Borrowing Trouble: Sovereign Credit, Military Regimes, and Conflict

ABSTRACT

Changes in relative military power in the international system are seen as an impediment to peace. This article will focus on one particular avenue for states to increase their relative military power: sovereign borrowing. States’ ability to borrow inexpensive credit can undermine credible commitments in international relations, but only for those states that habitually use credit for military purposes. I argue that military regimes are more likely to use fiscal resources such as sovereign credit toward military spending, which leads to sudden increases in military power. As a result, adversarial states have incentives to use preventive action against military regimes before these regimes use credit for military purposes. To test this argument, I examine target behavior in militarized disputes as a function of expected borrowing costs credit and regime type. The empirical analysis demonstrates that military regimes, expected to have improved borrowing costs, are more likely to be the target of militarized disputes.

KEYWORDS:

• Fiscal resources
• military regimes
• preventive war
• sovereign credit

Acknowledgments

I thank Matt DiGiuseppe, Justin Esarey, Songying Fang, Justin Kirkland, Ashley Leeds, Cliff Morgan, Paul Poast, Mark Souva, Ric Stoll, three anonymous reviewers, and the editor of International Interactions for valuable feedback and comments. A previous version of this paper was presented at the 2014 Midwest Political Science Association Annual Meeting in Chicago, IL, and the Rice University IR Workshop in Houston, TX.

Supplemental Material

Supplemental data for this article can be accessed on the publisher’s website.

Notes

¹ Military spending estimates are drawn from Singer (1987).

² Paler (2013) finds that taxes motivate citizens to monitor governments’ fiscal policies.

³ See DiGiuseppe (2015) for an analysis of the connection between sovereign credit and military spending.

⁴ There are examples of states’ secret borrowing, like Nazi Germany in the 1930s, but these examples demonstrate forced borrowing that is akin to taxation.

⁵ DiGiuseppe and Shea (2015) find that the use of credit is more important for nondemocratic leaders’ political survival.

⁶ Related to this, Kono and Montinola (2013) find that nondemocratic regimes will allocate foreign aid toward military spending.

⁷ Shea (2014) examines the benefits of credit when states enter war. Democracies enjoy more political benefits from credit, and thus better credit prices are associated with better outcomes for democracies. The argument in this article is concerned with whether certain regimes, particularly military regimes, are more likely to use credit for war purposes. If military regimes efficiently use credit toward military purposes, it would suggest that military regimes with low credit costs would do better in war. However, other states have incentives to disrupt a military regime’s creditworthiness before they can effectively use cheap credit for war purposes. As a result, we do not expect to observe military regimes with credit advantages in war samples.

⁸ Future research opportunities should consider the effect of dyadic dependence on the relationship between credit and conflict.

⁹ From the COW coding, a state is coded as a target if Side A = 0, and Orig = 1. One issue that could possibly complicate this coding is if these military regimes are targeted in time t in response to an action initiated by the military regime in either time t or time t−1. In a robustness test discussed later, I find that this is not the case.

¹⁰ Bond yields are commonly used as an indicator of creditworthiness (Dincecco 2009; Mauro, Sussman, and Yafeh 2006).

¹¹ These bonds are for governments’ domestic debt securities that are traded on a secondary market. The yield is calculated by the Global Financial Database by dividing the price of individual bonds by the coupon price. A reliability problem can arise when bonds mature because the risks to bond owners fall, decreasing bond yields. However, governments issue and retire new bonds on a somewhat regular basis. When the government issues a new bond of the stated maturity (10 or five years), it replaces the bond used for the index to keep the maturity as close to the stated time period as possible.

¹² This identification of adversaries is similar to how Leeds and Savun (2007) identify international threats.

¹³ As a robustness test, I use Cheibub, Gandhi, and Vreeland’s (2010) measure of military regimes, which has a wider conceptualization of military regimes. These results are found in the Web appendix and are consistent with the main results.

¹⁴ Note that the histogram for the nonmilitary regimes in limited to the range (–3, 3) in order to provide a comparison to the military regimes. One difference between the changes in bond yield data for military regimes and nonmilitary regimes is the range of the data. The range for changes in bond yields for military regimes is [–2.12, 1.94] while the range for changes in bond yields for nonmilitary regimes is [–72.59, 85.96]. The Web appendix examines how this larger range for nonmilitary regimes affects the results. I also explore whether certain observations (for either military or nonmilitary regimes) have too much influence on the results. I find no evidence that either scenario poses a problem to my inferences.

¹⁵ Measures of growth over a single year may mask instability in a state’s economic market (large declines followed by large upswings). This type of instability leads to lower long-term growth and discourages foreign investment (Martin and Rogers 2000). However, using a growth rate measure for a single year does not substantively change the main inferences. These alternative models can be replicated through the supplemental data materials.

¹⁶ If a state has never experienced conflict, this variable counts the number of years since it entered the data.

¹⁷ I interact Military Regime with changes in bond yields rather than level of bond yields because the latter signify an improvement in a state’s creditworthiness and change in the military regime’s future capabilities. Low levels of bond yields signify existing low borrowing costs, which may already have resulted in increased capabilities. Once capabilities have already increased, the incentives for other states to carry out a preventive strike is mitigated (Bell and Johnson 2015).

¹⁸ Note that models 1 and 2 in Table 2 have similar goodness-of-fit statistics (log-likelihood and AIC). However, this does not mean that the second model is not a better “fit.” Hosmer, Hosmer, Cessie, and Lemeshow (1997) examine several goodness-of-fit statistics using Monte Carlo simulations when logit models contained interactions between a continuous and binary variable. The authors found that all goodness-of-fit statistics had low power in determining the true value of the interaction model. In addition, the collinearity between the interaction term and the additive components of the interaction may limit the interaction term’s contribution to the goodness-of-fit of the model. However, without the interaction term, the hypothesis test would be theoretically underspecified.

¹⁹ I also consider different baselines of conflict in the Web appendix. Rivalries, allies, territorial disputes, and number of neighbors all affect the proclivity of conflict. I include these additional covariates in the analysis in the Web appendix and find similar results.

²⁰ This null result can be found in the replication materials.

²¹ However, the correlation between missing bond data and being targeted is low, with a correlation coefficient of only 0.05.

²² For example, see Chenoweth (2010), Fordham (1998), Gowa (1998), Hendrix and Salehyan (2012), McDonald (2007), Pollins (1996), Senese (1997), and Souva and Prins (2006).