The Politics of Preferential Trade Liberalization in Authoritarian Countries

ABSTRACT While autocracies constitute a third of all signatories of preferential trade agreements (PTAs), very little research has explained why some autocrats join PTAs while others do not. We argue that this variation reflects the leader’s degree of vulnerability to elite-led coups during leadership change–whether a leader enters power legally or extralegally. New extralegal leaders are more vulnerable than new legal leaders, which encourages extralegal leaders to use PTAs to both build support from exporters and pressure disloyal importers. We test our hypotheses using a dyadic data set of 120 autocracies from 1960 to 2014. Our results show that extralegal leaders sign more and deeper PTAs than legal leaders. Moreover, we find that extralegal leaders with a high risk of coups are more likely to form deep PTAs than extralegal leaders with a low risk of coups. In line with our argument, we also provide evidence that extralegal leaders sign trade agreements that are likely to be enforced. Our article has implications for the political economy of trade and for development studies.

While previous studies have greatly enhanced our understanding of the politics of trade policy, few studies account for the variation in trade policy among autocracies. 1 This is somewhat surprising given the recent body of literature on autocratic institutions developed by scholars working in comparative politics. For instance, Gandhi (2008a) argues that authoritarian regimes are motivated to create or empower institutions (for example, parliaments) that allow for limited policymaking ability. Other studies argue that autocratic institutions reduce the threat of collective action if elite powersharing arrangements are infringed (Gehlbach and Keefer 2011;Svolik 2009;Wright 2008). Such different institutional arrangements impact the protection of property rights (Gehlbach and Keefer 2011), increase domestic investment (Wright 2008) and bolster economic growth (Gandhi and Lust-Okar 2009). Strikingly, trade policy has been largely ignored.
This article takes a first step toward analyzing trade policy in autocratic regimes. Our article focuses on a specific type of trade liberalization: preferential trade agreements (henceforth, PTAs). 2 By narrowing our analysis to PTAs, we put forward a theory (and test it empirically) that explains why some autocratic leaders have incentives to institutionalize trade liberalization. We argue that leaders who enter office by extralegal means (for example, a coup) have credibility problems: promises made to their elite supporters are less credible than those made by leaders who negotiate their way into power. Lack of credibility threatens elite safety in the future and encourages them to plot against the dictator. Thus, extralegal leaders are more vulnerable to elite-led coups. Implementing preferential trade liberalization reduces leader vulnerability by building elite cooperation in two ways: (1) provide market access for exporters and (2) engender loyalty from importers through trade exemption clauses. In this manner, extralegal leaders use trade agreements to consolidate support among their elites. However, once extralegal leaders have consolidated power, incentives to further liberalize disappear.
We test this argument with a dyadic data set that includes 120 autocracies from 1960 to 2014. Our empirical results can explain the level of liberalization commitments in virtually all PTAs signed by autocratic regimes since World War II. Our empirical analysis uses both a dichotomous measure of PTA formation and a continuous measure of preferential liberalization to test both the presence and absence of a PTA as well as the level of commitment once a PTA is formed. Our results show that extralegal leaders are more likely to sign PTAs with more extensive liberalization commitments and stricter enforcement provisions than legal leaders. However, this effect is short-lived and only holds in the first 15 years in which an extralegal leader is in power. Importantly, our findings indicate that extralegal leaders with a high risk of coups are much more 1 See Hankla and Kuthy (2013) for an important exception. 2 PTAs are international arrangements under which each member grants special market access to all the other members' products (Bhagwati and Panagariya 1999). likely to form deep PTAs than extralegal leaders with a low risk of coups. We also find that extralegal leaders prefer to implement preferential trade liberalization with northern countries rather than with southern countries. Finally, we refute the alternative hypothesis that extralegal leaders commit to international treaties to boost their international reputation. Indeed, extralegal leaders are not more likely to sign other types of international treaties such as human rights treaties.
The somewhat counterintuitive conclusion is that autocratic leaders who prefer high tariffs actually end up committing to preferential liberalization through international treaties. While this effect is short-lived, PTAs, once signed, are there for good. Therefore, the liberalization commitments included in these PTAs that extralegal leaders sign early in their tenure are long-lasting and affect trade policies in autocratic regimes independently of their leaders' fates. Assuming that preferential liberalization erodes the rents of those groups favoring the status quo, such long-lasting trade policy effects reconcile our findings with those from previous studies documenting robust correlations between PTAs and democracies and between PTAs and democratic consolidation (Manger and Pickup 2014;Mansfield et al. 2002).

Leadership Change and International Trade Agreements
A fundamental characteristic of autocratic politics is the relationship between leadership survival and succession (Brownlee 2007). At the heart of this relationship lies a credibility problem during leadership change: new autocratic leaders cannot promise that the privileges and benefits shared with elites who supported the previous leader will be maintained when these leaders consolidate their rule in the future (Magaloni 2008). The creation of legislatures (Gandhi 2008b) and parties (Geddes 1994;Magaloni and Kricheli 2010) ameliorates this problem by imposing constraints on a leader's power. However, the degree to which institutional checks affect the leader will depend on the manner in which a leader entered power (Svolik 2009).
