Renegotiating in good faith: How international treaty revisions can deepen cooperation

Abstract

International agreements are often understood to help governments make credible commitments to future policy by limiting their ability to renege on their promises. Renegotiations of agreements are accordingly viewed as a threat to cooperation, since renegotiations call past commitments into question. But we know little about the frequency or nature of treaty renegotiations. When are international agreements renegotiated, and what effect does renegotiation have on international cooperation? Do most renegotiations indeed aim to backtrack on past commitments? Using the topical context of the trade regime, I collect new data on international treaty revisions, covering 310 preferential trade agreements signed since the year 2000. Around a quarter of these agreements have been amended in some form, and the supermajority of amendments result not in scaled back agreements, but in deeper commitments. Survival analysis shows that ‘like-minded’ countries with a shared language and similar voting patterns at the UN General Assembly are most likely to revise their commitments. In contrast, I do not find evidence to support the view of PTA revisions as ‘backsliding’ on past commitments. The effects of revisions on trade cooperation support the more cooperative view of revisions. An error-correction model shows revisions are associated with a long-run increase in export volumes. Renegotiations are not breakdowns in international relations, but opportunities for governments to renew their commitment to cooperation.

1 Introduction

A widely-held view in International Relations is that renegotiations of countries’ international commitments are borne of discontent, and augur ill for global cooperation (Copelovitch et al. , 2020; Copelovitch and Pevehouse , 2019). This view would seem to be supported by the ‘backlash’ against economic globalization exemplified by Brexit, the election of former United States (US) President Trump, and the subsequent US withdrawal from or renegotiation of international agreements such as the 12-member Trans-Pacific Partnership (TPP), the North American Free Trade Agreement (NAFTA), and the Korea-US Free Trade Agreement (KORUS). International Relations scholarship has long illustrated how membership in international institutions helps states to cooperate (Keohane , 1984; Koremenos et al. , 2001; Russett and Oneal , 2001); international agreements, and the institutionalization of those commitments in law through domestic ratification, constitute ‘credible commitments’ to cooperation (Goldstein et al. 2000, 393). States create certainty by limiting some of their policymaking ability, and this certainty aids cooperation. In the case of trade, economists and political scientists alike argue that policy uncertainty is reduced by membership in a preferential trade agreement (PTA) (Baldwin , 1994; Hollyer and Rosendorff , 2012), which promotes welfare-enhancing trade (Handley and Limão , 2017).

Yet we know little about when and why states renegotiate their commitments. We know even less about the effects of renegotiations on cooperation. The pessimistic view of renegotiations outlined above appears borne out in economic analyses of NAFTA renegotiations and Brexit, where trade policy uncertainty was associated with a reduction in international trade and investment (Crowley et al. , 2018; Graziano et al. , 2020; Caldara et al. , 2020). These prominent examples give the impression that revisions are unusual. They also suggest that revisions are likely to be destabilizing to cooperation as they generate uncertainty (e.g. over the future legal framework for cooperation). But other recent trade agreement revisions appear to have been conducted in ‘good faith’, without much political fanfare or opposition from economic or social actors, and with no destabilizing effects. A number of agreements are being or have been revised in the period since 2014, including the EU’s agreements with Chile and Mexico, the Australia-Singapore, Australia-Thailand, and China-New Zealand and New Zealand-Singapore agreements. Such renegotiations appear motivated not by a desire to backtrack on past cooperation, but instead by the goal of deepening cooperation.

These opposing accounts of PTA revisions make it clear that we need to understand more about their drivers and their effects. Which is the more accurate reflection of international treaty revisions in the trade regime? Just how frequently do governments revisit their international treaty commitments, and what are the effects of revisions on international cooperation? Moreover, the differing accounts of PTAs also suggests we need to better understand the varied ways in which PTAs are revised. Do PTA revisions tend to involve major changes to signatories’ fundamental commitments? If so, this would support the view of revisions as destabilising. Or do revisions tend to hew to the spirit of signatories’ commitments? If so, this would support an alternate view of revisions as more cooperative. To assess these questions, I analyse original data on trade agreement revisions, covering agreements originally negotiated in the year 2000 onwards. Previous pathbreaking work on renegotiations examined the inclusion of clauses to limit the duration of agreements, but this research did not focus in detail on the trade regime and moreover does not capture actual instances of renegotiation (Koremenos , 2001; Koremenos , 2005). Other research has explored instances of renegotiation, but has focused on international investment treaties (Haftel and Thompson , 2018). Moreover, that work does not distinguish between different possible outcomes from a renegotiation (e.g., more or less liberalizing), nor does it examine the effects of renegotiation on cooperation. The present data are collected using the Design of Trade Agreements (DESTA) dataset to identify trade agreements, as DESTA is the most comprehensive dataset on preferential trade agreements (PTAs) available (Dür et al. , 2014).

These data show that revisions of past commitments are strikingly common occurrences. Many PTAs include provisions mandating regular assessments of the agreement. Others have been ‘updated’ as part of country reviews of trade policy in general. The frequency of revisions in the trade regime seems to runs counter to the conventional wisdom that trade agreements function by creating policy certainty. I argue that PTA signatories are more likely to use revisions to deepen cooperation, expanding commitments into new issue-areas in the trade regime or extending cooperation into areas that may not have been politically possible at the time of the initial agreement. The empirical evidence supports this view. Most revisions in the trade regime appear to be aimed not at reducing market access but at increasing liberalization between treaty partners. Quantitative analysis indicates that revisions typically take place between countries with similar ideological commitments that would facilitate cooperative deepening of liberalization, and there is no indication that revisions take place during ‘hard times’. Turning to the effects of revisions, I show that on balance, revisions do not impede trade. They are instead associated with a long-run increase in trade between signatories. Far from a breakdown in cooperation, renegotiations are typically good-faith acts of cooperation.

