RESULTS

Table 5.1 reports the results for all four models. First, I will focus my discussion on the measures of US financial interests. On its own, bank exposure does not appear to impact the likelihood of a financially distressed country being selected for an ESF bailout (see Model 1 results). Although the coefficient is positive as expected, it is not statistically significant at conventional levels. However, when bank exposure interacts with systemic risk, the effect is both substantive and statistically significant across all remaining models. To illuminate the effects of systemic risk and bank exposure on ESF bailout selection, Figures 5.6 and 5.7 present the simulated predicted probabilities of a US rescue as SIFI bank exposure increases from its sample minimum (0 percent) to its maximum (10 percent).[1] The solid lines represent the mean expected probability of a bailout while the shaded gray areas represent 90 percent confidence intervals. Figure 5.6 shows the predicted probability of an ESF credit when systemic risk is low, held at one standard deviation below its sample mean (0.05). As the figure indicates, when systemic risk is low, there is almost no discernible effect of SIFI bank exposure on the probability of an ESF credit line. This suggests that when the risks facing the broader

Model 1

Model 2

Model 3

Model 4

Intercept

-11.820

(8.806)

-6.144

(8.723)

-4.629

(8.945)

-25.952

(19.513)

GDP (log)

0.295

(0.421)

0.106

(0.403)

0.124

(0.405)

0.074

(0.403)

GDP per capita (log)

-0.284

(0.672)

-0.466

(0.658)

-0.623

(0.634)

-0.526

(0.609)

External debt service/exports

0.032

(0.025)

0.029

(0.028)

0.034

(0.030)

0.031

(0.029)

Current account balance

-0.033^

(0.012)

-0.034“

(0.013)

-0.03Г

(0.013)

-0.03Г

(0.013)

Reserves/GDP

0.0382

(6.952)

-4.110

(7.039)

-2.871

(7.722)

-3.500

(8.499)

GDP growth

-0.033

(0.083)

-0.017

(0.090)

-0.014

(0.090)

-0.023

(0.091)

Speculative attack

1.142*

(0.569)

1.04T

(0.580)

1.093f

(0.632)

1.203*

(0.591)

Share of US trade

0.010

(0.244)

-0.094

(0.210)

-0.003

(0.305)

-0.046

(0.299)

Latin America

2.382**

(0.901)

2.4194'

(0.967)

1.935

(1.242)

2.017f

(1.197)

Democracy (PolityIV)

0.055

(0.067)

0.042

(0.067)

UN ideal point distance

-0.325

(0.864)

-0.262

(0.854)

US unemployment

1.387

(1.281)

US GDP growth

0.335

(0.250)

Republican president

1.586

(2.248)

Year

0.942

(0.747)

1.39T

(0.833)

1.367f

(0.806)

3.914

(2.786)

Year2

-0.089

(0.091)

-0.153

(0.097)

-0.149

(0.091)

-0.366

(0.283)

Year3

0.001

(0.003)

0.004

(0.003)

0.004

(0.003)

0.010

(0.009)

SIFIbank exposure

-0.023

(0.269)

-0.168

(0.229)

-0.187

(0.223)

-0.056

(0.277)

Systemic risk

-1.285

(1.792)

-3.305

(2.398)

-3.132

(2.433)

0.060

(3.528)

SIFI exposure * systemic risk

1.660'

(0.723)

1.886“

‘ (0.685)

1.991*

(0.815)

N

179

179

178

178

Number of countries

46

46

46

46

AIC

118.02

116.84

119.83

124.13

Figure 5.6

Predicted Probability of US Bailout (Low Systemic Risk)

Figure 5.7

Predicted Probability of US Bailout (High Systemic Risk)

Figure 5.8

Mean Predicted Probabilities of ESF Bailout (Model 4)

US financial system were low, SIFI exposure had little to no impact on US policymakers’ decisions to intervene on a distressed country’s behalf. For the sake of comparison, Figure 5.7 presents variation in SIFI exposure when systemic risk is elevated one standard deviation above its mean (0.65). The relationship is quite different in this case. When systemic risk is elevated, the effect of SIFI exposure on the probability of a US rescue is positive and quite substantial.

To further clarify this effect, Figure 5.8 presents the mean simulated predicted probabilities of a US rescue when systemic risk is low and elevated as SIFI exposure increases from its mean (0.9 percent of total foreign claims, slightly above the level of Turkey in 1999) to one standard deviation above the mean (2.8 percent, about the level of Argentina in 1984) to two standard deviations above the mean (4.7 percent, between the level of Korea in 1997 and Brazil in 1998). When systemic risk is low, increasing bank exposure from its mean to one and then two standard deviations above the mean has essentially no effect. Conversely, when systemic risk is high, increasing SIFI exposure from 0.9 percent to 4.7 percent of total foreign claims, the probability that Treasury will intervene as an ILLR and provide a bailout increases by a substantial 66 percent.[2] In short, the results indicate that US financial interests do correlate with ESF bailout selection. However, the results suggest that financial interests only impact US decisions to act as an ILLR when the private interests of major banks and the public interests that policymakers are mandated to protect are aligned. When risks facing the broader US financial system from international financial turmoil are high, policymakers are more sensitive to the exposures of major US banks and more inclined to respond to requests by the financial sector for protection. In such cases, policymakers may justify such actions as necessary to protect the broader national interests even though the banks reap a private benefit from the rescue operation. On the other hand, when the risks facing the system are low, ESF bailout selection does not appear to be substantially impacted by SIFI bank exposure. In such cases, policymakers appear to be far less sensitive to the banks’ private interests. The empirical results suggest that US policymakers’ sensitivity to the exposure of big Wall Street banks depends on the systemic context.[3]

Outside of the key financial covariates, the results indicate that countries with larger current account deficits and countries facing speculative attacks against their currencies were more likely to receive a US rescue. Thus, the results suggest that economic need played a role in the ESF bailout selection process. In all but Model 3, Latin American countries appear somewhat more likely to have been selected for an ESF rescue than countries in other regions. This implies that US policymakers may have been more sensitive to the needs of countries in their own geographic backyard. Neither the geopolitical variables nor the domestic economic controls are statistically significant at conventional levels.[4]

  • [1] This also allows me to avoid the problem of directly interpreting interaction coefficientsin nonlinear models. For more on this, see Ali and Norton 2003; Gelman and Pardoe 2007.
  • [2] At the elevated levels of SIFI exposure, the difference is statistically significant at thep<0.1 level.
  • [3] This is consistent with Oatley (2011), who cautions and demonstrates that within IPE,studies that focus on domestic politics “in isolation from international or macro processes”can generate inaccurate knowledge (p. 311).
  • [4] In separate specifications (not shown), I included a country’s share of US foreign aidas an additional control. The coefficient for US aid was negative and it was not statisticallysignificant. Additionally, its inclusion did not alter the statistical significance of the financialinteraction term. I omitted its inclusion here only because it resulted in the loss of 12 observations, including several cases where ESF bailouts were made. Given the small sample sizeand the scarce number of ESF credits in the model, I opted to exclude this from the analysis.
 
Source
< Prev   CONTENTS   Source   Next >