Kay and Quallenshots Fired? Foreign Aid and Retaliation in the WTO April 3, 2014

Kay and QuallenShots Fired? Foreign Aid and Retaliation in the WTO April 3, 2014

Abstract

Powerful states can exert themselves over less powerful ones, using all variety of policy instruments to assert themselves. Does this dynamic control in the World Trade Organization? Following mounting evidence that states use foreign aid as a political instrument, we ask whether one great power – the United States – uses reductions in bilateral foreign aid to punish states that file against it in the WTO. We compare immediate and delayed bilateral aid receipts from the United States with instances of state aid recipients filing or joining suit against the United States in the WTO, finding no significant evidence that any relationship – immediate or delayed – exists between the two variables. This finding complicates our understanding of the picture, suggesting that great powers may seek to be filed against and lose in the WTO, possibly as part of a two level game; that foreign aid is not so versatile a political instrument as many scholars suggest; or that these states are willing to restrain themselves to the WTO as an arena, avoiding extramural coercion.

Introduction

Great powers dominate the international system, or so the argument runs. They design institutions that serve their needs, and exert considerable power to maintain the status quo, even when that means exploiting emerging economies or circumventing the conventions of the international system. In this scheme, relationships reduce to exploitation; innocuous policy instruments become the means of political enforcement as states bring to bear every tool at their disposal. For states that vote against stronger peers in the UN, foreign aid dries up before election time; but for those that vote in line with them, the opposite occurs (Faye and Niehaus, 2011). For states that want IMF loans, UNSC alignment with its major stakeholders is key (Dreher, Sturm and Vreeland, 2009). These are the carrots and sticks of international enforcement and it is the most powerful states that wield them.

We consider the case of the World Trade Organization (WTO) – an organization which would appear to level the economic playing field. In the WTO, powerful states have no formal advantage over weak ones; Panels arbitrate disputes and states level penalties against one another. And yet, although the WTO seems to provide an arena in which states meet on a level playing field, the WTO is the subject of considerable concern. Most of these concerns orbit a single dispute: do economically advanced, more powerful states dominate the WTO? We are skeptical that an international organization can escape the will of its most dominant members. Even if the structure of the WTO encourages equality among its members, the levers of power do not cease to exist in Geneva.

We are not the first scholars to contemplate such a possibility. Guzmann and Simmons posit one possibility they call the “power hypothesis” (Guzman and Simmons, 2005). They propose the possibility that weak states are deterred from filing against powerful states in the WTO by the fear of costly retaliation. While they argue that we do not see clear evidence of this in defendant selection, they do not search for actual retaliations. In fact, by using defendant selection by weak states as a proxy for retaliation, Guzmann and Simmons’ overlook the possibility that retaliation still takes place within the WTO, but that it does not deter weak states from filing against stronger ones. Work by Davis and Bermeo seems to confirm the idea that states retaliate against each other in the WTO by directly search for and finding patterns of retaliation, implying that weak states are willing to risk retaliation and that such retaliation does indeed take place (Davis and Bermeo, 2009). This idea finds precedent in sanctions. Despite claims of their ineffectiveness, sanctions still exist as retaliatory mechanisms seeking to control state behavior. In sanctions, we see clearly that the relative ineffectiveness of a retaliatory technique does not necessarily rule out its use. Returning to the WTO case, we consider one such potential means of retaliation – foreign aid – which has proven versatile in its political uses in recent scholarship (cites). We imagine that states may use bilateral aid as a stick, punishing those states that file against them in the WTO by reducing bilateral commitments.

To investigate this possibility, we test whether the United States – for which the greatest number of WTO suits and foreign aid data are available – reduces foreign aid to those states that file against it. We draw data on American bilateral foreign aid, other than military assistance, from Vreeland and DreherThe Political Economy of the United Nations Security Council. We code our own dichotomous indicator using records from the WTO Dispute Settlement Gateway for country years in which states file or join suit against the United States. We scrutinize the relationship between these data using a variety of treatments and find that no significant evidence that one exists.

This finding keeps open a number of possibilities. Firstly, it bears out the belief that the power hypothesis cannot find significant evidentiary support. Two clear possibilities exist as to why this may be the case – either states are uninterested in using foreign aid as a political instrument in the WTO context, or they do not consider losing WTO cases detrimental to their interest, possibly as part of a two level game in which the WTO functions as political cover for highly accountable states to relinquish protectionist policy. In either case, our results fail to suggest that – in this case – states are applying external modes of power to control WTO filings, calling into question criticisms of the WTO based on such reasoning and suggesting that in fact even powerful states may respect the boundaries of multilateral mechanisms such as the Dispute Settlement Understanding.

