![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
ABSTRACT
Investor-state dispute settlement (ISDS) cases have increased dramatically in recent decades, and the institutions of investment arbitration to resolve investor-state disputes constitute the core of the modern investment regime. In this article, we seek to explain the variation in the host governments’ risk of being challenged by foreign investors before international arbitration tribunals. We argue that such risk is greatest at the intermediate level of democracy where some democratic elements such as elections are strong, but the system of rule of law remains weak. In those regimes, “regulatory risk” runs higher than in autocratic regimes as the host governments are under greater pressure for regulating matters of broad public concern. At the same time, more traditional political risk of arbitrary, discriminate, and abusive acts remains considerable at that level of democracy due to weak rule of law, exacerbating the former risk. Empirical analysis provides a good deal of support for the argument.
Supplementary material
Supplemental data for this article can be accessed on the publisher’s website.
Notes
1 There have been very few studies that look specifically at the determinants of investor-state dispute arbitration. Simmons (Citation2014) includes a brief statistical analysis of the investor-state dispute cases registered at the ICSID in her critical assessment of the current international investment regime, while Wellhausen (Citation2015a) uses investor-state disputes as an independent variable that captures expropriation risk. Freeman (Citation2013) is a rare exception.
2 Yet “the vast majority” of ISDS cases have come “to the public domain” at least in the sense that the very basic information on which government was sued and when ends up being publicly disclosed eventually, if not immediately (Poulsen and Aisbett 2013:19).
3 Raising a dispute against a host government might be particularly costly or even pointless when the host state is highly authoritarian, as the regime can simply ignore whatever awards the investor might win. Similarly, pursing a dispute against a highly developed democratic state might be seen as a losing proposition, as the presumably highly competent government is likely to prevail in the ensuing legal proceedings. If the aggrieved investor who suffers from a highly autocratic or highly democratic host government’s actions chooses not even to make a dispute out of these considerations, this would lead to an overstatement about intermediate regime’s political risk. We discuss this selection bias issue in the online appendix (Online Appendix A).
4 Saluka Investments BV (The Netherlands) v. The Czech Republic, UNCITRAL Arbitration, Partial Award, March 17, 2006, para. 255.
5 Singapore is perhaps the best-known counterexample for this, but so are the East Asian Tigers to a certain extent. In Latin America, too, Albertus and Menaldo (Citation2012) show that autocrats who create constitutions are more likely to uphold private property rights.
6 A testable implication of this assumption—that is, the more democratic the host state is, the more likely it is to be challenged for regulatory measures taken by it—will be empirically examined subsequently.
7 Legitimate regulatory measures taken according to domestic legal processes may be adopted and implemented, nevertheless, when there emerge strong popular concerns about the underlying issues, giving rise to occasional investors’ claims.
8 We elect random-effects models since 38 countries (about 45.2%) in the sample have all zeros on the dependent variable for the entire time period under study, and thus, those countries that have never had an ISDS case would all be dropped if fixed effects are used. This would leave us with a nonrandom sample consisting disproportionally of cases of high-dispute risk.
9 It is available beginning in the year of 1970 but only for every 5 years until 2000 from which its yearly data exist. Hence, we use 5-year moving averages to fill in those gaps in annual scores. Since rule of law is a system property that shows a degree of inertia, this interpolation would not pose a serious threat to valid inference. We use, however, three other measures to ensure the robustness of the results: the Rule of Law dimension from Worldwide Governance Indicators (Kaufmann et al. Citation2011), the Latent Judicial Independence (LJI) scores (Linzer and Staton Citation2011), and the Law and Order index from the International Country Risk Guide (PRS Group Citation2012:5). The results of these robustness checks will be made available in the online appendix (Online Appendix D).
10 There is evidence that strong checks and balances increase policy credibility, which then facilitates property rights protection (Zheng Citation2011). Firm-level FDI data also provide strong support for this connection (Vadlamannati Citation2012).
