1 Introduction

To forge stable international cooperation, states increasingly rely on legalized institutions and especially on treaties, which have proliferated in recent decades (Goldstein et al. 2001; Alvarez 2002). Given that their main function is to render obligations more explicit and to generate enduring commitments (Abbott and Snidal 2000; Simmons 2000), it is notable that many treaties change over time through various forms of renegotiation. Examples abound. In the security area, the North Atlantic Treaty has been amended repeatedly with formal protocols that expand its membership. In trade, the European Community signed an agreement with a group of African, Caribbean, and Pacific countries in 1975 but revised it in 2000. The revised treaty, known as the Cotonou Agreement, was renegotiated again in 2005 and yet again in 2010. Environmental agreements are often changed: the 1987 Montreal Protocol on ozone depletion has been “adjusted” five times and “amended” four times, and was itself an update to the 1985 Vienna Convention.

Why are some international agreements changed through renegotiation while others remain stable? The international relations literature has not addressed this question with much depth or rigor, despite a growing interest in the politics of international law. The influential studies of Koremenos (2001, 2005) provide a theoretical framework for explaining why some agreements contain limited-duration and renegotiation provisions while others are rigid and indefinite, making change more difficult. Related to this, various scholars have explored the general theoretical question of why states conclude agreements with flexibility to allow obligations to be adjusted over time (Rosendorff and Milner 2001; Helfer 2012). However, none of these studies takes the additional step of asking under what conditions states actually take advantage of their ability to renegotiate international treaties.

We argue that states are motivated to renegotiate when they have acquired new information about the consequences of their treaty commitments. We develop and test this learning logic in the context of bilateral investment treaties (BITs). Having proliferated over the last few decades, the network of almost 3000 BITs now forms the backbone of international investment law. However, some governments have grown dissatisfied with their original BITs and close to 200 have been renegotiated, a trend that began in the 1990s and has gained momentum in recent years. With its recent announcement that it would seek to terminate or renegotiate more than 50 investment treaties, India is the latest in a series of countries, developing and developed, to overhaul their BIT policies.Footnote 1 Indeed, the politics of investment agreements are more contentious than ever. This is illustrated by the negotiations over the investment chapter of the Transatlantic Trade and Investment Partnership, aimed at “improving and updating” the parties’ existing BITs (European Union 2015, 41), which have provoked heated debate on both sides of the Atlantic.Footnote 2 Similarly, perhaps the greatest obstacle to ratification of the Trans-Pacific Partnership has been its investment chapter (Nottage 2016).Footnote 3

Beyond their political significance, BITs provide a nice laboratory for studying treaty renegotiation. Almost all BITs include a duration provision that makes them eligible for withdrawal or renegotiation after a certain number of years, allowing us to control for this critical element of treaty design.Footnote 4 Moreover, BITs give us substantial variation across countries—almost every country in the world has signed one, representing all regions—while controlling for membership size, which is likely to affect the attractiveness of renegotiation.Footnote 5 Because the vast majority of international agreements are bilateral, BITs also have the virtue of being more typical and representative of treaties generally.

The next section reviews the literature as it relates to the question of agreement renegotiation and institutional change more generally. The third section presents our main hypotheses, which focus on the impact of learning on a state’s propensity to renegotiate treaties. Specifically, we propose that BIT parties are more likely to renegotiate when they are exposed to more investment disputes, which supplies new information about the precise consequences of their treaty obligations. A research design section introduces our data, including an original dataset on renegotiated BITs, and outlines our main explanatory variables and additional variables that capture alternative logics for renegotiation. Using a variety of model specifications, we find strong support for the role of learning, controlling for a wide range of other political and economic factors. We conclude by discussing implications of our study for the politics of investment agreements and the design and renegotiation of international treaties more generally.

2 Change in international agreements

Reflecting broader debates on institutions, political science work on international agreements has focused mostly on the motivation behind their conclusion and design, with competing logics of efficiency and power (Koremenos et al. 2001; Thompson and Verdier 2014; Krasner 1991), and their subsequent effects on compliance and other outcomes (Goldstein et al. 2007; Mitchell and Hensel 2007; Simmons 2009). Renegotiation and other forms of change have received much less attention. Most scholars are content with taking snapshots at a given point in time and do not explore the dynamic nature of institutions and the factors that lead states to seek adjustments, issues that remain theoretically and empirically underdeveloped (Fioretos 2011). Precisely because conventional approaches treat institutions as relatively stable and enduring outcomes, instances of significant and deliberate change are especially interesting and worthy of investigation (Weyland 2008; Jupille et al. 2013; Haftel and Hofmann 2017).