The type of autocratic leadership change, a process that we call the "mode of the entry," profoundly affects an autocrat's credibility with her elite supporters. We argue that there are two ways that autocrats can enter power: legally, according to the autocracy's established rules of succession, or extralegally, circumventing the rules through a palace revolt or coup. New legal leaders have more credibility because they are generally regime insiders and have negotiated their way into power through some recognized succession mechanism. Such mechanisms could be dynastic succession in monarchies, party leadership transition in dominantparty regimes, leadership rotation for military officers in military regimes, and so on. These regime insiders have a history of working with other members of the elite, which their preferences and loyalties known. In turn, this fosters trust and allows legal leaders to make credible commitments to fellow elites. For example, dynastic succession enables elites supporting the previous leader to enjoy the same privileges under the new leader (Tullock 1987). As a result, new legal leaders have greater support by elite supporters from the previous leader. Similarly, primogeniture rule or succession by the eldest male heir in monarchies exemplifies how legal leadership change promotes ruler-elite stability (Kokkonen and Sundell 2014). In sum, the insider status of new legal leaders generates a stable post-transition political environment.
In contrast, new extralegal leaders epitomize the political outsider: they enter power by challenging the rules of succession precisely because they could not do so under the established regime. Their loyalties and preferences are more uncertain, making ruling elites worry that they may break promises and conduct purges. According to Bove and Rivera (2015), when supporting coalition elites perceive that the leader's preferences are uncertain or even unpredictable, these elites begin to fear for their own safety and are more liable to plot against the dictator. Drawing on this insight, we argue that extralegal transitions with their lack of continuity and uncertainty make supporting coalition elites fearful for their future safety. As such, these elites have incentives to launch coups against the leader. This argument parallels a general argument by Brownlee (2007) that leadership change caused by coups, revolutions, or palace revolts often lead to violent conflict among elites in the post-transition period.
Empirical evidence of the relationship between mode of entry and insider status is displayed in Figure 1. Here, the vertical axis shows the proportion of legal and extralegal autocratic leaders who are coded as affiliated with the previous regime (Svolik 2012 Svolik (2012). Sample: 120 autocracies, 1948Sample: 120 autocracies, -2010 compared to only one-third of extralegal leaders, and this difference is statistically significant. From this, we can see that legal leaders overwhelmingly enter into power following the rules of succession from the previous regime, but this rarely occurs for extralegal leaders. This evidence confirms our theoretic approach of using the mode of entry to identify the stability of ruler-elite relationships: regime insiders negotiate their way into power but precisely for this reason, regime outsiders cannot do so.
As an illustration of the distinction between legal and extralegal leaders, consider the Gnassingbé's of Togo. Eyadéma Gnassingbé seized power by a coup in 1967 and managed to survive at least three assassination attempts, two of which came in the first 10 years of his rule (Glickman 1992). In contrast, his son, Faure Gnassingbé, who succeeded his father in dynastic fashion enjoyed the full support of his father's supporters: as the New York Times (Polgreen 2005)  While his father's entry into power involved fighting off assassination attempts from his elite supporters, Faure's ascension into power was welcomed by existing elites. The violent challenges that Eyadéma Gnassingbé experienced contrasted sharply with his son's experience. This case follows the logic of how mode of entry affects elite support: when leaders enter power by extralegal means, they face strong challenges to their rule but no such challenges occur for leaders who negotiate their way into power. Indeed, legal leaders have elites that are willing to rewrite the de jure rules to support a smooth transition.
Extending the mode of entry logic to trade policy is straightforward. Legal leaders without credibility problems are less vulnerable to elite-led coups, which affords the leader the freedom to select trade policy that maximizes government revenues. As long as elites can trust the new leader to share in the benefits that they enjoyed from the previous leader, high barriers to trade are the natural policy choice for autocracies (Hankla and Kuthy 2013;Milner and Kubota 2005). Thus, legal leaders have little incentive to liberalize trade and would be unwilling to take the further step of institutionalizing liberalization through trade agreements. This process is the conventional political economy narrative for autocracies.
However, this logic does not apply to extralegal leaders who lack credibility with their elite supporters. The lack of history between such leaders and their elite supporters makes it more difficult for these leaders to build stable political support coalitions. As such, extralegal leaders need credible policy instruments. We argue that international trade agreements are such an instrument for extralegal leaders.
Trade agreements in contrast to unilateral liberalization provide larger benefits to exporters by offering market access in the form of enhanced protection. In general, trade agreements expand markets for exporter goods by providing higher prices and greater profits (Grossman and Helpman 1995). However, a common effect of PTAs is a "trade diversion" effect whereby preferential access to a country's protected market can increase exporter profits of members within the trade agreement but adversely hurt producers of the same good for non-members (Aghion, Antrás, and Helpman 2007). Due to the additional diversionary gains that PTAs offer exporters, extralegal leaders can use trade agreements to build greater support from exporters as compared to employing unilateral domestic policies.
To take one paradoxical example, consider the formation of the EC-Algeria Association Agreement signed in 1969 between Algeria and the European Community. Algerian leader Boumédiene's rule involved massive expansion of the state and nationalization of key industries such as oil (Quandt 1969). The roots of Boumédiene's sudden desire to liberalize trade can be traced to his entry into power by a bloodless coup in 1965 and the subsequent internal violent posttransition politics that characterized the early years of his rule. Unsurprisingly, a key provision in the EC-Algeria agreement was the imposition of zero rates on two of Algeria's largest exports, textiles and refined petroleum products (Lawless and Findlay 1984). Since the European market is a customs union, such agreements would naturally worsen the terms of trade for these zero-rate sectors to all nonmembers of the agreement (Winters 1997). Although Boumédiene personally did not have liberal economic tendencies, the threat of coups early in his tenure pushed him to pursued a trade agreement, the EC-Algeria PTA, to curry favor with exporters.
