US–COOL regulation was implemented via an interim rule and then in January 2009 with a final rule. By 2009, US imports had fallen to about $1.7 billion.

Source: Constructed by the authors with data from the USITC's Dataweb.

Figure 1. US import values of products subject to the US–COOL dispute, 1989–2015

Shortly thereafter Canada and Mexico filed WTO disputes, and the COOL regulation remained in place through 2015. In 2015, Canada requested $2.4 billionFootnote ⁴ (CAN$3.1 billion) in retaliatory compensation and Mexico $713 million. The arbitrators granted Canada $805 million (CAN$1.054 billion) and Mexico $227.8 million, or a combined $1 billion in compensatory retaliation. Again, for perspective, prior to the COOL regulation, annual US imports of cattle and swine from Canada and Mexico were never more than $2.5 billion. In 2014, US imports reached $2.9 billion – despite the COOL regulation.

Next, consider Figure 2, which illustrates the US import volumes of cattle and hogs from each of its two major foreign sources. The United States imports substantial volumes of cattle from both Canada and Mexico. The total volume of cattle imports is relatively flat since at least in the mid-1990s, averaging around 2.1 million head per year, although this average masks considerable fluctuations in bilateral volumes on a year-to-year basis.

Source: Constructed by the authors with data from the USITC Dataweb.

Figure 2. US import volumes of cattle by source, 1989–2015

There are many apparent sources for the annual volatility in the bilateral volumes, some of which are associated with major shocks that have nothing to do with the COOL regulation. For example, in 1995 Mexico's exports plummeted in response to the peso crisis. In 2003, Canada suffered an outbreak of bovine spongiform encephalopathy (BSE) that resulted in a US import ban on the country's cattle until 2005.Footnote ⁵ In 1998, the United States initiated antidumping and countervailing duty investigations against Canada and an antidumping case against Mexico, and it is not uncommon for such cases to have a chilling effect on bilateral trade volumes.Footnote ⁶ In other years during this period, Mexico experienced droughts and Canada shut down domestic slaughterhouses, both of which exogenously increased the supply of exports of live animals to the US market. Furthermore, this figure also does not capture likely changes in the types of cattle (of different ages) being traded as the North American industry was developing into a relatively more efficient regional supply chain. Overall, the main point to keep in mind is that several factors shifted US import demand as well as Canada's and Mexico's export supplies for cattle during this period.

Figure 2 also shows why the timing of the imposition of the US interim and final COOL regulations was important. Beginning in summer 2008, US trade flows in general were in a free fall – indeed, the Great Recession led to a simultaneous collapse in trade for virtually all countries across almost all goods. Cattle and beef were no exception. Furthermore, the value of the Canadian dollar increased substantially relative to the US dollar, making it less attractive for Canadian livestock producers to export to the US market. In the end, the coincidence of the imposition of the COOL regulation and the Great Recession makes it difficult to separate a potential decrease in 2008–2009 trade volumes due to COOL from unrelated events associated with the macroeconomic shock of the Great Recession. Indeed, Figure 2 shows that while US imports of cattle from Canada were falling in 2009–2011 from their 2008 peak, US imports of cattle from Mexico were increasing.

Figure 3, illustrating the US import market for swine, shows a somewhat different story. The first distinction worth noting is that virtually all US imports of swine during the period were from Canada;Footnote ⁷ there were no imports from Mexico. Second, integration of the North American market in the 1990s led to a general increase in US import volumes from Canada until the trade collapse associated with the Great Recession in 2008.Footnote ⁸ US imports declined from a peak of 10 million head in 2007 to 6.3 million in 2009; again, the Great Recession's timing coincided with the implementation of COOL. Since then, volumes have stabilized at slightly more than half their peak levels.

Source: Constructed by the authors with data from the USITC Dataweb. The United States did not import swine from Mexico during this period and had sporadic imports of swine – at very small volumes – from other countries, mostly in Europe, during this period.