The remainder of this paper proceeds as follows. First, we discuss our data and how we choose to code for WTO filings. Next, we discuss the methods by which we treat the relationship between our variables of interest. We then present our results; and we position our results within the literature, drawing our conclusions on their importance and suggesting avenues for potential research. It is within this concluding section that we consider how to theoretically accommodate this result.

Data

We rely upon two datasets for our data: one from Vreeland and Dreher’sThe Political Economy of the United Nations Security Council and the other sourced from the WTO. The first dataset codes observations of foreign aid from OECD countries, several multilateral institutions, and a variety of other control indicators across most countries since 1960. We observe 11,033 country years in this dataset. In addition to this dataset, we code dichotomous indicator variables to correspond to suits filed against the United States in the WTO. We code indicators to represent the origination of a suit, the joining of a suit, and a variable to aggregate the two. We code instances of each type as 1 and all other country years in the Vreeland dataset we code as zero.

Our principle variables of interest are bilateral receipts of development assistance from the United States (response variable) and whether a given country filed a suit against the United States (causal variable). We drop instances where no foreign aid data from the United States is available. We are then left with 5039 observations of our chief variables.

Before discussing treatments of these variables, we summarize these most basic variables in Table 1.

Table 1 – Main Variables of Interest

Variable / Number of Observations / Mean / Standard Deviation / Maximum / Minimum
Development Assistance from the United States / 5039 / 122.4939 / 412.8584 / 12307.32 / -156.89
Suit against the United States? / 5039 / .0273864 / .1632227 / 1 / 0

*Table 1 outlines summary statistics for our key variables of interest. After dropping observations that do not include US foreign aid data, we observe 5039 country years for each variable. Development assistance from the United States is given in millions of constant 2009 dollars. Our dichotomous indicator is coded only 1 and 0 – mean value indicates that among nations who receive foreign aid from the United States, we find 138 instances of originating or joining suit against the United States with which to conduct our later analysis.

Methodology

We conduct multiple regressions testing for a relationship between differing treatments of our independent and dependent variables of interest.

Immediately, the distribution of our response variable – development assistance from the United States – demonstrates itself as problematic. The minimum lies within one standard deviation of the mean, while the maximum value lays almost thirty deviations above the mean. To normalize this distribution and negate potential regression leverage of extreme cases, we take the natural log of development assistance as our response variable instead of raw data. We are consistent with Vreeland and Dreher in this choice.

Figures 1 and 2 below demonstrate the original and normalized distributions of our aid data for the United States.

Figure 1 – Unadjusted US Aid Data Figure 2 – Normalized US Aid Data

Figure 1 is a histogram plot of aid receipts from the United States, in millions of constant 2009 dollars. In figure 2, these data have been normalized through the application of a natural logarithm, which is what we use as our response variable.

Our response variable requires further adjustment. We believe that foreign aid may be used a retaliatory mechanism, and yet neither WTO disputes nor foreign aid policy are resolved immediately. Retaliation, then, may not be contemporary with cause. We must consider the possibility that retaliation will require years to materialize. In order to do this, we carry out multivariate regressions using a variety of delayed treatments of our causal variable. We consider lagged and unlagged versions of our causal variable, going up to a seven year delay. We also consider the possibility that conference request dates do not accurately reflect the beginning of a dispute – that certain trade disputes may precede conference request by years, and that policymakers may attempt to reduce aid even before formal proceedings to signal the credibility of a threat. Therefore, we also consider versions of our causal variable that lead for up to seven years.

We source our baseline control variables from Vreeland and DreherThe Political Economy of the United Nations Security Council (2014). These variables closely align to our study because Vreeland and Dreher conduct analyses that also consider influences on bilateral assistance from the United States. We are unsure that Vreeland and Dreher capture all the potential important controls, and so we add imports, exports, total trade flow with the United States. In each case, heavily skewed distributions lead us to use natural logarithms to normalize the data.

Vreeland and Dreher’s original controls include:UNSC membership, pariah status, the presence of war (greater than 1000 deaths), real GDP per capita (normalized with a natural logarithm) polity data, and US military assistance in constant 2011 dollars.