As another example, consider Indonesia's involvement with the regional trade agreement, the Association of Southeast Asian Nations (ASEAN) in 1971. A few years earlier, General Suharto came into power by a coup in 1966 by deposing President Sukarno. One of Suharto's key pillars of support came from the Chinese cukongs or business elites. They provided the capital and financial acumen to support Suharto's New Order (Robison 1986). To cement this relationship, Suharto used ASEAN's preferential tariff rates for agricultural and commodity products for its members to entice several prominent Chinese cukongs who had large stakes in raw commodities such as rubber and timber (Chua 2008). For example, the rubber industry has been one of Indonesia's major agricultural exports since the 1940s, and the foundation of the Indonesian rubber industry happens to be Chinese capital (Thomas and Panglaykim 1976:159). ASEAN thus provided Suharto with a credible policy to obtain cooperation from Chinese business elites that had exclusive monopolies in these commodity markets. While the opaque nature of autocratic politics makes it difficult to draw strong conclusions, these brief examples show how the diversion-enhancing qualities of trade agreements can entice leaders to favor trade agreements if only to improve their survival prospects.
A second advantage from trade agreements is the commodity price protection given to exporters in developing countries. Until recently, many North-South trade agreements contained an export stabilization scheme that helped to prevent market fluctuations from hurting the price of exports from developing countries in the trade agreement. If the price of a particular export commodity fell below a minimum price threshold, then the exporter would be compensated through some type of disbursement process (Gruhn 1976). For example, many agreements between the European Community and African, Caribbean, and Pacific (ACP) countries were designed to remedy the harmful effects that market volatility would impose on ACP export earnings (Orbie 2007). By protecting them against market fluctuations, export elites have strong incentives to support leaders who push for such agreements.
Indeed, these stabilization schemes even enticed the most anti-liberal of leaders. One notorious example of anti-capitalist turned temporarily freetrader is Benin's Matheiu Kérékou. Three years after coming into power by a military coup in 1972, he advocated a Marxist-Leninist position for the state but also signed the Lomé Convention to join the European Community. Not only did Benin receive preferential market access to the European market, but it also obtained a compensation fund scheme that would compensate Benin's cotton exports if its earnings fell below a minimum threshold (Gruhn 1976).
While trade agreements provide obvious benefits to export elites, they can even induce support from elites in the import-competing sector. Here, autocrats can use trade agreements as a divide-and-rule tactic among import elites. In democracies, politically sensitive sectors obtain exemptions in trade agreements to make the conclusion of such agreements more feasible (Maggi andRodriguez-Clare 1998). Grossman and Helpman (1995) explain, "Exemptions allow the government to capture the support of some potential losers, while at the same time winning the favor of exporters. . .who would benefit from the agreement (684)." Applying this logic, autocrats can use exemption provisions to credibly build support with a portion of import elites. The trade agreement serves as a costly signal to import elites in the excluded sector that these elites will continue to enjoy their privileges in the future. PTAs provide two forms of costly signals of leader intent: external costs (international reputation) and internal costs (loss of support from export elites). Without a PTA, import elites would have little reason to trust extralegal leaders. Thus, the constraints imposed by trade policy reduce an autocrat's ability to make swift policy changes, which builds trust from import elites.
Consider again Suharto's willingness to join ASEAN in 1971. While he certainly needed the support of the Chinese cukongs, his first and major pillar of support came from the military, whose support he had difficulty initially obtaining. Indeed, early in his reign, Suharto was merely a "first among equals" with other military officers and had to foster trust among his fellow officers (Aspinall 2005). He attempted to gain the support of many military officers by granting bureaucratic positions to them. As many as 12 former military commanders occupied top posts in his 23-member cabinet. However, Suharto's personalization of power continued to raise suspicions among other military officers (Lee 2015). No amount of promises could curtail some members of the military from threatening Suharto's rule, especially those military commanders that hailed from Java, the main island where the government and two-thirds of its population reside (Wanandi 2012). The trade profile for Java heavily favored imports: for example, export shares in Java averaged around 14% per year from 1950 to 1965 and barely reached 18% in the 1970s (Booth 1998). Given the heavy reliance on imports, even the most modest of trade liberalization would entail high adjustment costs to Javanese military commanders tied with the businesses in Java (Robison 1986).
By willing to enter into a trade agreement, Suharto would thus be antagonizing the military, which had substantial stakes in many of the extremely uncompetitive import-competing industries (Pepinsky 2009). Trade liberalization would adversely affect the import sector (Tin 1981) and reduce the benefits obtained by Indonesian military officers tied with those industries (Robison 1986). To remedy these adjustment costs, ASEAN included a number of exemption provisions that allowed Indonesian manufacturing to receive a certain degree of protection (for example, imported goods must contain 60 percent local content) (Tin 1981:5). These exemptions did not apply to every industry, and Suharto could selectively decide how to reward those industries with supporters loyal to him. Hence, Suharto could use ASEAN to reward loyal import elites and punish disloyal import elites.
As a final advantage, trade agreements can directly increase the existing resources that leaders obtain from exports of cash crops and agricultural crops that are grown for profit rather than subsistence. This argument extends the oil wealth and regime consolidation argument discussed by Wright, Frantz, and Geddes (2013). Their study finds that autocratic regimes with oil wealth can use the rents from oil to gain the military's support by offering higher wages, better equipment, and other benefits. This reduces the likelihood that these military officers will participate in coup attempts against the regime. Applying this logic to cash crops more broadly, we argue that vulnerable, extralegal leaders can use trade agreements to increase the profits derived from cash crops and use these revenues to co-opt the military. Since trade agreements require concessions in the import-competing sector, the benefits from additional revenues versus the costs of adjustment by importers is not so clear for legal leaders. However, for extralegal leaders with credibility problems, the benefits of higher rents from cash-crop exports far outweighs the additional loss of support from import-competing elites.
