AN EMPIRICAL MODEL OF ESF BAILOUT SELECTION

The first step in constructing an empirical model of ESF bailout selection is identifying an appropriate population for statistical analysis. Ideally, I would have a list of all countries that approached the Treasury for assistance—both those that received rescues and those that were denied—during the period under investigation. Unfortunately, the US Treasury will not release any information regarding foreign requests for US emergency financial assistance that were denied; nor will it say whether the United States has ever extended a bailout only to be rebuffed by the targeted borrower.[1] In other words, the only data available on ESF bailouts are instances where the United States is known to have extended a loan to a country in distress. This makes the selection of "negative” cases necessary to build an appropriate population for statistical analysis more complicated than one would like. Without the true population of negative cases, I developed a second-best approach to sample selection. Treasury insists that if a country is to receive assistance from the ESF, it should have filed a letter of intent with the IMF.[2] Again, this is the point at which the borrower government and Fund staff have agreed to the terms of the loan and any required reforms. Based on this information, my sample comprises all countries that requested IMF assistance during the years under investigation. Although nearly all of these loan requests were approved by the executive board, a few were not. However, constructing a sample of requests is more suitable than basing it on actual loan approvals since, as discussed in the previous chapter, in most cases ESF credits were made prior to executive board approval. In other words, filing a letter of intent— not board approval—is the proper prerequisite for a developing country to receive an ESF bailout. Finally, the sample is restricted to requests from 1983 to 1999 because of data limitations.[3]

The presence (or absence) of an ESF bailout in a given year is the primary outcome of interest in this analysis. I confirm the presence of a US bailout through documentation of the ESF’s use for foreign credits (which I obtained directly from US Treasury by request) as well as additional secondary sources.[4] During the period under investigation, the United States intervened on 34 (yearly) occasions on behalf of 20 different countries via the ESF. However, due to missing data, these numbers drop to 28 and 17, respectively, in my sample.[5]

I rely on two key measures of US financial interests to test my argument. First, to capture the private interests of big US banks, I construct a measure of SIFI bank exposure. This simple measure accounts for the concentration of their foreign loans by country in percentage terms. Precisely, I divide the foreign claims of SIFIs on a given country by total foreign claims of those banks in that same year and then multiply this by 100. Thus, a value of 1 would indicate that 1 percent of SIFIs’ foreign claims are concentrated in that country at that time. A higher value indicates increased exposure. The second measure I construct is designed to account for systemic risk. The measure captures changes in the risks facing the US banking system from all foreign financial crises in a given year.

Figure 5.4

SIFI’s Foreign Claims and Capital Stock, 1982-1998

I base my construction of this measure on several assumptions. First, I assume that foreign claims of SIFIs are more at risk in countries that seek IMF assistance than in countries that do not. Second, by adding up the total claims of these big banks to all countries that sought an IMF rescue in a given year, we can capture the global risk climate facing major US banks at that moment in time. I assume that, all else equal, in a year when three or four countries to which SIFIs are highly exposed seek IMF assistance, the risks to the broader US banking system are higher than a year when only one of those countries approaches IMF. In such a case, there are multiple financial fires burning in the world and, thus, the risk climate is more threatening. Third, I assume that the vulnerability of the US banking system is not just based on the sum total of bank claims to economies facing financial crises. It is also based on the capital stock that banks hold in reserve. If the capital held by SIFIs rises relative to their total foreign claims, all else being equal, the system should be less vulnerable to foreign shocks. Banks should be more capable of weathering the storm by drawing down their own reserves. Figure 5.4 plots total foreign claims, total foreign claims on countries requesting IMF assistance each year, and total reported capital of SIFIs.

Building on these assumptions, we can estimate systemic risk facing the US banking system from foreign sources by taking the sum total of SIFI claims on countries seeking IMF assistance in a given year (the dotted line in Figure 5.4) divided by total capital held by SIFIs (the shaded area in Figure 5.4). The result is an index that reveals how US banks’ capital stacks up in relation to their outstanding loans to economies in crisis over time. Figure 5.5 displays this systemic risk index for US banks,