The same gap exists in the burgeoning literature on international investment agreements, which includes a number of sophisticated studies on why states conclude them (Elkins et al. 2006; Jandhyala et al. 2011; Poulsen 2015) and on how agreements, once implemented, impact investment behavior (Allee and Peinhardt 2011; Büthe and Milner 2008; Haftel 2010). The issue of why states sometimes change treaty terms through renegotiation has not been studied systematically, despite an interest in the evolving politics of the investment regime and the detailed provisions of BITs (Blake 2013; Allee and Peinhardt 2010; Alschner and Skougarevskiy 2016). This is an important omission given that many governments have expressed dissatisfaction with existing agreements and that the landscape of BITs is “undergoing a period of reflection, review and reform” (UNCTAD 2014, 2). Renegotiation is a key part of these reform efforts. Unlike more dramatic policy options, such as withdrawal from treaties and related institutions, renegotiation is designed to adjust rules through mutual agreement and thus to change the regime rather than undermine it.

Although not directly concerned with renegotiation, there is important research on the related question of why states sometimes prefer treaties that are easy to modify or exit. Studies in international relations and law have focused in particular on the role of uncertainty—about future political conditions and the distributive consequences of an agreement—in creating incentives to design agreements in flexible ways (Koremenos 2001, 2005; Rosendorff and Milner 2001; Raustiala 2005; Thompson 2010; Helfer 2012). As leaders learn more about the costs and benefits of their initial commitments, a flexible agreement can be adjusted accordingly. This echoes work on the value of adaptive governance structures in complex or uncertain environments, where the implications of rules are difficult to anticipate (Dietz et al. 2003; Cooney and Lang 2007). These insights have been applied specifically to investment agreements, where the ambiguity of investment law and risk of political shocks means that states prefer treaties with some flexibility and finite durations to allow for renegotiation and other forms of adjustment (van Aaken 2009; Pauwelyn 2014).

Although our goal is to explain renegotiation rather than agreement design,Footnote 6 the literature on institutional flexibility provides a fruitful theoretical starting point with its emphasis on the importance of uncertainty and updating. International treaties, like much domestic law, contain ambiguities that are only clarified over time as the rules are implemented and enforced (Suchman and Edelman 1996). This leaves significant room for learning, which, in turn, may create incentives for renegotiation. In the next section, we build on this logic to generate specific hypotheses on BIT renegotiation.

3 Learning and BIT renegotiation

We expect treaties to be renegotiated when governments acquire new information about the desirability of their existing treaty commitments. In the case of BITs, much of this learning centers around the meaning and legal implications of treaty provisions, both substantive and procedural, and the resulting political consequences—namely the benefits of appealing to investors measured against some loss of autonomy in important areas of public policy. These treaties were often poorly understood at the time they were originally concluded, leaving significant room for learning and incentives for adjustment.

Ignorance of the legal implications of BIT provisions existed for two main reasons. First, under the constraints of bounded rationality, policymakers sometimes failed to thoroughly consider the implications of alternative design provisions. For developing and transition economies, especially in the 1990s, there was a rush to sign BITs by governments seeking to remain competitive and eager to embrace the perceived standards of liberal reform (Elkins et al. 2006; Jandhyala et al. 2011). In this context, Poulsen (2014, 2015) and Poulsen and Aisbett (2013) provide evidence that negotiators from the developing world often relied on shortcuts and did not systematically assess the implications of specific provisions. A review of South Africa’s BIT policy, for example, notes “a lack of understanding regarding the real nature and consequences of BITs at that time [the mid-1990s]” (DTI 2009, 14). Negotiators from Eastern Europe faced similar challenges, sometimes pursuing BITs hastily with little experience and limited knowledge of their content and potential impacts (Fecák 2011).

A second source of ignorance—and subsequent learning—stemmed from the nature of the regime itself, which was based on evolving rules and a complex institutional landscape. BITs often contain broadly worded obligations, leaving host states unsure of their regulatory powers and dependent on arbitration to provide interpretations. In other words, there was some inherent uncertainty at the time of signing because BIT rules were only clarified with time and experience as the regime evolved (Schill 2009; Brown and Miles 2012; Pauwelyn 2014). Moreover, because of the overlap across investment agreements and conflicts with other institutions in the broader regime, such as regional trade agreements and taxation treaties (Alschner 2014; Bonnitcha et al. Forthcoming), the practical and legal implications of BIT rules were not always apparent initially.Footnote 7 Thus even high-capacity and fully rational governments failed to anticipate the range of consequences of earlier investment agreements and have sought to assess and implement lessons learned from their experience with them.Footnote 8

This learning has brought a better understanding of the costs and benefits that flow from BITs and a desire in some cases to “re-balance” their terms, often with greater precision (UNCTAD 2011; Manger and Peinhardt 2017; Henckels 2016). There is now more concern over the substantive provisions of BITs and how they interact with binding arbitration procedures, which have been influential and produced unanticipated results for some governments (UNCTAD 2010). For instance, the renegotiated China-Switzerland BIT, which entered into force in 2010 (replacing its 1986 predecessor), circumscribes which investors have standing to bring arbitration claims—for example, to reduce the possibility of using subsidiaries to “treaty shop” for the greatest protection—and more narrowly defines how the “national treatment” standard should be applied.Footnote 9 For its renegotiated BITs, India is attempting to carve out more regulatory space by including exceptions for politically sensitive areas of economic and social policy (Singh and Ilge 2016). As policymakers have received more information about the implications of BIT provisions for investor rights and government control, they have adapted accordingly by shifting their approach in new BITs and renegotiating old ones when possible.