Idi Amin and his participation in the Lóme I Convention captures this idea. After coming into power in 1971 by a coup, Amin nationalized over 50 industries (many with EC origins) and yet was a strong advocate for participation in the Lóme Convention to increase Uganda's coffee exports (Orbie 2007). Since coffee was the primary source of foreign capital, government revenues generated from increased coffee exports would help Amin consolidate his rule by providing higher wages and other additional privileges to his soldiers (Nurnberger 2003). This example as well as the previous illustrations highlight the ability of trade agreements to address the credibility and concomitant vulnerability to elite-led challenges that extralegal leaders face.
To recap, our argument proceeds in two steps. First, new extralegal leaders due to their lack of credibility are more vulnerable to elite-led coups and revolts. Second, this vulnerability forces extralegal leaders to use trade agreements to build elite support and consolidate their rule. This argument leads to our first hypothesis: H1: Extralegal leaders sign more PTAs than legal leaders.
Thus far, our discussion highlighted the myriad ways in which PTAs enable extralegal leaders to co-opt export and import elites. To credibly gain elite support, this argument relies on an implicit assumption that these agreements have to include sufficiently large commitments to shift away from the status quo. In other words, for our argument to be valid, the trade agreement must be sufficiently deep to obtain elite cooperation. The depth of the agreement matters as trivial increases in the exchange of goods neither compensate export elites nor sufficiently protect loyal import elites. For example, leaders could decide to sign "shallow" trade agreements, which do not improve market access, do not include market-friendly policies like eliminating behind-the-border barriers, and are essentially agreements on paper only (Downs, Rocke, and Barsoom 1996). Since such agreements are akin to "cheap talk" that is neither beneficial for elites nor particularly costly for dictators, extralegal leaders seeking to gain elite support would need to sign trade agreements with larger commitments than legal leaders. This leads to our second hypothesis: H2: Extralegal leaders sign PTAs with greater depth than legal leaders.
This distinction between legal and extralegal leaders, however, should not persist forever. Since the mode of entry is an event rather than a more permanent characteristic such as leader preferences (Colgan 2013), then over time, extralegal leaders can build their credibility with elite supporters and eventually establish a stable political support coalition. Figure 2 illustrates this point by showing the relationship between mode of entry and the risk of a coup attempt. The vertical axis shows the smoothed hazard rate of a coup attempt, while the horizontal axis displays the number of years that a leader has been in power. The solid and dashed lines are the hazard rates for legal and extralegal leaders, respectively, while the shaded areas show the 95% confidence intervals. Here, legal leaders are significantly less threatened by coup attempts than extralegal ones for roughly their first 10 years in office. After 10 years, the coup risk is indistinguishable between legal and extralegal leaders.
This figure indicates that the mode of entry only affects new leaders' credibility and subsequent vulnerability to elite-led coups in their initial years in power. Indeed, early in their tenure, extralegal leaders are unusually vulnerable. During their later years, they are no different from their legal counterparts, and thus their trade policies and propensity to form PTAs should converge over time, leading to our final hypothesis: H3: The propensity for new extralegal leaders to sign more and deeper PTAs should diminish over time.

Research Design
In this section, we test our hypotheses using a reduced form approach. We first present our data and then describe our empirical strategy. We explain the formation of up to 307 PTAs signed by 5,828 dyads of countries between 1960 and 2014. The data come from Desta (Dür, Baccini, and Elsig 2014). 3 We rely on two different operationalizations of preferential liberalization in the main analysis. First, we use a dummy for PTA as our dependent variable. We consider this specification as our baseline model, which facilitates the comparison with previous studies that also rely on a dichotomous operationalization of PTAs. We label this variable PTA. 4 The problem with this operationalization is that there is a great deal of heterogeneity among PTAs and that some shallow PTAs, especially between developing countries, are only valid on paper. Importantly for our argument, a dichotomous operationalization of PTAs is at risk of lumping together agreements that create the commitment device needed by extralegal leaders with agreements that do not create any credible commitments.
To account for the heterogeneity among PTAs and to test our second hypothesis, we rely on an alternative specification of our outcome variable: a continuous variable measuring the level of market access produced by PTAs. We label this variable Depth and the data come from Desta (Dür et al. 2014). 5 This indicator is built on 48 dummies that capture the inclusion of marketfriendly provisions in the design of a PTA. These provisions remove behindthe-door barriers, which enhance competition in liberalizing markets. For example, our variable captures whether a PTA has a chapter regulating investment or service liberalization and, if so, the specific provisions included (for example, the presence of a national treatment clause or of an enforcement mechanism to protect investment). Since a large number of these 48 dummies are highly correlated, we generate a single variable using latent trait analysis to avoid overestimating PTAs with many commitments. 6 The main independent variable, Extralegal Leader, is a dummy variable that scores one if an autocratic leader enters power irregularly and the leader was not part of the previous regime. Indeed, a scope condition of our theory is that extralegal leaders are regime outsiders. Moreover, since our theory argues that leaders sign PTAs to secure their rule, incentives to form PTAs should fade away as the length of the leader's tenure increases. Thus, in one model specification, we interact Extralegal Leaders with the number of years in which the leader has been in power (labeled Tenure). For Extralegal Leader and Tenure, the data come from Goemans, Gleditsch, and Chiozza (2009), whereas data on affiliation with the previous regime come from Svolik (2012). Besides the type of autocratic leader, several alternative causal mechanisms could drive participation in PTAs. Therefore, we add a set of control variables to avoid overestimating the 4 Figure A1 in the appendix shows the hazard rate of forming a PTA for legal and extralegal leaders. Latent trait analysis is similar to factor analysis. While factor analysis relies on a Pearson correlation, latent trait analysis is suitable for binary data. See Dür et al. (2014) for details on its operationalization. The kernel density distribution in the appendix ( Figure A2) shows that there is a great deal of variation among PTAs and that the large majority of PTAs are quite shallow. effect of the main explanatory variable. Since we have directed dyads, monadic variables appear twice on the right-hand side of our equation for both country i and country j. All covariates are lagged by one year to mitigate post-treatment bias.