Figure 5.5

Systemic Risk Index, 1983-1999

disaggregated across groups as well as in the aggregate. A point higher up the y-axis represents a higher level of risk, whereas moving down represents decreasing risk. For instance, the measure of 0.62 for SIFIs in 1989 means that the sum total of bank loans to countries requesting IMF assistance that year were equal to 62 percent of their total capital. Unsurprisingly, systemic risk was the highest in 1983 as the IMF was flooded with requests from developing countries during the first full year of the international debt crisis and US banks held a relatively small amount of capital in reserve. Systemic risk remained elevated throughout most of the 1980s before dropping in the 1990s. Risk was most elevated in that decade in 1997 as the Asian financial crisis erupted. To test my argument, I interact both measures of US financial interests: SIFI bank exposure and the systemic risk index. I anticipate that when both of these measures are elevated, the probability of ESF rescues will be the highest (corresponding to quadrant B in Figure 5.3). Both of these variables are discussed in more detail in the appendix.

I also include a number of additional covariates in the model to control for potentially confounding factors. Other considerations outside of financial interests may also motivate policymakers to bailout economies in crisis. For example, scholarly accounts of the decision by policymakers to provide ESF credits to South Korea in 1997 point to that country’s strategic value as an ally in Asia as influencing the decision.[6] To account for US geostrategic interests in a foreign country, I account for its United Nations General Assembly (UNGA) ideal point distance with the United States.[7] Smaller (larger) ideal point distance indicates that countries’ foreign policy preferences are closer to (farther from) US preferences. Although UNGA votes may not carry much actual weight in international relations, they may be useful symbols of support, solidarity, and common interests between countries. As has been argued elsewhere, "Even though UN votes may not be very important, they may still be an accurate signal of alliances and common interest ... . They may be correlated very strongly with important strategic interests.”[8] To control for the possibility that nonfinancial economic ties influenced US bailout selection, I control for countries’ share of trade with the United States.[9] Another factor that may influence US bailout decisions is a country’s domestic political institutions. For example, Treasury defended its 1984 Argentine bailout, in part, by pointing to Argentina’s transition to democracy. Consequently, I include Polity country scores to account for the regime type of each country in question.[10] I also account for Latin American countries in the model with a dummy variable because the vast majority of ESF credits were provided to countries in that region.[11] It may be the case that policymakers felt a greater responsibility to take care of countries in the United States’ "backyard” and less responsibilities for countries in other regions.[12]

It may also be the case that Treasury’s decisions to act as an ILLR were influenced by borrower-country need. Policymakers may have weighed the severity of the economic conditions facing a country in crisis when making financial rescue decisions.[13] Although all countries that approach the IMF for financial assistance are facing financial difficulty (otherwise they would not be asking for help), within this group considerable variation exists. Therefore, I control for a variety of macroeconomic conditions, including total national debt service costs over export earnings, the current account balance, total foreign exchange reserves (minus gold), and the annual GDP growth rate. I also include a dummy variable that accounts for whether a borrower country’s currency faced a speculative attack the same year, or the year before, they sought IMF assistance.[14] Variation in such factors may exacerbate or attenuate the severity of the crisis and, therefore, affect the extent to which countries are willing to ask for additional assistance. Such factors may also impact the willingness of the United States to help. In this sense, the Treasury or the Fed may act as international financial physicians that are more likely to treat patients suffering from especially severe forms of economic sickness. In addition to these need-based factors, I include standard macroeconomic control variables such as GDP and GDP per capita in all specifications.

I also include three domestic-level variables to account for the possibility that US political or economic conditions impact the ESF bailout selection process. As noted in the previous chapter, the Secretary of the Treasury decides when ESF resources ought to be marshaled in defense of a foreign economy, conditional on the consent of the president. Because variation in the political party affiliation of the president might affect bailout selection, I account for whether the executive is a Republican or Democrat.[15] It may also be the case that US economic performance has an effect on the likelihood that policymakers will decide to act as an ILLR. When the US economy is in the doldrums, administrations may be more reluctant to send resources abroad to rescue foreign sovereigns. Thus, I also account for the annual US GDP growth rate and the annual unemployment rate (in percentage terms), respectively.[16]

To address issues related to temporal dependence in models with binary dependent variables, I include a cubic polynomial.[17] Finally, in keeping with previous studies on international financial rescues, all the aforementioned covariates are lagged one year. This reflects the fact that the process of ESF bailout selection, like IMF loan selection, is based on information that lags behind the date of the actual decision.[18] Given the dichotomous outcome of interests (ESF rescue), I fit four logistic regression models: a first model with only the bank (SIFI) exposure measure and macroeconomic controls, a second that interacts bank exposure with the systemic risk measure as well as the macroeconomic controls, a third that adds in regime type and UN voting measures, and a fourth that adds in US economic controls.[19]