The remainder of the paper is as follows. Section 2 develops the paper’s argument. Revisions and renegotiations, like flexibility measures, have the potential to create considerable uncertainty about governments’ intentions, and therefore about the credibility of the commitments that they make through international agreements. They are therefore often viewed as bad-faith acts that reduce international cooperation. I argue however that there are political economy and normative factors that enable governments to engage in good-faith renegotiations. The increased domestic power of prior beneficiaries of liberalization and changes in the informational environment mean that protectionist groups may not be able to mobilize as readily to secure limits to prior liberalization. The lower public salience of revisions as opposed to initial agreements dampens possible opposition to further liberalisation, enabling like-minded governments with shared ideological commitments to deepen their cooperation.

Section 3 details the data and method used to test the argument, and Section 4 presents a typology of revisions and descriptive statistics on revision outcomes, which provide initial support for the argument. Sections 5 and 6 then test the argument quantitatively. I show that revisions are far more likely between like-minded country-pairs, and I find no support for the ‘backsliding’ account of PTA revisions. Most revisions to trade agreement terms result not in a decrease in trade, but in a boost to exports. Such revisions are not breakdowns in cooperation, but opportunities for treaty signatories to deepen their vows.

2 Why revise? Literature and theory

The view of renegotiations that prevails in the literature is that they are likely to be used to limit past commitment. In consequence, renegotiations are considered to be rare and risky. Countries should avoid them so as to maintain regime integrity. Should renegotiations occur they will likely result in trade-damaging uncertainty. I outline a contrasting theory of good-faith renegotiations as opportunities for like-minded governments to collude in advancing cooperation. In this alternative view, renegotiations may be used not to limit commitments, but to expand them.

2.1 The literature: Renegotiations as bad faith actions

A dominant view in political science equates renegotiations of international agreements with breakdowns in international cooperation. This view is premised on two related challenges to international cooperation. First, countries may face time-inconsistent preferences due to changed domestic or international circumstances. Second, the lack of centralized authority in world politics (anarchy) raises concerns that countries will ‘cheat’ their partners, due to these changed preferences or to incentives to defect. Formalized commitments (‘hand-tying’ or ‘self-binding’) may help governments to cooperate: commitments to cooperate are viewed as credible because countries give up the ability to take actions contrary to their promises (Elster , 2000; Goldstein et al. , 2000). In the trade regime, governments may use trade agreements to credibly commit (vis-à-vis international partners or their own voters) to low trade barriers where they may otherwise be tempted to give in to protectionist demands (Grossman and Helpman , 1994; Mansfield et al. , 2002). Doing so enables governments to sustain more socially-optimal trade policy than may otherwise be politically possible (Mansfield et al. , 2002), to internalise terms-of-trade externalities for exporters (Bagwell and Staiger , 1999; Bagwell and Staiger , 2011), or to adopt reformist domestic policy (Baccini and Urpelainen , 2014).

Yet the cost of lost policy space may sometimes be too high to bear. For this reason, signatories to international agreements design flexibility measures or ‘escape clauses’ that allow for temporary derogations from commitments due to unpredictable negative shocks (Downs and Rocke , 1995; Rosendorff and Milner , 2001). Thus, “any incomplete contract should include some form of flexibility to help its signatories deal with uncertainty” (Pelc 2016, 39). In the same way, states may design agreements that may be periodically renegotiated after a pre-specified duration. Koremenos argues that such ‘sunset clauses’ may make countries’ commitments more credible, as they allow for states to react to unpredictable shocks that affect the distribution of gains from the agreement (Koremenos , 2001; Koremenos , 2005). In this context, it is presumed that renegotiation is driven by dissatisfaction: states will either renege on an agreement (if it is inflexible), or will be able to address their grievances during scheduled renegotiations (if sunset clauses are incorporated) (Koremenos , 2005). In this casting, revisions may channel pressures to backslide on past commitments, driven by those who have been harmed by liberalisation. Protectionist social actors may have concentrated losses that make it easier to surmount barriers to collective action as compared to free-traders (Olson , 1965; Schattschneider , 1935), and as Autor, Dorn and Hanson have shown, social dislocations resulting from exposure to foreign competition may be severe and lasting (Autor et al. , 2016). Renegotiations are accordingly viewed as a pressure-relief valve that, like flexibility measures, are used to address unforeseen dissatisfaction with the terms of countries’ commitments.

Such perspectives echo in recent research on bilateral treaty renegotiation, which has largely focused on Bilateral Investment Treaties (BITs). Haftel and Thompson (2018) argue that states are likely to renegotiate BITs when they learn about the consequences of their commitments through their involvement in investment disputes. Similarly, host states may be in a position to renegotiate BIT terms when their bargaining power increases relative to home states (Huikuri , 2023). In this reasoning, host states push to revise their commitments when they are harmed because of them. The common theme in this research is that renegotiation is viewed as a means of exiting or amending a dissatisfying agreement; dissatisfaction is likely to have resulted from a host state agreeing to terms that it later finds unacceptable, possibly because the true impact of those terms was not clear at the time (Poulsen and Aisbett , 2013).