First, we include several "gravity model" variables: (log of) Distance, (log of) GDP, (log of) GDPpc, GDP growth, and (log of) Trade. Second, we add a variable capturing the type of regime of country j. Third, we include a dummy that scores 1 if country i is a GATT/WTO member and a dummy that scores 1 if country i and country j have a conflict. These control variables are commonly included on the right-hand side of empirical models predicting the formation of PTAs (Mansfield et al. 2002). Finally, we include variables capturing the type of autocracy, that is, a party-based dictatorship, a military regime, a monarchy, or a personalistic regime. Indeed, it might be the case that autocracies with extralegal leaders are less institutionalized than autocracies with legal leaders, which in turn might affect trade policy. We use monarchy as the reference category, and so it does not appear on the right-hand side of our models. Table A1 in the Appendix reports the descriptive statistics and source of all the variables used in our models.
We test our hypothesis with two main models. The unit of analysis for both model specifications is a directed dyad in which country i is always an autocratic regime, but country j can be either autocratic or democratic. Specifically, we drop from the data set countries i that have a Polity IV score greater than 5. 7 We choose this design since our theory applies only to autocratic leaders, who can form PTAs with either democracies or dictatorships. Our data set is unbalanced with (up to) 107 countries i's and 180 j's, although we lose many observations in our full model specification due to missing values. Standard errors are clustered at the dyadic level. When the dependent variable is binary, we use a Cox model, which is particularly appropriate to handle left and right censoring (Box-Steffensmeier and Jones 2004). The survival time is measured from the first year in which a dyad ij appears in the sample until the dyad ij signs a PTA. However, since several dyads form more than a PTA during our time span, dyads do not drop from the data set after signing the first PTA and the survival time restarts every time a new PTA is signed by the same dyad. In other words, we estimate repeated spells survival models. 8 For our continuous outcome variables, zeros in the variable may imply both the absence of PTAs or very shallow PTAs, that is, Depth equal to zero. This is a case of censoring from below. To remedy this, we estimate Tobit regressions, which are designed to estimate linear relationships between variables when there 7 Results are similar if use a dummy for democracy based on the coding rule by Cheibub, Gandhi, and Vreeland (2010). 8 We use the Efron option on the Cox model to deal with tied failures (results are similar if we use the Breslow method). Moreover, we stratify our analysis by year, which is equivalent to include fixed effects. Indeed, Cox models with year-fixed effects could be potentially biased due to the incidental parameters problem (results are similar if we include standard year-fixed effects).
is either left or right censoring in the dependent variable, using the following baseline equation: where Depth is the level of depth of a PTA signed by country i and country j at time t. Depth captures the level of depth of a PTA from the year of signature onward. If the same dyad signs more than one PTA, which is all but uncommon, the scores for depth of the new PTA replace the scores of the old one. β 0 is a constant, whereas β 1 , β 2 , β 3 , and β 4 are the coefficients. The key coefficient of interest is β 1 , which we expect to be positive. X ij;tÀ1 is a vector of dyadic control variables, Z i;tÀ1 and W j;tÀ1 are vectors of monadic control variables related to, respectively, country i and country j, and θ t are year-fixed effects. Finally, ε ij;t is the error term. In our main model specification, we are unable to include country-fixed effects, since for several countries they correlate perfectly with ExtraLegal Leader. That would leave us with limited within-country variation to explain. However, to account for the fact that autocracies are heterogeneous, we implement a robustness check in which we include country-fixed effects and rely on ordinary least squares estimates. Our main results survive this test.
The second model is an error correction model (ECM), which distinguishes between the short-term and long-term effects of Extralegal Leader on Depth. Moreover, since the ECM relies on first differences, it accounts better for temporal dependence. Its specification is: þ β 3 ΔX ij;t þ β 4 X ij;tÀ1 þ β 5 ΔZ i;t þ β 6 Z i;tÀ1 þ β 7 ΔW j;t þ β 8 W j;tÀ1 þ β 9 Depth ij;tÀ1 þ ε ij;t ; (2) where β 1 captures the short-term effect and β 2 β 9 captures the long-term effect (De Boef and Keele 2008). Since our first differences are at the dyadic level, we are able to include country i fixed effects, the coefficients of which turn out to be statistically significant.