Figure 3. US import volumes of swine by source, 1989–2015

3.1 A model-based formulation of reciprocity to limit retaliation

To evaluate the arbitration approaches proposed by Canada and Mexico, as well as that adopted by the arbitrators, we introduce a very simple model (Bown and Ruta, Reference Bown, Ruta, Bown and Pauwelyn2010) that illustrates the retaliation limit implied by the Bagwell and Staiger (Reference Bagwell and Staiger2001) mathematical formulation of reciprocity.Footnote ⁹ The Bown–Ruta model was previously used to evaluate how close earlier Article 22.6 arbitrators were in relying on the Bagwell–Staiger formulation in the first ten disputes that reached that phase of the DSU process.Footnote ¹⁰

For ease of exposition, we limit the discussion to two countries: the United States (respondent) and Canada (complainant).Footnote ¹¹ Figure 4 illustrates the Canadian domestic market for livestock (left panel) and the international market for livestock trade between Canada and the United States (right panel), including Canada's export supply of livestock, given its domestic market conditions (shown in the left panel), and US import demand for livestock given its domestic market conditions (not shown). A WTO-consistent regime for the United States would yield an import demand curve of $MD_1^{US} $ , the market-clearing price would be P ₁, and the equilibrium volume of trade – Canadian exports of livestock to the United States – would be given by Q ₁. In the Canadian domestic market (left panel), Q ₁ is equivalent to the difference between domestic quantity supplied (q ₁) and domestic quantity demanded (d ₁) at price P ₁.

Figure 4. The retaliation limit under the mathematical formulation of reciprocity: Canadian domestic livestock market (left) and Canada–US livestock trade (right)

Now suppose the United States implements a non-tariff barrier, such as the COOL regulation. This shifts the US import demand curve for livestock from $MD_1^{US} $ to $MD_2^{US} $ , causing the volume of US imports from Canada to fall from Q ₁ to Q ₂ (right panel) and total livestock production in Canada to fall from q ₁ to q ₂ (left panel).

Under the Bagwell and Staiger (Reference Bagwell and Staiger2001) mathematical definition of reciprocity serving as a limit to the tariff retaliation, the Bown–Ruta model shows that this would be given by the shaded area in the right panel, or equivalently by the combination of the two shaded areas of the left panel. These rectangular areas are determined by the initial world price (P ₁) multiplied by the difference between the WTO-consistent level of exports (Q ₁) and the level of exports under the WTO-inconsistent COOL measure (Q ₂) i.e., P ₁[Q ₁ − Q ₂] in the right panel, which is equivalent to P ₁[q ₁ − q ₂] + P ₁[d ₂ − d ₁] in the left panel.

It would be easy for arbitrators to implement this theoretical formula. Provided that post-violation import volumes (i.e., quantities Q ₂) can be disentangled from post-violation import values (i.e., P ₂ Q ₂), this reciprocity formula is no more difficult to implement than the trade effects formula that the arbitrators have frequently chosen, including in US–COOL, as we describe next.Footnote ¹²

In both this case and in the arbitrators’ choices, Q ₁ and P ₁ are unobserved or ‘counterfactual’ (WTO-consistent) levels of trade volumes and prices. The arbitrators need to derive these counterfactual levels for Q ₁ and P ₁ through economic modeling techniques, as we describe in Section 4. For now, it is sufficient to note that the arbitrators could come up with estimates for Q ₁ and P ₁, and this implies that they could have implemented the theoretically motivated Bagwell–Staiger formula of Figure 4 if they had wanted to do so.

3.2 The Arbitrators’ decision on the formula in US–COOL

The arbitrators in US–COOL could have implemented a level of retaliation consistent with this definition of reciprocity, but they did not. Instead, they followed earlier panels, such as EC–Bananas III and EC–Hormones, and used a trade effects formula under a slightly different definition of a change in export revenues. The alternative formula is illustrated in Figure 5.

Figure 5. The retaliation limit under ‘trade effects’

In US–COOL, the arbitrators authorized the complainants to retaliate by an amount equal to [P ₁ Q ₁ − P ₂ Q ₂] (right panel of Figure 5). This is equivalent to[P ₁(q ₁ − d ₁) − P ₂(q ₂ − d ₂)], illustrated in the left panel. The change in export revenue under this trade effects approach allows not only for volumes to change(Q ₁ → Q ₂), as is also the case under the reciprocity formula, but also for prices to change (P ₁ → P ₂). The latter is different from the reciprocity formula, which evaluates the change in export volumes at the fixed price of P ₁.

The level of retaliation under the trade effects approach illustrated in Figure 5 will be at least as large as the amount of retaliation authorized under the reciprocity approach illustrated in Figure 4.

It is, however, instructive to consider two scenarios in which the trade effects and reciprocity approaches are equivalent. The first occurs when Q ₂ = 0, i.e., the WTO-inconsistent US non-tariff barrier is prohibitive and cuts off all imports from the complainant of the disputed product. The second occurs when P ₂ → P ₁, i.e., the WTO-inconsistent US non-tariff barrier results in very little change in the exporter received price, which could happen if the export supply curve is very elastic. Otherwise, the trade effects approach permits a larger amount of retaliation by the complainant than the strict definition of reciprocity implied by the Bagwell–Staiger formulation.