11 Similar to Checks, Polcon measures the number of institutional veto players but also takes into account their preference differences. While Polcon III includes the judiciary as a veto player, Polcon V does not. When either of these is used in place of Checks, no noticeable changes are made. These results as well as the ones without any measure of veto points are available in the online appendix (Online Appendix E).
12 The data for fuel, ores, and other metals exports along with other economic variables, unless noted otherwise, are taken from the World Bank’s World Development Indicators.
13 There might be a concern about multicollinearity when GDP growth, per capita GDP, and log GDP are all included together. We run models with only one of them at a time. No material changes are made. The results are available in the online appendix (Online Appendix F).
14 Our sample consists of only 84 non-OECD countries due to missing data. To guard against selection bias arising from missing data, we rerun our main models using multiple imputations. This enables us to recover 66 countries that are left out in the case-wise deletion method, making the total 150. The total number of observations increases more than threefold from 1,255 to 4,028. Yet the results show no material changes in our main findings. They will be made available in the online appendix (Online Appendix G).
15 The parallel results of the negative binomial regressions will be made available in the online appendix (Online Appendix B).
16 Since Polity2 is a composite measure that comprises three conceptual aspects of democracy—namely, executive recruitment (Exrec), executive constraints (Exconst), and political competition (Polcomp)—it would be interesting to see which components of Polity drive the outcome. Of particular interest would be to discern any differences the participation and competition elements (Exrec and Polcomp) and the executive constraints element may make in explaining the odds of ISDS as the theory can be interpreted as suggesting such a distinction. Thus, we test each component separately as well as the composite variable that combines Exrec and Polcomp (Reccomp). The results show that Polcomp is what primarily drives the outcome, while Exrec also seems to contribute to the pattern albeit insignificant of itself. On the other hand, Exconst has little to do with the inverted-U pattern; it alone tends to depress the likelihood of ISDS. The results are available in the online appendix (Online Appendix H).
17 The parallel results using count models will be made available in the online appendix (Online Appendix C).
18 We also add various measures of “history” to control for a host state’s past records with regard to ISDS: the lagged dependent variable, the cumulative number of past ISDS cases a host has faced, a host state’s past wins, and its losses of the cases. The Polity’s squared terms remain highly significant in all logit models but become weakly significant in most count models. The results are available in the online appendix (Online Appendix I).
19 We test the hypothesis that democracy leads to the rule of law using the same data set as used in our main empirical analysis. Again we employ the four measures of rule of law (dependent variable) as well as the four measures of democracy (independent variable) to guarantee the robustness of the findings. For control variables, we consider veto points, log GDP per capita, trade openness, and resource endowments. Overall, the results overwhelmingly confirm the findings of previous studies that democratic institutions help to consolidate institutions of rule of law. These results will be available upon request.
20 It is worth mentioning that rule of law serves as a mediator, not a moderator, in our theory. In the latter case, the effects of democracy should vary depending on the level of rule of law, and to test such relationships, we should include an interaction term between democracy and rule of law in addition to both of the variables at once. While we acknowledge that such an interaction theory is not infeasible, that is not what we propose and test here. We maintain that rule of law is causally closely related with democracy and that the outcome of our interest is affected in part by a change in rule of law that is caused by an increase in democracy. Baron and Kenny (Citation1986) is a classic piece on the distinction between moderator and mediator variables.
21 We repeat this probe using three other measures of rule of law. The overall results, taken together, are rather weak and inconsistent across all models with different measures to prove the intervening role of rule of law. Nonetheless, the constraining effects on the dependent variable of all the measures of rule of law used are generally highly significant, and there is some evidence that its addition weakens the inverted-U pattern that democracy has in relation to the dependent variable. These results will be made available in the online appendix (Online Appendix D).
22 We also fit ordered logit models assuming that the Risk Types variable is an ordinal variable as well as multinomial logit models treating it as a nominal variable with three categories. Both models yield largely similar results, which are available in the online appendix (Online Appendices J and K).