Although some learning takes place on a routine basis among officials who specialize in investment policy, we propose that one aspect of BIT practice is especially informative to political leaders in government: investor-state dispute settlement (ISDS). In general, rulings made by dispute settlement bodies help to interpret ambiguous or contested treaty provisions and thereby clarify their meaning and consequences (Goldstein and Steinberg 2008). Because of its fragmentation and the broad standards in many treaty texts, the role of interpretation has been especially important in the investment regime, with ISDS tribunals playing a “critical role” (Roberts 2010, 179). Schill (2009) points to ISDS as the primary mechanism for resolving the uncertainties surrounding BITs and van Aaken (2009, 532) refers explicitly to the “learning effect” of arbitration, which has allowed states to approach BITs with more information—and with more scrutiny and caution.

There is anecdotal evidence of this learning process in several important cases. In response to a series of ISDS claims in the late 1990s and 2000s, Canada updated its “model” BIT in 2004 and renegotiated several existing BITs in the direction of providing more room for social and economic policy objectives. Similarly, Indonesian officials explained their intention to renegotiate BITs by referencing the number of arbitration claims they were facing, which led the Indonesian government to realize that “some of its treaties are not beneficial for it” (OECD 2014, 9). Studies of South Africa’s investment policy reforms show explicitly that it was exposure to international arbitration—specifically, a claim brought by European investors over “black economic empowerment” requirements in the mining sector—that led the government to re-evaluate its approach to BITs (Schlemmer 2016; Poulsen 2015). Even without losing the case (which was settled), South African policymakers learned from the experience and saw potential implications of their existing BITs for development plans and social goals. These examples illustrate a key piece of information derived from arbitration: The extent to which BIT rules extend to the domestic realm and the resulting boundaries of regulatory space available to host governments.

To be clear, lessons from ISDS are not always negative. After observing the dramatic rise in investor-state disputes in the late 1990s, China responded with a new generation of BITs that granted investors higher standards of protection and access to binding international arbitration for the first time (Gallagher and Shan 2009). As a growing sender of FDI, Beijing saw the virtues of greater investor protection and the value of ISDS. To protect their investors, home governments sometimes seek to strengthen protections in agreements they deem too weak, even if the same rules are used against them in some cases. For example, commenting on an important dispute involving Germany as a defendant (Vattenfall v. Germany), German officials claimed that, “Overall, the German government’s experience with the international dispute settlement in this case was positive” and “stressed the benefit of the international arbitration mechanism in creating an environment in which the disputing parties face strong incentives to find constructive solutions of their disputes” (OECD 2011, 5). This view helps explain why many renegotiated German BITs introduce strong ISDS provisions, replacing older BITs that lack them.

Although policymakers can derive lessons from the experience of other countries, we assume that the most important learning takes place as a result of a government’s direct exposure to arbitration. Received wisdom tells us that organizations and individuals respond more to information that is salient and readily available, enhancing the importance of direct experience (Levy 1994; March 2010). In the case of investment disputes, when host governments are targeted by a claim this raises the specter of a costly award and mobilizes interest groups in defense of the policy measure in question. The political salience of these events is likely to draw the attention of political leaders at the highest level, not just specialized bureaucrats, provoking a concerted assessment of existing BITs. This helps explain why most governments did not systematically examine their BIT obligations until they were targeted by ISDS claims (Poulsen and Aisbett 2013). To assess the role of learning through direct exposure to arbitration, we propose the following hypothesis:

Learning Hypothesis (Respondent): States involved in more ISDS cases are more likely to renegotiate their BITs.

Of course, to the extent that information diffuses internationally, states may also learn from ISDS cases that do not involve them. Perhaps a general increase in ISDS activity promotes learning across the system. In a variety of domains, including investment, governments have shown they can derive policy lessons from salient examples in other countries (Weyland 2008; Elkins et al. 2006). The international diffusion of policy lessons faces a variety of obstacles (Gilardi 2010; Solingen 2012) and the relative opacity of investment arbitration limits the flow of information to uninvolved parties. Nevertheless, it is certainly possible that direct experience is not necessary and that disputes have a more general effect by raising awareness among policymakers around the world of the legal and political implications of BITs. To consider these broader learning dynamics, we propose a second hypothesis:

Learning Hypothesis (Global): When there are more dispute initiations world-wide, states are more likely to renegotiate their BITs.