3.3 Evaluating the complainant's request for a different formula

Next consider what Canada, the complainant, requested in the arbitration for additional compensation – beyond the trade effects illustrated in Figure 5 – for the negative impacts on domestic sales (in the Canadian market) due to the WTO-inconsistent COOL measure. In Figure 6, we use this simple economic model to describe potential ways of interpreting this request to evaluate the arbitrators’ decision not to include it in the formula for determining Canada's retaliation limit.

Figure 6. The retaliation limit under ‘trade effects’ plus ‘domestic price suppression’

One possible way to interpret the request is that Canada seeks to be compensated for lost domestic revenue for its livestock producers for their livestock sold at home. Recall that in the model, the Canadian price of livestock falls from P ₁ to P ₂ when the United States implements COOL. However, as Figure 6 illustrates, because of the standard assumption that the Canadian demand curve for livestock is downward sloping, quantity demanded (i.e., the volume of domestic sales) increases from d ₁ to d ₂ with the decline in price. Thus, it is not necessarily the case that Canada's domestic revenues fall with the implementation of COOL. The change in revenue associated only with the change in domestic (Canadian) sales is given by [P ₂ d ₂ − P ₁ d ₁], or the area [b − a] in Figure 6. If b > a, the change in domestic revenues associated with the decrease in price is positive because of the COOL requirement. Specifically, revenues associated with domestic sales increase with a decline in the domestic price when the initial equilibrium is situated at a relatively elastic part of the demand curve, so that the percentage increase in quantity is larger than the percentage decrease in price.

An alternative way of interpreting the request is that perhaps Canada seeks to recoup the lost revenues, associated with the drop in price, which would be limited to the domestic sales volume that would have existed under a WTO-consistent policy. This approach ignores the (positive) change in revenues associated with the increase in Canadian quantity demanded due to the fall in price (area b in Figure 6) and only considers the losses due to the original level of sales (area a in Figure 6).

The problem with such an approach is that it ignores the fact that when the Canadian price of livestock falls from P ₁ to P ₂, Canadian purchasers of livestock (meatpackers and final consumers) benefit. In Figure 6, their economic well-being is measured by the economic concept of ‘consumer surplus’: with the decline in price, Canada's consumer surplus in the livestock market increases by the area given by [a + c]. These gains to Canadian consumers of the lower price more than offset the losses in producer surplus associated with the change in domestic sales [a].Footnote ¹³

Furthermore, the complainant's proposed approach is ad hoc and moves further from reliance on an economic modeling framework. Such a framework is beneficial in that it imposes discipline on the arbitrators through both consistency of analysis and the requirement that all changes be taken into account.Footnote ¹⁴

The complainant's proposal seeks to focus only on the ways in which COOL imposed losses on Canadian economic well-being, ignoring benefits elsewhere in the Canadian economy. Such a non-model-based approach leads to double counting. It also raises follow-up questions regarding where to draw the line.

First, if one were to adopt the complainant's approach, the next logical question would be: Why stop at only trade losses (the shaded areas of Figure 6) and losses in domestic sales (area a in Figure 6)? For example, in a different modeling framework, one could show that there were also losses to domestic factors of production that are specific to the livestock industry. From that perspective, why not also add into the calculation the workers whose wages are lower or the investors in livestock production (capital owners) whose rents are lower, each of which also loses in the short run because of falling prices of Canadian livestock? Our reason for leaving them out of the analysis here is that they are not included in this particular economic model, and the choice of economic model determines what data are needed, and what are not, to compute the retaliation. Put differently, reliance on an economic model clarifies not only where to begin the analysis but also where to draw the line and end the analysis.

Second, the complainant's non-model-based approach ignores that any change in trade policy has distributional impacts for a country that must be considered in an economic welfare calculation. Specifically, there are segments of the Canadian economy in which the change in economic well-being due to the COOL requirement is positive, as observed in Figure 6. As mentioned, Canadian meatpackers and consumers of beef gain from COOL because more domestic production is retained locally and prices are lower. However, just as an economic model provides a framework to show where to draw the line in counting up losses, it imposes discipline that appropriately limits the scope of those who benefit from the COOL measure as well. In a different economic model, other domestic groups could be shown to benefit from the lower Canadian consumer prices for beef from the COOL measure. For example, Canadian producers of ketchup and mustard – complements to the consumption of beef – are better off as the price and consumption of such goods increase with the increased consumption of beef.

