Data and Econometric Approach
In the analysis, the additive Index of Government Coerciveness is used as a dependent variable so that our sample comprises crisis countries only. Specifically, we use the agreement-based dataset described by Enderlein et al. (2010). The time frame is 1980 to 2005, a period for which we coded 103 individual debt restructurings in thirty-six developing countries. Each of these agreements has an individual index value indicating how many of the nine subindicators of coerciveness have been fulfilled during the restructur?ing process. In other words, the dependent variable captures the degree of coercive actions that governments impose on their foreign creditors in the negotiations preceding each agreement. The minimum index value is 1 (very cooperative debtor behavior) and the maximum is 10 (very coercive). Across all 103 cases, the average index value is 3.58. Interestingly, the standard deviation is quite high (2.01) underlining the large variability in debtor behavior across countries and time.
To test our hypotheses systematically, we use the governance indicators by Kaufman et al. as key explanatory variables. More specifically, we group the indicators as follows: Hypothesis 1 will be tested based on the indicators “Government Effectiveness” and “Regulatory Quality” Hypothesis 2 is tested using the “Rule of Law” and “Control of Corruption” indicators. Hypothesis 3 is based on the measure of “Voice and Accountability,” while Hypothesis 4 is tested using the indicator on “Political Stability and Absence of Violence” Higher values for each of the indicators reflect a higher quality of governance.
The governance indicators were first released for the year 1996 and are available on an annual basis only since 2002. Given these constraints, we simply construct average values of each of the six indicators based on the available data from 1996 to 2005. The rationale for using these time-invariant average indicator values is that we are mostly interested in cross-country differences in governance levels. This will unveil general empirical regularities, while we explicitly disregard within-crisis dynamics. However, we do include year fixed effects to capture some of the variation in our dependent variable and to improve the preciseness of estimation.
Given the categorical character of our dependent variable (ranging from 1 to 10), we estimate an ordinal probit model. Ordinal probit models are a generalization of the simple two-outcome probit model. In a first step, we therefore regress our coerciveness index for each of the restructurings on the six governance indicators and a set of cutpoints.