4 Research design

We employ complementary statistical methods to test the propositions presented in the previous section. We first use a Cox proportional hazard model with robust standard errors to estimate the effect of the independent variables on the duration of treaties before they are renegotiated.Footnote 10 This type of survival analysis estimates the “risk” that an event will take place as time elapses and is thus especially useful to account for variation in the timing of events. This method is not well-suited to investigate the effect of time in and of itself, a factor we are substantially interested in. We address this shortcoming with a probit model with a cubic polynomial approximation, which accounts for temporal dependence (Carter and Signorino 2010). Standard errors are robust and clustered by dyad.Footnote 11 This model estimates the probability of BIT renegotiation. Finally, BIT renegotiation occurs rather infrequently: only in 0.6% of the observations in the data set. We address this potential concern with rare event (RE) logistic models (King and Zeng 2001). The three statistical techniques produce similar results, suggesting that our findings are robust to alternative methods. To reduce the risk of endogeneity, all time-varying independent variables are lagged one year. In addition, as discussed below, we include a spatial lag to account for potential diffusion effects as well as various time lags to account for longer temporal dynamics. We now turn to a description of the dependent and independent variables. Tables A1 and A2 in the online Appendix report summary statistics and bivariate correlations, respectively.Footnote 12

4.1 Data and dependent variables

Our data set includes 2043 mutually-ratified BITs from 1962, the year in which the first four BITs entered into force, to 2007.Footnote 13 The unit of analysis is the treaty-year: each BIT includes an annual observation from the year in which it had entered into force until the year in which it was substituted by a revised treaty or up to 2010 (if the original BIT remained intact). To account for the possible effect of the initial duration of the treaties, which is ten years for most BITs, we repeated the analysis while excluding the first nine years after the treaties entered into force (during which about 25% of renegotiations took place). There were no major changes in our results.

The main dependent variable—labeled Renegotiate—is a categorical variable that scores one if the treaty was renegotiated at year t, and zero otherwise. The coding of this variable requires a list of all renegotiated BITs and the year in which they were signed. We assembled a new data set of all renegotiated BITs for this study. Given the originality of this data, we elaborate on the definition of renegotiation and present some descriptive statistics.

We define a renegotiated BIT as a signed treaty that replaces an existing mutually ratified BIT or a signed amendment to an existing mutually-ratified BIT.Footnote 14 Renegotiation can take two primary forms. First, governments may keep the old treaty in place but amend it with a new protocol. In such cases, the changes to the treaty commonly address a small number of specific issues, which may nevertheless be quite important to the parties. For example, many of the Czech renegotiations took the form of adding a protocol to an existing BIT. While some of the provisions in these protocols address matters related to the Czech accession to the EU, others have far-reaching implications. In particular, many Czech protocols include a provision that allows the parties to take action to protect their “essential security interests.” In another example, the 2007 Sweden-China Protocol to their 1984 BIT specifies ISDS rules, which were completely absent from the original treaty. Second, the parties to an existing BIT may sign a new investment treaty and add a clause that terminates the prior BIT. In such instances, parties typically overhaul the entire agreement.

With these criteria, using UNCTAD records and government websites, we created a comprehensive list of all renegotiated BITs, which total 177. About 60% of the renegotiations are new investment treaties and the rest are amending protocols. On the whole, we believe that the data set provides an inclusive and accurate description of renegotiated BITs. A glance over the data provides a number of insights into the phenomenon of BIT renegotiation. First, this is a rather new trend and is becoming more significant over time. The first renegotiation we identify is a 1986 amendment to the 1974 France-Egypt BIT. Renegotiation became more widespread in the mid-1990s with the renegotiation of four treaties in both 1994 and 1995. As Fig. 1 shows, the number of renegotiated treaties has increased steadily since then, reaching a peak of 22 in 2009 and 2010.

Fig. 1
figure 1

The Annual Number of Renegotiated BITs, 1986–2010

Full size image

Close to a hundred countries from all continents were involved in BIT renegotiation. Some countries, such as Guinea and the Philippines, have renegotiated only one treaty. Others have renegotiated BITs recurrently, such that fifteen countries account for more than half of these outcomes (see Fig. 2). The “champion” is Romania with close to 40 renegotiated BITs. It is followed by the Czech Republic, Germany, China, and Egypt, each of which renegotiated at least 15 BITs. Interestingly, the top renegotiators come from a number of regions, including Europe, the Middle East, and East Asia.

Fig. 2
figure 2

Number of Renegotiated BITs for the Top 15 Renegotiating Countries

Full size image

We examine two additional actions taken by states that are akin to BIT renegotiation. First, the parties may substitute a BIT with a free trade agreement (FTA) that includes an investment chapter. Although the dynamics of renegotiation may be quite different in such instances, given the greater political salience of FTAs, the outcome is usually similar to a BIT renegotiation. We have identified only thirteen cases of such renegotiation by 2010, mostly by countries in the Western Hemisphere. Second, states may opt to terminate a BIT rather than renegotiate it. Parties can either unilaterally denounce and withdraw from a BIT after its initial duration or terminate it by mutual consent. Here the outcome is somewhat different from BIT renegotiation, but the dynamics described in the previous section probably hold. A government that learns from arbitration may prefer to terminate rather than renegotiate the agreement (especially if the partner disagrees about renegotiation). We have counted 31 instances of termination by 2010, split in half between unilateral denouncements and mutually-consented terminations, and present the results in auxiliary models. Both actions were rather uncommon until recent years, however their growing number and political significance call for more investigation in future research. We now turn to the explanatory variables.