These points push back against the complainant's proposal in the US–COOL arbitration, in which the complainant cherry-picked a longer list of groups in Canada that suffered negative distributional effects from the measure, without considering what would likely be an equally long list of those who enjoyed positive distributional effects.

However, these observations also reinforce the argument against the approach that the arbitrators have undertaken in US–COOL and in other disputes (e.g., EC–Hormones). While the Bagwell and Staiger (Reference Bagwell and Staiger2001) modeling framework provides logic-based and model-based consistency for defining a level of retaliation limited to the WTO-consistent policy's world price multiplied by the change in export volumes (e.g., P ₁[Q ₁ − Q ₂]; Figure 4), it does not follow for the change in the export revenue rule that the arbitrators have seemingly followed (e.g., [P ₁ Q ₁ − P ₂ Q ₂]; Figure 5) since the early WTO decisions (Bown and Ruta, Reference Bown, Ruta, Bown and Pauwelyn2010).

Thus, while the arbitrators made the right decision to reject the complainant's proposal to include domestic price suppression effects, it is still somewhat unclear why they have continued to apply the trade effects formula that they apply.

4.1 The scale of the trade effects in US–COOL

Before going into details of the determination of which elements would feed into the formula, we step back to consider the scale of retaliation that Canada and Mexico requested, as well as what was authorized, to put these into perspective.

In the 2015 arbitration, Canada requested roughly $2.4 billion (CAN $3.1 billion) and Mexico $713 million in retaliatory compensation. Even when focusing on the requests limited to perceived export revenue losses (under the trade effects formula), Canada requested $1.6 billion (CAN $2.0 billion) and Mexico $515 million. The United States claimed much smaller trade effects due to the COOL measure, estimating the lost export revenue for Canada at $43 million and for Mexico $48 million – i.e., less than $100 million combined.

The Article 22.6 arbitrators ultimately granted Canada $805 million (CAN $1.054 billion) and Mexico $227.8 million. Again, retaliation granted is based on a formula that gives the value of export revenue lost annually because of the COOL regulation. These figures thus represent approximations of the value of additional annual bilateral exports to the US market in 2009–2015, had the COOL regulation not been in place.

Figure 7 puts these amounts into perspective by plotting them with the information on actual US imports of cattle and swine from Canada and Mexico over the 1989–2015 period. The solid lines represent US imports from Mexico (grey) and Canada (black), the dashed lines represent what the arbitrators granted plus the actual level of imports, and the lines with boxes represent the ‘trade effects’ portion of the Canadian and Mexican requests plus the actual level of imports.Footnote ¹⁵

Source: Constructed by the authors with data from the USITC's Dataweb and WTO reports. *indicates the retaliation requests limited to only the ‘trade effects’ component and thus does not include the additional request for the ‘domestic price suppression’ component (that the arbitrators denied formulaically).

Figure 7. US imports of livestock and implications of proposed WTO retaliation levels, 1989–2015

Consider Canada beginning in 2009, the year of COOL implementation. The arbitrators’ authorized level of retaliation puts their approximation of what Canada's exports to the United States would have been without the COOL regulation at $2 billion, roughly equivalent to its peak level of exports in 2007. This is relevant because the arbitrators are essentially indicating that Canada's cattle exports would not have declined at all in 2007–2009, even though there had been a global trade collapse in 2008–2009 – in virtually all products and in all countries – due to the Great Recession.

The story for Mexico is even more dramatic. The arbitrators’ decision implies that Mexico's exports of cattle to the United States would have increased by roughly 30% in 2007–2008 but for the COOL regulation.

Before imposition of the COOL regulation, total US livestock imports from Canada and Mexico peaked at $2.5 billion in 2007. Given that exports in 2014 – with the COOL regulation still in place – were $2.9 billion, this implies that total combined exports without COOL in 2014 would have been $3.9 billion, or that – but for COOL – Canada's and Mexico's exports would have increased by 87% over seven years. These numbers are clearly difficult to rationalize against the US market reality and thus raise important questions about how the arbitrators arrived at such figures.

It is worth pointing out that at least the arbitrators did not grant the even higher levels that Canada and Mexico had requested, which are also plotted in Figure 7.