  • [1] I filed a Freedom of Information Act (FOIA) request with the US Treasury askingfor (1) a complete list of all ESF loan requests by country/date between 1972 and 2012 and(2) information regarding whether or not ESF loans were ever offered to countries withouta request. Treasury’s response in a letter to the author was as follows: “A search has beenconducted by this office and no records responsive to your request have been located.”
  • [2] In rare cases, Treasury provided assistance prior to a letter being completed (i.e.,Mexico in 1982); however, in such rare cases the borrower had been in negotiations with theIMF staff working toward agreement.
  • [3] Full details of the statistical sample are available in the appendix.
  • [4] Osterberg and Thompson 1999; Wilson 1999.
  • [5] Coding method of the dependent variable is discussed at length in the appendix.
  • [6] While defending his department’s decision to use ESF resources in defense of Koreaand Indonesia in the midst of the Asian financial crisis, then Treasury Secretary RobertRubin explained, “Our nation’s economic and national security are vitally at stake in the situation in Asia” (Stevenson 1998a; emphasis added).
  • [7] Specifically, I employ data from Bailey, Strezhnev, and Voeten 2015. The authors showthat ideal point estimates improve upon conventional dyadic similarity indicators such asAffinity or S-scores (Gartzke 1998; Signorino and Ritter 1999) by allowing for more validintertemporal comparisons: distinguishing UN agenda changes from changes in statepreferences.
  • [8] Dollar 2000, p. 38. A related body of research on the determinants of IMF lendinghas consistently found evidence that Fund loan selection (Thacker 1999), loan size, andloan terms are influenced by the borrower countries’ voting record in the UNGA (Dreherand Jensen 2007; Dreher, Strum and Vreeland 2009; Oatley and Yackee 2004; Stone 2004,2008; Thacker 1999; Vreeland 2003, 2005).
  • [9] Specifically, I calculate each country’s total trade (imports + exports) with the UnitedStates divided by total US trade with the world. I then multiply this by 100. Thus, the variable accounts for each country’s annual share of US trade.
  • [10] Marshall, Jaggers, and Gurr 2012. Polity consists of a 21-point scale and range from -10(least democratic) to 10 (most democratic).
  • [11] Specifically, this is a dummy variable equal to 1 if a country is classified as being part ofLatin America (including the Caribbean) by the UN, 0 if otherwise.
  • [12] A conversation during a 1982 Federal Open Market Committee (FOMC) suggeststhere may have been a perceived geographical division of labor between the United Statesand Europe when it came to financial crises. At one point during the meeting, ChairmanVolcker noted, "It certainly is in [Europe’s] mind as are some of these other things: LatinAmerican is your area” (FOMC 1982c, p. 65).
  • [13] For example, former Undersecretary of the Treasury for International Affairs DavidMulford once explained to Congress that ESF credits are “considered on a case-by-casebasis, based on a demonstrated need for liquidity and evidence of adjustment efforts in cooperation with the IMF” (US Senate 1984, p. 13; emphasis added).
  • [14] This variable, equal to 1 in the event of a speculative attack, 0 otherwise, is based onthe exchange market pressure (EMP) index developed by Eichengreen et al. (1995) andKaminsky and Reinhart (1999) as described in Leblang (2003). I discuss how I code thisvariable, as well as the source of the data, in the appendix.
  • [15] Specifically, I employ a dummy variable equal to 1 if the president at the time of an ESFcredit was a Republican, 0 if otherwise.
  • [16] I rely on the World Bank’s World Development Indicators (WDIs) for each of the aforementioned economic need-based and domestic-level variables.
  • [17] Carter and Signorino 2010.
  • [18] Knight and Santaella 1997, p. 413. The only exception is the covariate accounting forthe president’s party, since this would have obviously been known at the time the decision was being made.
  • [19] To address potential problems due to clustering in countries and years, I computeHeteroskedastic and Autocorrelation Consistent (HAC) standard errors. All models are fitted using the R package Zelig (Imai, King, and Lau 2007, 2008).
 
Source
< Prev   CONTENTS   Source   Next >