4.2 Explanatory variables

Our primary hypothesis suggests that involvement in ISDS is likely to prompt treaty renegotiation by revealing new information to those countries that take part in such proceedings. We conceptualize this variable with Dispute Respondent, which tests whether states that are taken to arbitration by foreign investors in the previouse year are more likely to renegotiate their BITs.Footnote 15 It is a count of all new investment claims filed against both governments in a given year. For example, in 2005 investors filed four new complaints against Russia and none against the United Kingdom (UK). In the following year, the UK and Russia faced one new claim each. The values for this dyad are therefore four and two for 2005 and 2006, respectively. This variable captures the idea that governments respond to new information that emerges from their direct experience with ISDS. Government responses and treaty renegotiation may require more time, however. We account for this possibility with two variables, Dispute Respondent t-3 and Dispute Respondent t-1 – t-3 . The former is a three-year lag of the original variable and the latter sums up its values over the previous three years, thereby testing the effect of cumulative ISDS experience over several years.Footnote 16

A second dispute variable, labeled Dispute Global, is the total number of new disputes worldwide in a given year. It captures the logic of the second hypothesis, that governments react to and learn from the experiences of other countries.Footnote 17 The value for the UK-Russia (and every other) dyad in 2005 is forty-three, which is the number of new investment claims filed that year worldwide. The dispute variables are coded with data compiled by Allee and Peinhardt (2011) and Dupont and Schultz (2013) and include 538 cases through 2010.

4.3 Alternative logics and control variables

To isolate the effect of our proposed learning dynamic, we include a variety of additional independent variables in the analysis. Some of these are designed to capture changes in circumstances that could render old BITs outdated, while others capture changes in the political and economic characteristics of states that could influence preferences vis-à-vis BITs. We also control for factors identified in the literature that make states more or less likely to conclude BITs and treaties generally.

When it comes to the desirability of existing BIT, the range of relevant circumstances is expansive and each country faces its own, possibly idiosyncratic considerations. For purposes of our cross-national empirical test, we focus on general factors that could affect a range of states. First, because conditions change more as time passes, we expect older treaties to be more outdated on average and thus more prone to renegotiation. Older BITs that lack certain features, such as binding arbitration, may be viewed as inadequate and are more susceptible to renegotiation (Vandevelde 2009, 29). The talks between Germany and Pakistan regarding the replacement of the first-ever BIT illustrate this point. Explaining the rationale, one German diplomat reported that German investors believed the old treaty “does not reflect newer developments and international standards,” including ISDS provisions.Footnote 18 We capture this with an annual counter variable, labeled Time, in the probit and rare event logit models.

Second, EU accession is an important political and legal change for several states with active BIT policies. In some cases there are conflicts between existing BIT obligations and EU law and the European Commission has urged new members to scrap or renegotiate their existing BITs (Olivet 2013). We account for this with the variable New EU Member, which scores one for BITs that involve at least one country that joined the EU in the 2000s from the year of accession onwards.

Third, we consider changes in domestic political circumstances. Shifts toward democracy could affect preferences toward investment and BITs. In general, we expect democratic governments to be more sensitive to public and interest-group concerns and more jealous of their policy autonomy in areas such as social regulation and sustainable development, which are often restricted under investment rules. In addition, moves toward democracy are likely to entail a strengthening of institutions and rule of law, which could reduce the need to attract investors with restrictive international agreements (Stasavage 2002; Jensen 2006). We capture democratization with Democratic Transition, which scores one if at least one party is going through such a transition in a given year. We consider a state to have gone through democratic transition if it experiences at least a three-point increase in its Polity score over a period of three years or less (Marshall et al. 2010, 35–36).

Changes in the ideological orientation of governments could also drive a desire to renegotiate. Partisanship is correlated with policy preferences regarding foreign capital and FDI in particular (Pinto 2013; Brooks and Kurtz 2007), giving leaders an incentive to alter BITs concluded by predecessors of a different ideological orientation. The variable Party Change scores one if at least one country experienced a shift in the ideological orientation of the government, from right to left or from left to right, since the BIT’s entry into force. We employ the World Bank’s Database of Political Institutions (Beck et al. 2001)—which codes the executive’s party as right, center, or left—to operationalize this variable.