4.2 Difficulties for Arbitrators in attempting to implement the COOL formula

As indicated in Section 3, once arbitrators decide on a formula, they need information on the values of the parameters that will be used to compute the formula. In particular, post violation import volumes (i.e., quantities, Q ₂) are known – see, for example, the data for the US cattle market (Figure 2) or US swine market (Figure 3). Thus, the arbitrators only need to determine counterfactual levels of Q ₁ and P ₁, or the projected trade volumes and prices if the United States had implemented a WTO-consistent COOL regulation.

The intellectual exercise is thus to predict Canada's exports of cattle and hogs and Mexico's exports of cattle to the US market in 2009 and beyond, in order to compare those predicted values with the actual volumes of exports – in other words, to determine what the 2009–2015 export volumes would have looked like if the United States had not imposed the COOL measure. In such an exercise, economists typically use data from the past to predict the future.

In Figures 8 and 9, we illustrate just how divergent the outcomes to this basic prediction can be when we change only one element: the past years used to predict the future. We rely only on publicly available data, which will not match exactly what was described by Canada, Mexico, the United States, or the arbitrators in the report – data to which we do not have access. Furthermore, although we refer to decisions made by the arbitrators in implementing these economic models, we recognize that the arbitrators are frequently constrained by the quality of the data, analysis, and models that the parties put before them. This is an important factor, but we focus here on other challenges that arbitrators face.

Source: Constructed by the authors with data from the USITC's Dataweb.

Figure 8. Using 1989–2008 total trends to construct counterfactual cattle export volumes

Source: Constructed by the authors with data from the USITC's Dataweb.

Figure 9. Using 2004–08 bilateral trends to construct counterfactual cattle export volumes

Figure 8 illustrates the supposition that post-2009 total import volumes of cattle followed the trend of total US imports of cattle over the prior 20 years (1989–2008). As we noted above, notwithstanding substantial annual fluctuations in bilateral trade during this period – e.g., when US imports from Canada declined in a year in response to a shock at home, imports from Mexico would tend to increase to fill the gap – total US cattle import volumes from the two were fairly flat. Thus, using the past data from the whole period to predict the counterfactual would suggest that US imports in 2015 were not much different from what they would have been as predicted by this 20-year pre-COOL trend – about 2.2 million head.

Suppose instead that we modify the approach that generates the counterfactual prediction very slightly along two dimensions. First, rather than data from the 20-year period we use data only from a more recent period – say, 2004–2008. Second, we construct the prediction for each US trading partner individually.

The result is illustrated in Figure 9, which shows the other extreme. This model predicts a level of Canadian exports of 4.5 million head of cattle by 2015 – twice as much as Canada had exported to the United States in any given year before COOL, and twice as much as the total predicted to be exported by Canada and Mexico combined in Figure 8 under the alternative approach.

Why is this prediction for Canada so different? Canada had a BSE crisis that left its cattle exports at zero in 2004. Once the US BSE import ban was lifted, because Canada already had a highly developed production capacity, it experienced a substantial increase in export growth to the United States in 2005–2008. However, this growth was not driven by some underlying economic fundamental, like a boost in productivity that might occur after adoption of a new technology. The sharp increase in export growth over 2005–2008 was simply due to Canada's resuming cattle exports from zero in 2004 to the 2002 (pre-BSE) levels. Put differently, it is unlikely that Canada would have kept up the same level of export growth that it experienced in 2005–2008 after 2008 – i.e., once it had reattained its pre-BSE levels of exports to the United States and was again producing and exporting at capacity.

The years chosen to illustrate this argument were selected deliberately, of course. The point is that even the selection of years to use in the historical sample to generate a prediction can make a sizable difference in the results.

To see this even more clearly, suppose that the years 2004–2008 were also chosen to predict Mexican export volumes to the United States. As Figure 9 illustrates, Mexico's cattle exports to the United States were declining slightly during those years. Some of this decline was a normal market response – US imports from Mexico were declining slightly as US imports from Canada were resuming after the US lifted the BSE import ban on Canada.

The challenge of determining counterfactuals raises at least two additional complications for arbitrations such as US–COOL – that involve multiple complainant countries – to have to address.

First, to what extent should it be necessary for arbitrators to demand consistency in approach across the complainant countries? For example, the demand for consistency in generating the predictions that we have just described would require using the same years for Mexico as for Canada. That would seem like a reasonable rule of thumb, if not a requirement. But, in this instance, when that approach is taken, the trend predicts that Mexico's exports to the United States (even without COOL restrictions) would have declined to zero by 2014 (Figure 9). Would this then invalidate the use of the approach for the Canadian data?