Empirical Results
Table 1 presents the baseline and extended models. Extralegal Leader is always positive and statistically significant at the conventional level across all models. Specifically, while there is some evidence that extralegal leaders are more likely to sign PTAs of any type, extralegal leaders are especially prone to form deeper PTAs. In other words, extralegal leaders have both an effect on the extensive (that is, signing any type of PTAs) and intensive (that is, signing comprehensive PTAs) margins. Our main explanatory variable remains statistically significant when we include dummies for the type of autocracy. Thus, Extralegal Leader does not proxy for institutional features of dictatorship. Taken together, these results confirm our main hypothesis: how dictators enter into power affects their decisions to commit to deeper preferential liberalization as well as stringent enforcement mechanisms. The control variables, which are statistically significant, usually have the expected sign in line with previous studies. Since the results for the control variables are similar in all the estimations, we omit them from subsequent tables. Next, we turn our attention to the magnitude of our results. First, extralegal leaders are up to 23% more likely to form PTAs than legal leaders are. Moreover, we find that extralegal leaders, ceteris paribus, form PTAs that are (up to) 22% deeper than PTAs signed by legal leaders. Thus, not only is our treatment statistically significant across all models, but its impact is quite substantial. Furthermore, we note that extralegal leaders formed 79 PTAs, which account for 28% of the total number of PTAs signed by autocracies according to Desta (Dür et al. 2014). Some of these PTAs involve several countries and cover a healthy amount of trade flows. Thus, although democracies tend to sign more PTAs than autocracies and extralegal leaders are less likely to sign PTAs than   Let us now explore the timing of PTA formation. Our argument predicts that extralegal leaders have incentives to form deep PTAs only at the beginning of their tenure when their grip on power is shaky. As leaders consolidate their rule, incentives to form PTAs in general, and deep PTAs in particular, should fade away. Table 2 reports two model specifications that test this proposition by exploring the timing of PTAs. The first two models show the interaction between Extralegal Leader and Tenure with PTA and Depth, respectively, as outcome variables (Models 5 and 6). The coefficient of the interaction term is negative as expected, but it is significant only with Depth as the outcome (Model 6). These results are in line with our theory: incentives to form deep PTAs decline as extralegal leaders secure their power. However, this effect is weaker when we do not take into account the design of PTAs. Indeed, some of the agreements are so shallow that they do not change the status quo, that is, they are cheap talk, and therefore extralegal leaders may keep forming some of them even after having consolidated their power. 9 Figure 3 displays the interaction of Tenure and Extralegal Leader for Model 6. Here, the figure shows that extralegal leaders are more likely to form PTAs with deep commitments when their tenure is less than 15 years. After (roughly) 15 years, this effect is not statistically significant and actually becomes negative. 10 This result validates the logic of our theory. A 15-year window might seem long, given that authoritarian leaders only require a few years to secure their power. This relatively long window might be explained by the fact that informal and formal negotiations to sign PTAs last several years. Therefore, PTA signatures might only materialize relatively late when leaders have already consolidated their hold on office. More importantly, we note that trade agreements tend to empower exporters and to weaken importcompeting industries (Ornelas 2005a). As a result, trade liberalization becomes easier after a country is a member of several PTAs. 11 This might explain why extralegal leaders keep signing trade agreements even after their power is relatively secured. Incidentally, we note that the 15-year window is roughly in line with Figure 2's length of coup risk difference between extralegal and legal leaders.
Finally, the results for our second model specification, the ECM, are reported in Table 2  Note: Data on tenure come from Goemans et al. (2009). Sample: 120 autocracies, 1948 The slow change in the interaction variable for the binary PTA measure might also be explained by the length of time it takes for informal and formal negotiations to conclude a PTA, which could suppress the effect of tenure on formation of any time of PTA. 10 The results are similar if we use a logarithmic transformation of Tenure. 11 For a formal model on the ''rent destruction'' effect of trade agreements, see (Ornelas 2005b). statistically significant only in the short term, which is captured by the first difference of Extralegal Leader and ΔExtralegal Leader. Since the depth of PTAs does not vary much over time, the significance of Extralegal Leader is weaker than in previous models. Conversely, the long-term effect is not statistically significant and it remains positive only for Depth. The long-term effect is given by the ratio between Extralegal Leader (lagged) and PTA Depth ðlaggedÞ, and the standard error of the long-term multiplier is estimated using Bewley's transformation (Bewley 1979). In sum, our results indicate that extralegal leaders are more likely to form PTAs with deeper commitments than legal leaders. However, this effect is only significant in the short term.

Exploring the Mechanisms
Our main results show a significant and substantive effect of Extralegal Leader on the probability of forming PTAs in general, especially deep PTAs. Below we explore the main mechanisms highlighted by our theory to further corroborate our findings.

Coups, Extralegal Leaders, and PTAs
The most important test concerns the mechanism linking mode of entry and the probability of facing a coup. Specifically, our argument is that extralegal leaders face a higher probability of coup than legal ones because extralegal leaders are regime outsiders and are less credible. However, not all countries have the same structural risk of coups (Belkin and Schofer 2003). Thus, we expect that the probability that extralegal leaders form deep PTAs increases with their coup risk.
To test this mechanism, we interact Extralegal Leaders with a variable capturing the probability of a coup. This variable parallels the interaction logic applied with our Tenure variable by showing how structural vulnerability to a coup affects the timing of extralegal leaders' willingness to form trade agreements. It is built by Belkin and Schofer (2003) and its coverage is between 1960 and 2000. Simply put, Coup Risk is the predicted values of a probit model estimating the probability that each country i experiences a coup. Belkin and Schofer (2003) construct this predicted value using a number of indicators such as the ratio of paramilitary to military, the size of the military, previous number of attempted coups, and so forth. 12 This variable is not a substitute for Extralegal Leader as evidenced by the low correlation between Extralegal Leaders and Coup Risk (that is, ρ ¼ 0:2).
Results are reported in Table 2 (Model 8). The sign of the interaction term is positive and statistically significant as expected. To ease the interpretation of the interaction term, we plot the marginal effect of Extralegal Leaders on the probability of forming deep PTAs for different values of Coup Risk. Figure 4 shows that extralegal leaders are more likely to form deep PTAs than legal leaders only for large values of Coup Risk. Specifically, the interaction term becomes significant in the upper quartile of Coup Risk.