We also consider changes in the economic positions of the parties. As developing and transition economies grow and become more attractive destinations for FDI, they rely less on highly restrictive BITs and may improve their bargaining position relative to home states (Urpelainen 2011). We capture the shifting economic relationship between the two partners with ΔGDPPC Gap. This variable is measured as the cumulative change of the difference in logged GDP per capita since the BIT’s entry into force. We use Penn World Tables 7.0 GDP data to generate this variable. To account for changing patterns of actual investment flows, we also test for the cumulative change in the host country’s FDI inflows as a percentage of GDP since the BIT’s entry into force.Footnote 19 Data for this variable, labeled ΔFDI Flows, is provided by UNCTAD’s UnctadStat data base.Footnote 20

We control for two additional factors that might influence the propensity of states to renegotiate. Recent studies indicate that countries with a common law system are less likely to sign or ratify international agreements than countries with other legal traditions.Footnote 21 This could apply to treaty renegotiation as well. We control for this possibility with Common Law, a dichotomous variable that scores one if at least one party has a common law system and zero otherwise.Footnote 22 Next, countries with shared history and institutions may find it easier to renegotiate their agreements. We account for this possibility with the dyad’s shared colonial heritage, another variable that is often included in research on international institutions. Colonial Ties scores one if the BIT parties share a common colonial history, and zero otherwise.Footnote 23

Finally, the propensity to renegotiate may be driven by a country’s own previous action or the behavior of its peers. We account for such dynamics with three variables. First, states that renegotiated in the past may be more likely to do so in the present. Indeed, as noted above, a relatively small number of countries appear to dominate the trend of renegotiation. We address this issue with Lagged Renegotiated BITs, which is a one-year lag of the cumulative number of renegotiated BITs. Along similar lines, governments may renegotiate more treaties simply because they have a larger pool of BITs in force available for renegotiation. Lagged BITs in Force is thus a one-year lag of the number of BITs in force for a given state.

Elkins et al. (2006) show that the propensity of a government to sign BITs hinges on the actions of its peers. A similar diffusion dynamic may take place with respect to their renegotiation. We account for this possibility with a spatial lag. For any given country, this variable is the sum of all renegotiated BITs of all other countries (lagged one year) weighted by the geographic distance between the two capitals. Spatial Lag ranges from zero to one, and increases as a state’s closer neighbors renegotiate more BITs. For all three variables, we take the higher value in the dyad, assuming that the country with a more robust BIT program, greater experience with renegotiation, or neighbors that are more enthusiastic about renegotiation is more likely to determine the outcome.

5 Results and discussion

Table 1 presents the results of six event history models estimating the effect of the independent variables on the time that lapses until renegotiation. With the exception of Model 2, all models examine the effect of being a respondent to a dispute on renegotiation. Model 1 uses the Dispute Respondent variable with a number of basic controls. Model 3 adds Party Change and Model 4 replaces lagged renegotiated BITs with lagged BITs in force. Models 5 and 6 consider alterations of the main independent variable; the former is a three-year lag of Dispute Respondent and the latter is the sum of its values over the previous three years. Model 2 examines the effect of Dispute Global on renegotiation.Footnote 24 The table reports the hazard ratio, which estimates the hazard rates for different values on the independent variables. A hazard ratio that is greater (smaller) than 1 indicates a higher (lower) probability of renegotiation, such that, as values on the independent variable increase (decrease), the likelihood of survival decreases (increase).

Table 1 Cox Proportional Hazard Estimates of BIT Renegotiation
Full size table

Table 2 reports four probit models and two rare event models with a cubic polynomial approximation. In these models, the estimates represent the likelihood of renegotiation conditioned by values on the independent variables. Models 7–10 and 11–12 include the same variables included in models 1–4 and models 1–2 in Table 1, respectively. Table 3 presents additional robustness checks. The first four models in this table consider more expansive definitions of BIT renegotiation: Models 13 and 14 add BITs that were replaced by FTAs and models 15 and 16 include terminated BITs. Model 17 includes a spatial lag and Model 18 excludes the first nine years after the treaty’s entry into force.

Table 2 Probit and Rare Event Logit Estimates of BIT Renegotiation
Full size table
Table 3 Additional Estimates of BIT Renegotiation
Full size table

In support of our theoretical expectations on learning, we find that governments that find themselves on the receiving end of ISDS are more likely to renegotiate their BITs. Dispute Respondent is larger than one and highly statistically significant in all models irrespective of the statistical technique used. Furthermore, Tables 1 and 3 indicate that this result is robust to different specifications and lags of the dependent variable, to the inclusion of a spatial lag variable, and to the exclusion of the first nine years of the treaty’s life. In addition, the results remain intact, and are even stronger, after adding BITs that were incorporated into trade agreements or adding terminated BITs. Figure 3 illustrates the substantive effect of this variable. It plots the estimated survival rate of mutually-ratified BITs at the minimum and at one standard deviation above the mean of Dispute Respondent t-1 – t-3 .Footnote 25 Put differently, it displays the percentage of BITs predicted to be renegotiated at a given point in time (over a thirty-year span), conditioned on values of this variable. It reveals a meaningful substantive effect: about 15% of BITs that involve parties which experienced no investment complaints are predicted to undergo renegotiation within thirty years. In contrast, about 25% of BITs that involve parties which faced at least three complaints over the last three years are expected to be renegotiated within thirty years.