Second, and regardless of whether such consistency of approach is required, should there be an ‘adding up’ constraint to impose discipline on the arbitrators? For example, in a dispute such as this, should the arbitrators be required to ensure that each country's individual model does not aggregate up to result in a combined counterfactual volume of total US cattle imports that is unrealistic, given the underlying market conditions?

Suppose that the arbitrators had adopted the approach illustrated in Figure 8 to determine Mexico's counterfactual exports and the approach illustrated in Figure 9 to determine Canada's counterfactual exports. An adding-up constraint would reveal that the combination of the two models’ predictions provides an implausibly large increase in total exports of cattle to the US market to nearly 6 million head.

While the cases illustrated in Figures 8 and 9 are extremes, these simple exercises starkly illustrate important issues that arbitrators face in these types of disputes. Using only a very simplistic economic (time trend) model to predict the counterfactual volumes, we showed that even slight (and arguably quite reasonable) differences in the methods of applying that model can yield wildly divergent results.

Furthermore, in the application that the arbitrators adopted in the US–COOL dispute, there were many complicating factors besides the COOL regulation – including the Great Recession – that impacted Canadian and Mexican exports of cattle (and swine, not shown) to the United States. We documented several of them in Section 2. Constructing a level of counterfactual exports depends heavily on which of these factors the arbitrators took into consideration. As outsiders, we are unable to discern exactly which of the arbitrators’ choices led to the authorized retaliation levels that we argue were implausibly large.Footnote ¹⁶ Identifying the precise factors would require access to the underlying data and models.

4.3 Additional concerns associated with implementation of the formula

Other important issues arose during the arbitration. We do not have sufficient space to assess them all here, but flag two that raise additional concerns about the process.

The first involved the discrepancy between Canada's provision of counterfactual estimates based on weekly data and US counterfactual estimates based on monthly data. The data led to quite different results.Footnote ¹⁷ Canada argued that the weekly data were preferred because the higher frequency provided more variation that allowed for more precise estimation. The United States argued that there was likely to be more measurement error in the weekly data, which were collected by a different government agency (USDA's Animal and Plant Health Inspection Service, APHIS) than the one tasked with collecting data on official US trade statistics (Census Bureau). Unlike the monthly official trade statistics, the weekly APHIS series was not checked for errors, corrected, or revised.

Each side has a legitimate argument on theoretical (statistical) grounds. The question is: Which effect is larger and more likely to significantly bias the results? Unfortunately, that may only be discerned empirically through careful examination of the data, if at all.

Furthermore, there is the potential long-run impact associated with the US government not being allowed to prevent certain data from being used in the arbitration. Would this encourage the United States to become less transparent or have other unintended consequences? Indeed, this case involved a different US government agency (APHIS) collecting unofficial data on cattle border crossings for its own purposes of tracking animal health. But such data inadvertently ended up harming the United States. Could this result in one US government agency (e.g., the Census Bureau) being forced to instruct another US government agency (e.g., USDA) to not collect its own data on trade flows if it does not have the resources to revise the data and reconcile them with other official US data series? Given that APHIS was collecting the data with good intentions, a byproduct of such a policy could be that such a prohibition inadvertently harms public (and animal) health.

The second important concern arose because Canada and the United States used very different empirical approaches to establish their estimates.Footnote ¹⁸ Canada relied on econometric regression techniques, whereas the United States chose to use a partial equilibrium simulation model referred to as an equilibrium displacement model (EDM). The arbitrators did not rule out the US approach a priori, but they did not consider it seriously, under the argument that they considered the US implementation of the model flawed because the United States assumed that the compliance costs of the COOL measure were non-discriminatory and applied to US industry.

An open question is whether the arbitrators would have been satisfied with a US model in which the United States assumed that a US meatpacker using (some) imported Canadian livestock as an input faced a higher cost of compliance than a US meatpacker using only domestic US livestock.Footnote ¹⁹ While the United States attempted such an approach in response to questions from the arbitrators, it did so only for the original COOL measure, not the amended COOL measure. The arbitrators found that to be enough to dismiss it.

In our view, it is unfortunate that more attention was not paid to exploring the results of such an alternative estimation approach. This is mainly because the parties and arbitrators constructed such widely divergent estimates for the size of the trade effects. A better understanding of the US model might have provided more insight as to the source of the large differences, and this information might have allowed the arbitrators to use an approach that provided trade effect estimates more in line with the market reality. Again, as described in Section 3, the figures they settled on seem far too large.