The negative effect of Extralegal Leader on the probability of signing deep PTAs in authoritarian regimes with low coup risk is in line with our theory. Indeed, if the coup risk is low, extralegal leaders are likely to keep their power even at the beginning of their tenure. Thus, they have no incentives to buy off the support of exporters. Absent these incentives, forming trade agreements produces unnecessary constraints and reduces extralegal leaders' rents, which can be used to consolidate their power (Liu and Ornelas 2014). Moreover, PTAs (especially with North countries) might include provisions on human rights, which are against the interest of extralegal leaders. In sum, when the coup risk is low and incentives to credibly commit are limited, extralegal leaders are less likely to sign deep PTAs compared with legal leaders. Taken together, these results are the central result of our article: incentives for extralegal leaders to form deep PTAs depend on a leader's risk of a coup.

Exporters, Extralegal Leader, and PTAs
Another claim of our theory is that extralegal leaders form deep PTAs to obtain the support of exporters. Here, extralegal leaders as regime outsiders Note: Data on coup risk comes from Belkin and Schofer (2003). Sample: 120 autocracies, 1960Sample: 120 autocracies, -2000 must rely on deep PTAs to sufficiently compensate elites in order to gain their support. Incentives to rely on exporters to stay in power should be particularly high when countries are open to trade and have a sizeable exporting sector. In particular, we expect that the probability of extralegal leaders to form deep PTAs increases as countries become more open to trade. Following the convention in the literature, we use Trade GDP to capture trade openness. We then interact Trade Openness with Extralegal leaders. 13 Results are reported in Table 2 (Model 9). The sign of the interaction term is positive and statistically significant as expected. To facilitate the interpretation of the interaction term, we rely again on a graph. Figure 5 shows that extralegal leaders are more likely to form deep PTAs than legal leaders only for large values of Trade Openness. Specifically, the interaction term becomes significant in the upper quartile of Trade Openness. This test shows that there is a link between the economic structure of a country and extralegal leaders' incentives to form deep PTAs in order to gain the support of the export constituency.

PTAs as a Commitment Device
An important claim of our theory is that PTAs act as a commitment device. Given the distrust faced by extralegal leaders, PTAs are a credible commitment if and only if relevant actors (for example, exporters and trade partners) perceive that the agreements can be enforced and constrain extralegal leader behavior. Absent enforcement devices, extralegal leaders would hardly obtain the support of exporters to remain in power. While providing a precise analysis of the credible commitment argument is empirically quite challenging, we perform a series of tests that all point in the same direction: extralegal leaders tend to form enforceable PTAs.
First, we use a third dependent variable capturing another dimension of the PTA design, which is closely related to our argument about credible commitment: enforcement. The justification for looking at this outcome variable is that if extralegal leaders defect from agreements with strong enforcement mechanisms, they face reputation costs and other forms of sanctions. The reputation and financial costs make reneging on the agreement costly even for extralegal leaders (Allee and Elsig 2015;Baccini and Urpelainen 2014). Moreover, the external costs imposed by agreements with strong provisions also enable extralegal leaders to provide a costly signal to their elite supporters. In short, enforcement provisions should mitigate the lack of credibility faced by extralegal leaders both domestically (vis-á-vis exporters) and internationally (vis-á-vis trade partners).
To operationalize enforcement, we rely on an index measuring the strength of dispute settlement mechanisms for all PTAs in our sample. The index is based on six components: (1) the extent to which dispute settlement authority is delegated to a third-party, legal body; (2) the ability of a complainant state to choose the dispute settlement venue; (3) the method by which the chairman of any judicial panel is selected; (4) whether the DSM in a given treaty specifies any time limits for the dispute settlement process; (5) the extent to which postaward sanctions can be used to effectively implement awards; (6) whether the dispute settlement provision applies broadly to all areas covered by the agreement or only to some areas. These six components are standardized on a 0-1 scale, resulting in an indicator ranging from 0 to 6. 14 Data come from Allee and Elsig (2015). We label our first variable Enforcement. 15 Results in Table 3 (Model 10) indicate that extralegal leaders are more likely to form PTAs with stricter enforcement mechanisms than legal leaders.
Moreover, we show that extralegal leaders strategically pick trade partners that maximize the probability of enforcing PTAs. We offer three pieces of evidence. First, we find that extralegal leaders are more likely to form PTAs with democracies than with autocracies (Table 3, Model 11). This result is in line with studies, which show that democracies tend to enforce international treaties more than autocracies do (Leeds 1999). Thus, incentives to cooperate with democratic regimes are much higher for extralegal leaders who are in need of credible commitments. Second, we find that extralegal leaders are more likely to form PTAs with North countries than with South countries (Models 12 and 13). 16 There are two explanations for this finding. Since North countries are more likely to enforce PTAs than South countries due to power asymmetry and monitoring capacity (Baccini 2012), this result confirms that extralegal leaders make an extra effort to form enforceable agreements. In addition, since our theory claims that extralegal leaders use PTAs to gain market access in foreign markets, we expect that such leaders form North-South PTAs more frequently than South-South PTAs. North-South PTAs emerge due to a comparative advantage setting in which firms in southern countries are more competitive in the production of labor-abundant products than firms in northern countries. If this is the case, trade liberalization with northern countries should remunerate both capital owners and workers in exportoriented labor-abundant sectors (Kono 2008).