Fig. 3
figure 3

Survival Estimates at Different Values of Dispute Respondent t-1 – t-3

Full size image

With respect to the broader diffusion of information, Models 2, 8, and 12 indicate that an increase in the number of international investment disputes worldwide does not increase the likelihood of treaty renegotiation. The estimates of Dispute Global are actually smaller than one in the survival models and negative in the probit and rare event models, suggesting that, if anything, the total number of global disputes discourages renegotiation. This result is not statistically significant in the rare event model and is sensitive to model specification.Footnote 26 It therefore appears that governments are not motivated by the ISDS experiences of other countries when they decide to renegotiate BITs. These findings suggest that investment disputes do induce renegotiation but that direct experience is key: governments tend to revise their BITs only after becoming involved in disputes themselves. Poulsen (2015) refers to this as “narcissistic learning” and points to the case of South Africa, which began seeking new BIT terms after being targeted by a potentially costly claim. Similarly, after facing a barrage of investment claims, the Czech Republic pushed for changes to its existing BITs that would limit the scope of national treatment and MFN clauses and expand national security exceptions.Footnote 27 And Australia’s BIT reforms followed a high-profile complaint filed against it by Philip Morris over marketing regulations of tobacco products.

Turning to other variables, we find strong empirical support for the expectation that accession to the EU results in BIT renegotiation. As indicated by the very high level of statistical significance of New EU Member across all models, BITs that involve at least one country that joined the group are much more likely to be revised compared to treaties that do not.Footnote 28 The substantive impact of this variable is meaningful as well. After thirty years, about 28% of the BITs with at least one new EU member are expected to be renegotiated, compared with only 16% for the remainder of the sample.

The proposition that domestic political transformations lead to treaty renegotiation obtains modest empirical support. Democratic Transition is negative but statistically insignificant, implying that democratization does not lead to the renegotiation of investment treaties. Changes in the ideological orientation of the government, on the other hand, appear to make a difference. The estimates of Party Change are positive and statistically significant, indicating that ideological shifts are associated with a higher likelihood of renegotiation. These findings are consistent with political-economy explanations for other areas of economic policy, where ideology and partisanship seem to matter a great deal (Milner and Judkins 2004; Grieco et al. 2009; Pinto 2013).Footnote 29 The age of a treaty, as a general indicator of changed conditions, proves to be an important predictor of renegotiation. In all the models reported in Tables 2 and 3, Time is positive and highly significant.

The analysis indicates that changes in the gap between the BIT partners’ level of economic development has no effect on the chance of renegotiation.Footnote 30 ΔGDPPC Gap switches signs across models and is never statistically significant. Thus, a host country that catches up with its BIT partner economically is not more likely to push for renegotiation and convince its counterpart to do so. The second economic variable, changes in the flow of FDI into the host country, also has no bearing on renegotiation. This is a surprising finding given that the nominal purpose of BITs is to attract investment. This is consistent with the observation of one UNCTAD official that negotiators, typically from foreign ministries, are focused on diplomatic concerns and do not track FDI closely.Footnote 31 In tandem, these results suggest that shifts in a government’s relative economic position rarely drive the decision to renegotiate.

In line with previous studies we find that BITs involving at least one common law country are less likely to be renegotiated. Given the more profound consequences of international treaties in common law legal systems, it is possible that such countries are more reluctant to conclude BITs (Elkins et al. 2006) but, to the extent they do, put more effort into carefully designing them and thus feel less need to renegotiate. The estimates of colonial ties point to a greater propensity of partners that share a common heritage to renegotiate their BITs but do not pass conventional thresholds of statistical significance in most models. We therefore cannot say with confidence that this factor matters for BIT renegotiation.

Lagged Renegotiated BITs is highly significant and indicates that past experience with renegotiation has a strong positive effect on current behavior. Similarly, Lagged BITs in Force has a positive sign and is (weakly) statistically significant, suggesting that, as expected, governments with a larger pool of ratified BITs also have more renegotiations. Finally, the analysis offers no empirical support for the presence of spatial diffusion. As Model 15 in Table 3 shows, Spatial Lag is smaller than one and statistically insignificant. There is thus no indication that governments emulate their neighbors’ policies in the realm of BIT renegotiation. The inclusion of these various control variables does not undermine our findings with respect to the positive effect of investment arbitration claims on the propensity of governments to renegotiate BITs.

In summary, it appears that states are more likely to renegotiate when learning takes place. Direct experience with ISDS, which reveals new information about the political and legal consequences of BITs, is strongly associated with renegotiation. It is fair to say that many BIT signers were only boundedly rational (Poulsen 2015) and became better informed when confronted with arbitration claims. At the same time, our findings with respect to Dispute Global and Spatial Lag suggest that learning is rather parochial and that renegotiation has not been driven by a broader diffusion of lessons and practices. This is especially interesting because it seems to contrast with studies on why states conclude BITs to begin with, which show diffusion and competition dynamics at play (Jandhyala et al. 2011; Elkins et al. 2006). Further research would be needed to explore more specific patterns of learning and diffusion that are not considered here.