Other recent research on renegotiations has focused on multilateral regimes. Here too, the dominant view is that renegotiations will take place as a result of dissatisfaction with the terms of a prior treaty. Parties pushing for a renegotiation are not inviting more cooperation, but are instead seeking to shift outcomes to better suit them (Morse and Keohane , 2014; Lipscy , 2017).¹

What might be the effect of revisions, according to the dim view of their motives outlined above? Like flexibility provisions built into treaties, renegotiation clauses offer governments the means of backsliding on past commitments, hence the need for careful construction and continuous efforts by contracting parties to avoid granting too much leeway to would-be backsliders (Pelc , 2009; Pelc , 2016). Particularly if revisions arise regularly, they introduce the possibility that domestic groups who have been harmed by the initial terms of the agreements will lobby the government to scale back their past commitments. We see this in the narrative around renegotiation of NAFTA and the changes to the UK’s relationship with the EU. Renegotiations created uncertainty about the nature and credibility of the US’ and UK’s commitments to their respective agreements; this uncertainty had a measurable negative effect on trade and investment (Crowley et al. , 2018; Graziano et al. , 2020; Caldara et al. , 2020).

2.2 The theory: Renegotiating in good faith

Renegotiations may be a pressure relief valve, and they may provide opportunities for those harmed by prior liberalization to attempt to reduce openness. But in contrast to the prevailing view outlined above, they may also present opportunities to deepen cooperation between signatories. Renegotiations may aim at recommitting, rather than reneging. I first discuss factors relating to the domestic political economy environment during a renegotiation that suggest a less hostile environment for revisiting PTA terms. I then argue that PTA signatories’ cultural similarities and shared normative commitments are likely to create a permissive environment for increased cooperation during a PTA revision. The former discussion suggests important control variables for the analysis, while the latter leads to more precise empirical expectations, which I formalise as hypotheses in the following section.

The discussion in the previous section suggests that revisions are likely to come about because of dissatisfaction with an existing agreement, and might therefore be dominated by those harmed by liberalisation. This is doubtless sometimes the case. But there are also political economy reasons why renegotiations may provide opportunities to increase cooperation. First, prior liberalization is likely to increase the ability of ‘winners’ to influence trade policy, as opposed to ‘losers’. The basis of this view dates at least to Milner’s seminal work on firms, which describes how as firms become more internationalized, they have greater incentives to resist protectionism (Milner , 1988). As international economic integration increases, governments may face increasing pressure to maintain liberal trade policy. More recently, firm-based research emphasises how the largest and most productive firms experience the largest expansion of trade following liberalization through PTAs, while market concentration in US PTA partners also increases following a PTA (Baccini et al. , 2017).² Consequently, protectionist interests gradually lose political power, and pro-free trade groups should gain it in line with their increases in material resources.

Second, information about potential gains and losses resulting from a PTA revision may be different as compared to an initial negotiation, in a way that increases the incentives for winners to mobilize. Greater precision about the distributive impact of trade policy should lead us to expect exporters to resist losses of market access, thereby increasing their ability to mobilize in favour of maintaining liberal trade policy once it has been negotiated (Goldstein and Martin 2000, 608–609). Similarly, because international treaties like trade agreements often constitute what amounts to a mechanism for enacting a package of domestic regulatory reforms (Baccini and Urpelainen , 2014), the public salience of new treaties is particularly high. Once agreements have been implemented, however, the prospect of further dramatic policy outcomes may be less likely, leading to lower public interest in revisions. Thus, while PTA revisions may reflect or enable protectionist pressure, PTA revisions may also reflect or enable mobilization on the part of the winners from the status quo agreement.

The foregoing discussion suggests some indeterminacy about the effects of economic and political economy factors on the likelihood of a PTA revision or its outcomes. Economic downturns, sudden trade shocks, or persistently one-sided benefits from liberalization may prompt social actors to mobilize to seek to reduce the liberalization their state has agreed to in a PTA. Conversely, positive economic growth, successful experiences of trade cooperation, and the accrual of welfare gains from prior mobilization may create a powerful domestic constituency in favor of expanding prior trade cooperation. What then determines the likelihood of revisions? I argue that it is the character of the relationship between PTA signatories. Specifically, I argue that PTA signatories’ cultural similarities and shared shared normative commitments are likely not only to motivate cooperative revisions, but also to create a permissive environment for increased cooperation during a PTA revision. Such a permissive environment may allow states to respond cooperatively to specific exogenous or endogenous drivers of revisions, which themselves are likely to be idiosyncratic and thus resistant to generalizability.

Renegotiations may be opportunities to advance states’ normative as well as political or economic agendas. As Erik Voeten has argued (2021) states’ ideological commitments may motivate the creation of and membership in international institutions. International institutions may reflect efforts to promote states’ ideologies, and states that share membership in international institutions are more likely to have shared ideological commitments. In consequence, we might expect that governments that sign PTAs with one another have a shared desire to institutionalise norms that reflect these shared ideological commitments. Particularly when it comes to non-trade issues, states may seek ideological or principled commitments that reflect normative concerns (for instance on human rights in trade) (Hafner-Burton , 2009).³ This view is expressed in the well-known literature on ‘normative power Europe’ (Manners , 2002). A more general antecedent may be found in studies in the tradition of John Ruggie, which underscore how international economic cooperation is underpinned by specific (shared) normative commitments (Ruggie , 1982; Lang , 2006). These shared norms may become an increasingly powerful component of states international economic ties as cooperative relations become densely institutionalised, with overlapping membership in multiple PTAs.

The normative character of PTA signatories’ wider relationship, beyond formal institutions like PTAs, also constitutes the social context in which policymakers, trade negotiators, and economic actors from PTA signatories are ‘embedded’. As Mark Granovetter has argued, “concrete ongoing systems of social relations” shape the behavior of actors, building trust and discouraging reneging on commitments (Granovetter 1985, 487). Policymakers and politicians who are more deeply ‘embedded’ with their counterparts in other PTA parties may be more likely to seek to enhance cooperation and to build rather than break trust with one another. In this way, shared normative commitments may support ‘good faith revisions’ to PTAs, even where the specific trigger of a revisions may be one or several of any number of a large range of exogenous or endogenous factors. Importantly, this perspective is neutral about the precise nature of signatories ideological positions; it suggests only that where such positions are shared, PTA revisions are likely to be more cooperative.