Third, we control for major colonizers. Specifically, we include dummies scoring 1 when country B is France, Spain, or the U.K. Indeed, if extralegal leaders are more likely in former colonies and if former colonies are more likely to form (deep) PTAs with colonizers, this confounder could bias our estimate. Moreover, we include a dummy when country B is the United States. Indeed, it may be that extralegal leaders are weak and therefore they are targeted by the United States to exploit political and economic concessions. Accounting for these incentives of the trade partner to form PTAs with autocratic leaders does not change our main results (Model 14). Fourth, we re-run our main model including a dummy for the Cold War period to account for differences in the international system. While we always include year-fixed effect in our models, this further check mitigates any concerns on heterogeneity across historical periods. This variable has a negative sign as expected, since deep PTAs have mostly been formed over the past two decades. However, our main results are unchanged (Model 15).
Finally, we assess whether PTAs formed by extralegal leaders increase trade flows. Increasing trade flows would provide evidence that PTAs are not only signed but also enforced. In other words, it would show that PTAs are de facto and not merely de jure devices to lock in trade liberalization. Given the emphasis of our model on the credible commitment argument, this constitutes an important piece of evidence. We run a standard gravity model with (the log of) exports between country i and j as the dependent variable. PTA, Depth, and Enforcement are the main explanatory variables. Moreover, we include export-year, importyear, and dyad-fixed effects as this is becoming state of the art in this literature.
Thus, we are unable to control for any time-varying and time-invariant monadic variables and for any time-invariant dyadic variables. However, we do include a dummy for joint membership of GATT/WTO. In running the gravity model, we limit the sample to extralegal leaders, since our theory is agnostic about PTAs signed by legal leaders. We find that exports increase significantly after the formation of PTAs, deep PTAs, and PTAs with strict enforcement mechanisms (see Table 4). In sum, this test provides evidence that PTAs help extralegal leaders to increase exports, remunerating export-oriented constituencies.

Robustness Checks
We implement many robustness checks to further validate our findings. Here, we discuss the key findings (the tables and details of the analyses can be found in the online appendix). First, a skeptical reader may argue that extralegal leaders form PTAs to improve their reputation in the international system. If so, we should see that extralegal leaders sign PTAs with major powers. In doing so, these extralegal leaders may hope to use such ties to help stabilize their regimes. To address this concern, we include a dummy scoring 1 when country B is a major power. Data on major powers come from State System Membership (v2011), the Correlates of War (COW) project (2011). 17 Our main findings hold and the dummy for major powers is never significant (see Table A2, Model A1). Second, we implement a placebo test to further rule out the competing hypothesis that extralegal leaders sign international treaties to boost their international reputations. We do not find evidence that extralegal leaders are more likely to sign human rights treaties (see Table A2, Model A2). Third, we show that our main results hold even if we use year of negotiation instead of year of signature as our dependent variable (see Table A2, Model A3). Fourth, we re-run our main model including a variable capturing oil rents to better control for rentier states. The data on oil rents are from the WDI (2012). Our main results still hold and Oil Rents is not statistically significant (see Table A3, Model A4). Fifth, we run OLS regressions with country-and dyadfixed effects. Even in this case, our main results are unchanged (see Models A5 and A6 in Table A2). However, Extralegal Leader has limited within-country variation once we include country-fixed effects. 18 Finally, signing a PTA is a decision by (at least) two countries. Therefore, the dyad-year unit of analysis appears to be the right decision to control for important dyadic variables such as distance. However, a skeptical reader might argue that we are inflating the outcome variable in the case of plurilateral PTAs. To address this concern, we replicate our analysis at the (authoritarian) country-year level. The results are shown in Table A3 in the online appendix. Here, it is enough to say that our main results are unchanged, though Extralegal leader loses significance in Model A7. Our dyadic results show that extralegal leaders are very careful in choosing their trade partners: for example, forming deep PTAs with democratic countries and North countries. Thus, it makes sense that the monadic analysis produces somewhat weaker results than the dyadic analysis since, with the former, we are unable to account for trade partner characteristics.

Conclusion
While some scholars have recently shown interest in the international political economy of autocracies (Hankla and Kuthy 2013;Milner and Kubota 2005), we still lack knowledge on why autocracies institutionalize trade. Our article has put forward a simple theory to answer this question: extralegal leader changes induce autocrats to institutionalize preferential trade liberalization as a means of securing power. This mechanism, supported by empirical evidence, helps explain why some dictators are more liberal than others.
These results have three important implications. First, our finding on the relationship between extralegal leader transitions and PTAs could shed light on when autocrats are most vulnerable to pressure. Extralegal entries thus have implications for policymakers seeking to change the behavior of autocrats. Our results suggest that the international community can promote democratization efforts by engaging with autocrats in PTAs, which vulnerable autocrats are motivated to enter into even if such agreements will increase the likelihood of competitive politics emerging in the long run.
Second, our results have implications for a broader set of political-economic outcomes. While we focus on PTAs, our argument could be extended to a number of other policy domains, such as macroeconomic management and compliance. For example, our theory suggests that extralegal leaders should be more inclined to support a fixed exchange rate system, which is preferred by export sectors.
Third, our argument and empirical findings touch on a plethora of recent studies concerning the political economy of authoritarian countries. Many of these articles highlight the importance of coup susceptibility for economic performance (Besley and Kudamatsu 2008), military intervention (Svolik 2013), and economic transparency to the public (Hollyer, Peter Rosendorff, and Vreeland 2015) in autocracies. Our findings suggest that autocratic leader vulnerability to coups, like democratic leader vulnerability to elections, may also potentially generate good policies.