We discuss the findings of these techniques in our ‘Results’ section

We find it worthwhile to note that our choices to scale or not to scale certain variables have tremendous influence on outcomes in the context of our results. Table 2 demonstrates that the decision to include raw or normalized population and foreign aid data each change both the sign and significance of our correlation. This ought to both inform our choices to normalize most skewed data and make clear to the reader now that our results prove seriously inconsistent.

Table 2 – Effects of Variable Treatments on Results

(Sign, dependent variable, control variable) / POP / Pop_ln / No Population Control Variable
US Foreign Aid (Raw) / +, sig, sig / -, sig, ~sig / -, sig, N/A
ΔUS Foreign Aid (Raw) / +/-, insig, sig / +, sig, insig / +, sig, N/A
US Foreign Aid (Natural Log) / -, sig, sig / +, insig, insig / +, insig, N/A
ΔUS Foreign Aid (Natural Log) / +, insig, insig / -, insig, insig / -, insig, N/A

Table 2: This chart illustrates the results of regressing the various forms of our dependent variable, US bilateral foreign aid, with the raw and natural log forms of our “population” control variable included in our baseline specification. The first value represents the sign of the correlation between our independent variable, whether or not a country filed or was a third party to a WTO dispute against the USA, with a lag of three years, and different dependent variables. The second value declares whether or not our independent variable proved significant to at least the |P|< .10 level. The third and final value expresses whether or not the control variable was significant to at least the |P|< .10 level.

Results

After a battery of analysis and controls, our results fail to suggest that filing a suit against the United States, or being a third party to one, has any significant effect upon the foreign aid that a country receives from the United States. This non-finding is robust across a variety of controls, including “fixed effects”. We summarize these findings in Table 3

In each of our models, we present a three year lagged version of our independent variable, meaning that we search for evidence of retaliation three years after the conference request date. While we tested for both delays and leads—seven years in each direction— in our independent variable, we have chosen to present the three-year lag as it produced the best results of the various lags and leads that we examined. Despite representing the best results, our independent variable fails to achieve statistical significance in every of our four models and regardless of controls, implying a robust non-finding.

In our first model, we test our baseline specification without fixed effects and find no significant correlation between our independent and dependent variables. Regarding our control variables, we see that imports, UN Security Council membership, real GDP per capita, pariah status, and US military assistance the only statistically significant controls. These results largely remain in line with our expectations, as pariah status and a higher real GDP per capita would sensibly lead to a decrease in bilateral assistance from the US, while receiving US military assistance and strong trade relations, represented by imports, would logically lead to an increase in aid.

Where our results surprise us in the first model is in that UNSC membership appears to have a significant negative effect on bilateral aid, contradicting Vreeland’s findings that show a significant positive effect on US foreign aid (Vreeland 2014). This finding only holds in models one and two, which exclude fixed effects. Because the result does not hold with fixed effects, it is likely due to an endogeneity issue. States that accede to the Security Council tend to be more powerful and prosperous, and thus receiving less aid from the United States. However, once a country gains a seat on the Security Council, it tends to receive more aid, hence the elimination of the significant negative effect when examine models that include fixed effects, suggesting our results do not actually conflict with Vreeland’s.

In the second model, we drop each of our insignificant control variables, although this does not improve the significance of our independent variable. The only notable result here is that the “Pariah dummy” control loses its significance once we drop the insignificant controls from the first model. While this might suggest an oddity in the data, the small number of pariah states more likely means that there exists some phenomena related to a few pariah states, not any larger trend in the data.

In our third model, we include fixed effects and find that as a result, not only does significance elude our independent variable, but nearly all of our controls as well, save for real GDP per capita. We choose to use fixed effects in this model because by creating separate intercepts for our dependent variable for each country, they allow us to examine exogenous trends in foreign aid as opposed to ones potentially exhibiting endogeneity.

Moving to our fourth model, we retain fixed effects but remove each of the insignificant control variables from the third model, meaning all except real GDP per capita. In this final model, our independent variable, filing against the US in the WTO, again proves to be an insignificant factor in determining US bilateral aid. As our independent variable remains insignificant across each of our four models, this seems to suggest a robust non-finding and a strong lack of evidence in favor of the hypothesis that the US uses its foreign aid as either a stick or a carrot to react to those who file against it in the WTO.

Table 3–Effect of WTO Filing on US Foreign Aid Receipts