To recap, shared normative commitments may motivate cooperative revisions to PTAs and may emphasize cooperative outcomes when they occur. Above, I discussed how political economy factors could predict more rather than less liberalization, and that this indicated important control variables. Yet the foregoing discussion about normative factors also suggests how political economy forces might be structured by normative commitments. First, like-minded policymakers may have their own interests in negotiating or revising a PTA (Castle et al. , 2016), for instance if this advances wider shared objectives. They might accordingly set the agenda for revisions; as is well known, agenda-setting power is likely to shape outcomes (Tallberg , 2008). Second, like-minded policymakers may be reluctant to seek to reduce PTA commitments in a revision if this would damage ongoing social relations with their PTA counterparts. In both cases, to the extent that policymakers’ shared commitments predict greater cooperation, anti-trade social actors may be less influential during PTA revisions.⁴

Should like-minded states not simply sign deep agreements to begin with? While like-minded policymakers may wish to do so, even farsighted political leaders and negotiators cannot know, ex ante, what an ideal agreement would look like, and are unlikely to be sufficiently farsighted that they can write rules that will stand the test of time in all issue-areas.

2.3 Empirical expectations

From the above discussion I derive the following hypotheses, which are formulated in terms of average effects–negotiations are complex interactions that depend on a range of domestic and international factors, only some of which are captured by the above theory. Firstly, renegotiations should, on average, result in more liberalization, rather than less (H1). H1 captures the main claim of the paper, that PTA revisions tend to be cooperative.

The discussion about normative commitments leads me to expect that ‘like-minded’ countries with similar ideological orientations should be more likely to revise their PTAs, both because PTA revisions may be motivated by advancing shared norms, and because similar ideological commitments may provide for a cooperative context in which to review institutional commitments. I develop two specific empirical expectations. First, I expect countries with similar ideological commitments to be more likely to revise their PTAs. To capture ideological commitment I use data drawn from countries’ voting patterns at the United Nations General Assembly (Bailey et al. , 2017). As Voeten sets out, UNGA votes can provide a good proxy for countries’ foreign policy orientations, and map closely on to ideological commitments (Voeten , 2021).⁵ The commonly-used absolute distance of countries’ idealpoints, gives a measure of the similarity of countries’ ideological orientations in their foreign policy. I use the change in idealpoint distance since date of entry into a dyad’s first PTA. By using the change in idealpoint distance relative to that at the time of PTA entry, I capture the extent to which countries have become more or less similar in their foreign policy orientations over time: we might expect even culturally dissimilar states, who experience positive interactions due to cooperation through a PTA, to develop shared normative positions that might facilitate a revision to their agreement. As a robustness check I also simply use the idealpoint distance.

Thus, revisions will be more likely between countries whose voting behavior in the UNGA becomes more similar in the time following PTA signature (H2a). Similarly, to the extent that cultural similarities enable cooperation to be achieved more readily, I expect that revisions will be more likely between countries that share a common official or primary language (H2b).

The theory also has testable implications regarding the effects of renegotiations. If PTA commitments are credible because of the predictability they provide, renegotiations might reduce trade, since they create uncertainty. The argument outlined above however suggests that renegotiations may have predictable (cooperative) outcomes given the context of shared understandings between negotiators about the purpose of agreements–and the purpose of revisions. To the extent that PTA revisions tend to reflect cooperative dynamics between signatories, I expect that revisions should signal renewed commitment to cooperation between signatories. Thus, revisions will on average increase trade (H3).

3 Data and method

In this section I describe the methodology used to gather and analyze data on trade agreement revisions.

3.1 Data

To identify PTAs, I use the most comprehensive dataset on trade agreements, the Design of Trade Agreements (DESTA) data (Dür et al. , 2014). I originally searched for agreements in DESTA signed between 2000-2017, using DESTA’s March 2017 version.⁶ Using DESTA as a guide, I collected data on revisions by searching through ProQuest for newspaper and trade journal articles signalling the revision of an agreement. I focus on cases where the terms of an agreement that has already been signed by negotiating parties are subsequently re-negotiated between those same parties. During the negotiation phase of an agreement, new versions of agreements or agreement-chapters are sometimes released (or leaked) (Castle and Pelc , 2019), and these are often referred to as ‘revised versions’ of an agreement or chapter. This is different from the sort of renegotiation I am referring to here. Agreements might also be ‘renegotiated’ if a new member joins an existing institution, for example in the case of member state accession to the European Union. This also falls outside the meaning of a renegotiation for the purpose of this paper. Here, I focus on instances where the original contracting parties to an agreement decide to change the terms of that agreement. As a concrete illustration, I identify the US-Mexico-Canada Agreement (USMCA) as a revision of NAFTA,⁷ but do not identify NAFTA as a revision of the prior Canada-US FTA (CUSFTA).

In coding agreements, I followed the above definition of a revision as an amendment to a signed agreement. This definition captures the three forms of revisions described further below and in the appendix:⁸ pre-ratification amendment; amendment following a scheduled review; and amendment outside of a scheduled review. In total, identified 77 out of 310 agreements that had been revised (25%), and a total of 147 instances of a revision (with multiple revisions per PTA). I coded each revision into a different outcome category (described in depth in Section 4), in order to understand the results of renegotiations. This coding was based on available information in associated publications, press coverage, or through comparison with the previous agreement.

To test hypotheses about the causes and effects of revisions, I add the revision data to a gravity dataset built at the undirected-dyad-PTA-year level. I use import and export figures from 2000 to 2022.⁹ I include data on GDP, GDP per-capita and other country-level economic variables;¹⁰ distance and other geographic measures;¹¹ and regime type.¹² I also include a variable measuring countries’ alliances¹³ and a measure of global economic business cycles (Mansfield and Milner 2012, 75), measured by the year-to-year change in global economic output.