We also find support for the effect of some other variables. As might be expected, time itself is a good predictor of renegotiation. As the conditions surrounding a BIT have more time to change, its provisions produce less desirable effects and renegotiation becomes more appealing. In this sense, time may capture both changed circumstances and accumulated learning by the parties to the agreement. In addition, we find that EU accession—an important change in the political and legal landscape for many European countries—and a change in the political orientation of the government are associated with BIT renegotiation. In contrast, democratic transition and changes in a country’s relative economic position have no effect on renegotiation. These findings suggest that some of the standard explanations for other aspects of investment policy, based on domestic political institutions (Jensen 2006; Tobin and Rose-Ackerman 2011) and relative power (Guzman 1998; Allee and Peinhardt 2014), are not as useful for understanding BIT renegotiation. One implication is that treaty renegotiation is a truly distinct phenomenon with its own political determinants.

6 Conclusion

Treaties are a snapshot in time of an institutional outcome and reflect a particular configuration of interests, circumstances and information facing states. Our research considers how these snapshots become outmoded and when states take advantage of their ability to adjust them through renegotiation. Building on theories of institutional flexibility and anecdotal evidence of BIT practice, we hypothesize that states renegotiate when they acquire new information about the legal and political consequences of their treaty commitments, and that this learning occurs especially when states are involved in investment arbitration. Our findings in support of this argument are robust and substantively significant, even controlling for a variety of changes in domestic and international conditions that could reduce the appeal or relevance of existing treaties.

International relations scholars, who have always been interested in why and how states create institutions to promote cooperation, are increasingly focused on how these choices are modified over time to produce institutional change (Rixen et al. 2016; Jupille et al. 2013). Some of our findings have potentially broad implications that could serve as questions for this research agenda. First, although we find significant room for learning in investment agreements, given how little states knew about their consequences initially, other agreements might be entered into with more complete information. In these cases, does change occur less often or for different reasons? Second, we find that states learn in interesting ways. They respond most to their own experiences and less to the experiences of other states, including their neighbors. Moreover, they seem to learn when they are forced to consider the consequences of their existing commitments, in our case through involvement in dispute settlement. An interesting question is whether, in other institutional contexts, states seek information less myopically or more routinely. Finally, we find that the factors that motivate states to sign agreements may not be as relevant at the time of change. This raises the question of whether renegotiation and other types of change are subject to distinct processes in some or all issue-areas.

With respect to BITs, studying their renegotiation expands our understanding of these important treaties, which are the main vehicle for regulating the rapid growth of international investment. While there is an important and growing scholarly literature on BITs, much of this research examines the effect of these treaties on foreign direct investment, with mixed results (Sauvant and Sachs 2009). Some skeptics suggest that BITs are often ignored by governments and investors and are therefore of little practical import (Poulsen 2010; Yackee 2008).

The growing number of renegotiated BITs belies this view and indicates that some actors are concerned enough with the consequences of these treaties to push for renegotiation. In this respect, the practice of treaty renegotiation appears politically significant—perhaps more consequential and revealing than the signing of the initial treaty. Renegotiation is a strong indicator that governments are concerned with the implications of treaty design, which was not always the case when they were initially concluded.

At the same time, the phenomenon of renegotiation indicates that governments accept the general principles of the regime but wish to adjust specific rules as they learn more about their consequences. BIT renegotiation thus reflects a change within the regime rather than a change of the regime. Indeed, the ability of states to recalibrate BITs to satisfy the evolving concerns of host governments and investors is an important mechanism for maintaining their legitimacy and relevance. This is especially important if the main alternative to renegotiation is denunciation of BITs or withdrawal from ICSID—in other words, a rejection of the regime altogether. For now, these options are rarely exercised and have political and legal disadvantages compared to a consensual renegotiation (UNCTAD 2010; Lavopa et al. 2013). The inclusion of a high-standard investment chapter in the Trans-Pacific Partnership, modeled largely on existing BITs (Alschner and Skougarevskiy 2016), is further evidence that investment agreements remain popular even if some provisions are controversial.

Important questions remain. First, what are the effects of renegotiated BITs? Further research might explore whether renegotiated agreements generate more or less FDI, and whether governments that renegotiate find it easier or harder to attract BIT partners in the future. Second, what is the impact of renegotiation on investment rules and procedures? Do the revised treaties offer investors greater protection or do they give governments more regulatory space and flexibility? Some renegotiated BITs, such as those revised by China and Germany, contain stronger provisions with respect to investor rights and dispute resolution. In contrast, the new U.S. model BIT limits certain types of claims and incorporates environmental and labor rights concerns. Addressing this issue would require a comparison of the initial and the revised treaties’ design and a systematic coding of their provisions on various dimensions. This task would help us understand with greater depth the evolution of the international investment regime and the role of renegotiation in shaping it.