3.2 Analytical method

To understand when and why revisions are triggered, I use survival analysis. I fit a lognormal model, as this distribution best approximates the time-to-revision. The lognormal model also performs better than other survival models. I present a parsimonious model in the paper with robustness checks in the appendix, including a Cox proportional hazards model and a logistic model.¹⁴

I then turn to the effects of revisions on signatories’ cooperation. In order to understand whether revisions have an effect on bilateral trade, I use an error-correction model (De Boef and Keele , 2008). The ECM models the equilibrium relationship between dependent and independent variables. In this case, we can assume that trade flows and the nature of trade institutions between countries have a long-term equilibrium relationship. But we can also assume that changes in the nature of trade institutions disrupt that equilibrium in the short-term. The ECM estimator includes differenced variables which capture change in a state, e.g. from ‘non-revised’ to ‘revised’. The estimator also includes lagged dependent and independent variables. This allows me to distinguish between short-run and long-run effects.

4 Typology of revisions and descriptive statistics

I first describe revisions and present descriptive statistics, as this provides an initial opportunity to evaluate the argument.

4.1 Types of revisions: Process and outcome

I identify three different ways in which agreements can be amended once they are signed, which I describe in more depth in the appendix.¹⁵ The first is renegotiation in the pre-ratification stage, sometimes referred to as the ‘legal scrubbing’ phase. The legal scrubbing phase of negotiations refers to the period in which countries’ legal teams will examine an agreement and clarify its legal language to ensure that it is sufficiently unambiguous. Revisions at this point are likely to be driven by domestic changes between signature and ratification, such as a change in government, or sudden domestic backlash.

The second way that an existing agreement can be modified is through scheduled or periodic review of an agreement. Many agreements–especially more recent agreements–mandate a review of the implementation or functioning of the agreement after a specified time period. At such points, agreement members may decide to revise the agreement. Such revisions may be relatively minor, as in the case of updating tariff schedules to reflect new tariff classification nomenclature. They may also be more substantive, as in the case of an amendment to concessions or to the terms of the treaty itself.

A final form of revision takes place without being foreseen. In this case, one or more (or all) of the agreement members decides to trigger a revision of the agreement if they are not dissatisfied with the deal, or if changes could be made to improve it. The Trump administration’s renegotiation of NAFTA falls into this category of revision,¹⁶ but so also do revisions to ‘modernize’ past agreements, such as between Mexico and the European Union.

I also distinguish between six possible revision outcomes, as summarized in Table 1. The first is an administrative revision. Administrative revisions amend how an agreement is governed. Common administrative revisions include changes to tariff schedules to reflect new tariff nomenclature. Other administrative revisions include institution-building, such as the creation of a secretariat to oversee the agreement. Administrative revisions are not necessarily liberalizing, although they may facilitate trade through the good functioning of the agreement. Revisions may also result in increased access, as when tariffs on already-liberalized goods are further reduced. This is clearly liberalizing in nature. Revisions resulting in increased scope are also liberalizing, and may in fact be more so than revisions to increase access, as the latter may take place within a partial scope agreements where overall liberalization is more limited. Examples include revising an agreement to cooperate in new issue-areas (such as trade in services, or investment). A further type of liberalizing amendment is an upgrade, which involves both increased access and increased scope. Many of the above types of revision are enacted through a ‘protocol to the agreement’. Where this is the case, I classify the revision based on the outcome (e.g. increasing access, or increasing scope). Yet, I also describe a further type of revision as a protocol, which I use to identify a side agreement to deal with a specific issue-area in isolation of the rest of the agreement. Protocols may or may not be liberalizing in nature. Finally, not all revisions aim at further cooperation; they may also be limiting. Such revisions seek to reduce previously-granted access. Examples include increasing the regional content required for a good to qualify for duty-free access under the agreement.

Table 1 Overview of different PTA revision outcomes

What of the extent of revisions? Changes to increase access, or limiting revisions tended to involve relatively minor changes to the agreement rather than changes to the parties’ fundamental commitments. For example, most moves to increase access entailed the gradual extension of liberalization, such as adding products to a country’s schedule of tariff elimination commitments, or increasing the number of professional qualifications qualifying for mutual recognition. Limiting moves were infrequent and typically minor in nature, such as removal of select products from a liberalization schedule. Revisions to increase scope were more significant, as they typically added a new chapter to an existing agreement. Common examples were the addition of services or foreign investment liberalization. The appendix elaborates with further examples.

4.2 Descriptive statistics

Revisions occur surprisingly often. In the sample of 310 PTAs signed after 2000, 77 agreements (25%) are revised at least once. As Fig. 1 shows, the mean time to a PTA revision is a little under seven years (6.86 years). This includes all forms of revision, both major and minor. Limiting the sample to those revisions that are not administrative in nature, the mean time to a revision is slightly longer at 7.1 years. As is confirmed below with diagnostic tests, time-to-revision most closely approximates a lognormal distribution.

Fig. 1

Time to revision in years

Figure 2 illustrates that the vast majority of revisions to agreements are intended to improve the access or scope of an agreement, with a minority aimed at reducing or limiting access. This figure includes multiple instances of revisions to agreements, as a single agreement may be revised several times. Of a total of 147 instances of a revision, administrative revisions account for close to a third of revisions (31%; 46 instances). Those revisions aimed at improving on access already provided were the most common, accounting for 40% of revisions (59 instances). Revisions to increase the scope of a deal were also fairly common (16%; 23 instances). Upgrades accounted for 7% of revisions (10 instances). Three instances of a revision (2%) had mixed outcomes, with a combination of both increasing and decreasing liberalization. There were two agreements (1.4%) with protocols that codified some other aspect of cooperation not strictly related to the agreement, or whose intent was unclear (because of no primary data and very little published information). Finally, limiting revisions–those seeking to claw back access granted–were very few in number (4), accounting for a mere 2.7% of revisions. This low number is all the more striking when contrasted with those revisions that aimed at further liberalization (increasing access or scope, or upgrading), which together accounted for 62.5% of revisions.

Fig. 2

Types of changes to PTAs

I examine revision type at different levels of agreement depth (based on DESTA’s ‘Index’ measure of depth) (figure presented in the appendix). Expanding market access appears common across the board, but accounts for a particularly large proportion of revisions of lower-depth PTAs. This may reflect revisions to (lower depth) partial scope agreements with a ‘positive list’ of liberalization. Upgrades are also undertaken at all levels of depth, and as one might expect, revisions to increase scope appear more frequently in agreements that address fewer issue-area to begin with. Agreements aimed at limiting cooperation are most commonly (though not uniquely) found in the most ambitious deals–those with a depth score of 7. Given the small number of limiting revisions, they may be driven by idiosyncratic factors and it is difficult to draw general inferences.

Taken together, the descriptive statistics clearly support H1, that renegotiations should, on average, result in more liberalization, rather than less. We can see that liberalizing revisions (increasing access or scope, or upgrading) accounted for the supermajority of changes to PTAs. The descriptive statistics also allow me to address the alternate ‘backsliding’ view of revisions in the literature. As the descriptive statistics reveal that only a very small percentage of revisions result in some scaling back of prior commitments, they offer very weak support for this view when seen in wider perspective. The following two sections turn to quantitative evidence on the predictors and effects of PTA revisions.

5 Predicting treaty revisions

I first use survival analysis (using an undirected dyad-PTA-year unit of analysis) to test expectations about why revisions come about. In this analysis, ‘failure’ is given by an amendment to an agreement (i.e., the broader category of revision). I structure the dataset at the dyad-PTA-year level, since the same pair of countries might be party to more than one agreement, each of which is at risk of revision, and I control for features of the dyad as well as features of the agreement. It is not always possible to know which party initiated the revision, so I use an undirected rather than directed dyad structure in which each country-pair only appears once per PTA. The PTA-dyad enters into the analysis following signature of a post-2000 trade deal (i.e., one which was subject to research). In the example of Singapore and New Zealand, the country-pair first enters the dataset in 2000, the year of signature of their first agreement (the New Zealand Singapore Closer Economic Partnership). The pair then enters the data again in 2005 following signature of the Trans-Pacific Strategic Economic Partnership (P4); in 2009 following signature of the ASEAN-Australia-New Zealand Free Trade Agreement (AANZFTA); and in 2016 following signature of the Trans-Pacific Partnership (TPP). I do not include the 2018 Comprehensive and Progressive agreement on Trans-Pacific Partnership (CPTPP) as a separate agreement. Guided by model fit criteria, I model the time-to-revision as a lognormal distribution.¹⁷ Table 2 presents the results.

In the table, I present parsimonious models in which, in addition to the independent variable of interest, I include as covariates the sum of logged imports and exports and PTA depth (using DESTA’s ‘rasch’ measure). The online appendix presents models with additional control variables; the main results hold. I also include simple spatial lags to account for country-specific and PTA-specific factors that may be driving renegotiation. Namely, I include the sum of PTA revisions of country A and country B in the dyad (lagged by one year) as well as the mean count of PTA revisions by other countries in the same PTA in question, in the case where the PTA has more than two members. The latter variable is also lagged by one year. The main specification is model 1, which models time to any revision. In column 2 I model survival time to an administrative revision. In column 3 I model survival time to ‘expansionary’ revisions, which group together access, scope, upgrade and protocol revisions. In column 4 I model survival time to scope revisions.

The lognormal survival model is an Accelerated Failure Time (AFT) model in which the natural logarithm of time-to-event is assumed to follow a normal distribution. The coefficients are point-estimates for the log time to revision. A one-unit increase in a covariate is associated with a multiplication of the failure time at a value of the exponentiated coefficient. The change in UNGA idealpoint distance variable is calculated in such a way that an increase in the variable indicates that the countries in the dyad have more similar voting patterns in the UNGA. Where the coefficients are negatively signed, this means that increasingly ideologically aligned countries have shorter time-to-PTA revision. This is the case in all models, although in model 3 the effect is significant only at the 90% confidence level. Looking to model 1, the effect of a one-unit increase in the Change in UNGA idealpoint distance is estimated to shorten time to a PTA revision by 58% (exp(-0.865) = 0.42). Change in UNGA idealpoint distance ranges from has a mean of .02 and a standard deviation of 0.32. Thus, a one standard deviation change in this variable would correspond with a reduction in time to a PTA revision by around 24% (exp(-.865*.32) = 0.758). These results support H2a, that revisions will be more likely between countries whose voting behavior in the UNGA becomes more similar in the time following PTA signature.

I find that a common official or primary language also tends to predict a faster time to PTA revision (H2b), although the result is only significant at the 90% confidence level in model 3 and is not significant in model 4. In model 1, having a common official or primary language is estimated to shorten the time to a PTA revision by around 36% (exp(-0.450) = 0.637). Figure 3 depicts this graphically. The effect is a little larger in model 2 (time to an administrative revision).

Table 2 Predicting treaty revisions: survival analysis

Fig. 3

Survival analysis: effects of common language on treaty revision

These findings are robust to a number of different model specifications, specifically: including the two independent variables separately; using an alternative measure of ideological similarity (cumulative PTA membership); including a dummy variable capturing the presence of a clause enabling a review of the services chapter (as a proxy for a general review clause); excluding ‘pre-ratification’ revisions; using the annual UNGA ideal-distance measure rather than a change measure; running a Cox regression; and running a logistic regression. These robustness checks are all presented in the online appendix (Appendix Table 2).¹⁸

The results also allow me to address the dominant alternative narrative around treaty revisions, that they are likely to reflect ‘backsliding’ in hard economic times. The models in Table 2 do not support this account of PTA revisions. In the models, increases in trade flows (such as during a trade shock) do not correspond with shorter time-to-revision. One would need more fine-grained data on trade at the sectoral or product level (and more detailed coding of revisions) to completely rule out localised or industry-specific trade shocks as a driver of revisions, but in general, increases in trade have no statistically significant effect. Where there is an effect (models 3 and 4), it runs counter to expectations, corresponding with a longer time-to-revision. Otherwise put, if anything it is reductions in trade that seem to invite a PTA revision.

Downturns in GDP also do not appear to precipitate PTA revisions. While the variables for both countries are rarely jointly significant, the results suggest that GDP growth is in fact correlated with shorter time-to-revision. Similarly, the ‘backsliding’ account of revisions might suggest that as the difference between imports and exports grows (i.e. as trade imbalances widen), revisions should become more likely, since one country may perceive that the agreement is not bringing equal benefits to all sides. The results are not statistically significant in all models, but they suggest the opposite.

Together, these results provide strong support for the main claim of the paper, that countries’ ideological similarities should predict PTA revisions. The result in model 4 indicates that increased similarity in UNGA votes, the main measure of ideological similarity, is a particularly strong predictor decisions to expand the scope of PTA cooperation to new areas. Meanwhile, the results offer no support for the long-running alternate account of revisions as ‘backsliding’ on commitments. The following section examines the effects of PTA revisions on trade flows.

6 Effects of renegotiations on cooperation

Does the uncertainty created by revisiting past commitments act as a brake on trade? Anecdotal evidence from the process of renegotiating NAFTA would suggest that uncertainty discourages firms from investing or making other commercial commitments. Is this evidence of a wider trend? I use an error-correction model to differentiate between short- and long-term effects. As for the survival models, I present a parsimonious model, and add a large number of controls in additional models presented in the appendix.

In Table 3, the coefficients on the differenced variables indicate the short-run effect of those variables on the outcome (here, logged exports). The revision disturbs the equilibrium relationship between trade institutions and export flows, causing exports to be too low (or high, in the case of a review that reduces trade). Accordingly, exports increase (or decrease) until equilibrium is reached. The ultimate long-run multiplier effect (the extent of the increase or decrease) is given by the ratio of the coefficient of the lagged independent variables over the (negative) coefficient of the lagged dependent variable (De Boef and Keele , 2008). And the rate at which this increase (or decrease) occurs is given by the coefficient on the lagged dependent variable. While the ratio of these two variables provides the substantive effect, we cannot determine the statistical significance directly. To do so we can use Bewley’s (1979) transformation.¹⁹ Conveniently, this provides us with the substantive effect and statistical significance of the long-run multiplier directly, given by the coefficient on the lagged independent variable. Finally, because the independent variable of interest (PTA revision) is binary, differencing this variable captures the shift from non-revised (‘0’) to revised (‘1’) states. But it also captures the return to a ‘normal’ state of affairs (from ‘1’ to ‘0’), returning a value of ‘-1’. This has no theoretical meaning for current purposes, and so like (Chow and Kono 2017, 898) I use a differenced variable coded ‘1’ on revision, and ‘0’ for subsequent, non-revision years.

The four columns examine the effects of four different types of revision. I first make no distinction between different types of revision (model 1). Model 2 examines only administrative revisions. Model 3 examines the effect of ‘expansionary’ revisions, described above. Finally, I examine the effects of revisions to increase access (model 4). Short-run effects are given by the bolded coefficient on the first-differenced ‘PTA revision’ variable, while the long-run multiplier effect is given by the bolded coefficient on the lagged ‘PTA revision’ variable, following Bewley’s transformation. If revisions have a varied effect on trade depending on their nature, we should see an increasingly positive effect as we move from Column (2) through Column (4), since this differentiation captures revisions that bear increasingly on trade liberalization (as opposed to other types of liberalization, including the addition of other issue-areas).

Table 3 Long- and short-run effects of PTA revisions on bilateral trade

This is indeed what we find. For all revisions excepting administrative revisions, there is a short-run increase in trade, ranging from around 0.11 (model 1) to 0.23 (model 4). We can manually calculate the long-run effect on trade. For models 1 and 4, the effects are calculated as

$$\begin{aligned} 0.060/-0.090\times -1 = 0.66 \quad (\text {model}\;1) \end{aligned}$$

$$\begin{aligned} \text {and for a liberalizing revision} \quad 0.167/-0.090 \times -1 = 1.86 \quad (\text {model}\;4) \end{aligned}$$

These results are equivalent (accounting for rounding error) to those given directly on the lagged variable in the second part of the table (following Bewley’s transformation). Thus, we can see that PTA revisions result in a long-run increase in trade that ranges from 0.47 (model 3) to 1.86 (model 4). Bewley’s transformation allows us to see that these effects are significant at the 99% confidence level. The exception is model 2 (administrative revisions), for which there is a long-run decrease in trade (-0.196), significant at the 95% confidence level.

How large are these effects in substantive terms? The logged trade variable has a mean of 34.0 and a standard deviation of 8.3. In the model with the largest effect (model 4), the long-term effect of a PTA revision is around 22% of a standard deviation in logged trade.

In sum, I find that most revisions to trade treaties are associated with a statistically significant increase in exports in both the short- and long-term, in strong support of the hypothesis that revisions will on average increase trade (H3). We can also see that as expected, the largest effects on trade are created by revisions that increase market access.

The results are a little more ambiguous for administrative revisions. The negative effect disappears in the fuller model specification presented in the appendix and in fact turns positive (though it remains modest when compared to the effect of other revisions). Yet in the parsimonious model presented here, administrative revisions appear to result in a long-run decrease in trade. It is possible that such revisions may sometimes place an additional burden on exporters, thereby reducing trade. It is also possible that such revisions are used as protectionism in disguise, if they result in lower compliance with the terms of an agreement. Finally, it may be that if these revisions are not sufficiently well-signalled as enhancing cooperation they generate uncertainty about country commitments, in line with the conventional ‘credible commitments’ view of cooperation. If this is the case, it suggests an important role in signalling in communicating government intentions.

7 Conclusion

High profile renegotiations of countries’ treaty commitments have recently been viewed as a major source of economic uncertainty and as a sign of a liberal order under threat. Yet beyond a few prominent cases we know little about when and why countries choose to revisit their past treaty commitments. This paper contributes to remedying this gap in our knowledge by introducing a new dataset of revisions to international trade agreements signed after the year 2000. Based on a few prominent cases, such as the revision of NAFTA under the Trump administration, renegotiations are generally understood as a breakdown in cooperation. The new data presented in this paper show that in fact, governments usually use renegotiations to deepen cooperation. And while renegotiations are usually seen as rare events, the data show that they are surprisingly common.

Descriptive statistics, a survival model, and an error correction model provide support for the paper’s hypotheses. I find that PTA revisions are overwhelmingly aimed at building on prior commitments, rather than scaling them back (H1). ‘Like-minded’ countries that share ideological commitments and a common language are more likely to undertake revisions, consistent with the view that revisions may be enabled by the character of the relationship between partners (H2a and H2b). And finally, I find that PTA revisions tend to increase rather than decrease trade (H3).

These findings provide ample scope for further research. First, future research can build on anecdotal evidence by delving into case studies of renegotiations. Since the 1980s, New Zealand has progressively deepened trade liberalization with partners through revisions to its agreements. This began with revisions in 1988 to the Australia New Zealand Closer Economic Relations Trade Agreement (ANZCERTA, popularly CER), and now includes major revisions to the agreements with China and Singapore. In line with the discussion in section 2.2, evidence from the Australia-New Zealand case illustrates how policymakers from the two countries cooperated to manage their (often reluctant) domestic groups during the CER negotiations (Andre et al. , 2003), including by choreographing a sequence of liberalization and postponing challenging issues until a later point. As a senior official from the period has recounted, ‘[o]fficials were careful not to bite off more than they could chew at the outset. For example, we did not deal comprehensively in the first iteration with services or investment, but we provided for a comprehensive review of what were called ‘second generation’ issues, and this proved wise and successful’ (Groser 2020, 11–12). Moreover, the historical record of the 1988 CER review reveals how Australian and New Zealand policymakers overrode business opposition where it did not align with their own agendas for further liberalization (Castle et al. , 2016).

The European Union is also upgrading agreements, including with both Chile and Mexico. In the European, as in the New Zealand cases, trade agreement upgrades have taken place with relatively little public attention, and have resulted in a gradual increase in commitments. This contrasts with the NAFTA experience, and illustrates the importance of social context in sustaining cooperation in trade agreements. More research into policymaker motivations for initiating revisions–especially large ones–would provide a deeper understanding of the conditions under which the ‘winners’ and ‘losers’ from prior liberalization may be likely to influence outcomes.

Second, I have found that preferential trade agreements are typically revised in the direction of further liberalization. This is at odds with the evidence from revisions of bilateral investment treaties, where revisions more typically seem to stem from dissatisfaction with the agreement (Haftel and Thompson , 2018; Huikuri , 2023). Further research would valuably ask what accounts for this difference. Similarly, more quantitative research could zoom in on the distributional impact of trade liberalization at the micro level. Finer-grained details of trade agreement changes at the product level could be combined with changes in trade flows at a similarly micro level. Under what conditions are those harmed by prior liberalization able to initiate a PTA revision or influence its outcomes?

The findings also allow for a broader reflection on the relationship between global order and formal institutions like PTAs. To the extent that PTA revisions are frequent, this would suggest that it is important to nuance the discussion of PTAs as commitment devices, and that we should understand PTAs within their relational context: what is the character of the relationship between signatories? This may tell us about the extent to which formal commitments are complements rather than substitutes for other sources of stability, such as trust, goodwill, and shared normative commitments between policymakers in different countries. While order in international regimes certainly relies at least in part on the creation of binding international commitments, as in the ‘hand-tying’ account, it likely also owes to governments’ shared normative commitments. As the institutionalization of such commitments becomes increasingly dense, as in the trade regime where international economic agreements increasingly overlap, the general commitment to similar norms may become more stable.

This is not a deterministic argument leading to ever more liberalization: renegotiations may also be taken as opportunities to shift the balance between domestic policy space and international openness. There too though, one might envisage like-minded partners sharing similar rebalancing objectives. What is more, not all governments share commitments to market-liberal values. In such cases, negotiators and policymakers will be less inclined to use renegotiations to further commit to liberal trade policy. The case of the Trump administration’s renegotiation of NAFTA and KORUS suggest that higher salience of renegotiations, combined with a lesser commitment to liberal values, may result in downwards revisionism of past commitments.

To restate the main argument, renegotiations are usually synonymous with backsliding. I show that in fact, they result in increased cooperation. Renegotiations are not breakdowns in cooperation, they are the renewal of countries’ vows. In the trade regime, countries renegotiate in good faith.