Free to Trade: Democracies, Autocracies and International Trade

Relatively little research has focused on whether countries’ political institutions affect international trade relations. Here, we address this issue by analyzing the relationship between regime type and trade policy. We begin by presenting a formal model of commercial policy that establishes that the much more important role played by the domestic legislature in democracies than autocracies leads pairs of democratic countries to set trade barriers at a lower levels than pairs composed of an autocracy and a democracy. We then test this hypothesis by analyzing the effects of regime type on trade during the period from 1960 to 1990. The results of this analysis confirm that democratic dyads have more open trade relations than dyads composed of a democracy and an autocracy. The effects of domestic political institutions on trade policy have been a longstanding source of controversy among social scientists. Despite the widespread interest that this debate has prompted, remarkably few studies have addressed the links between political regime type and commercial policy. Most existing studies of the political economy of trade policy that examine institutional factors tend to focus only on variations in policy among democracies, and avoid comparing democracies with other types of regimes. Moreover, many analyses of trade barriers, especially those by economists, ignore the effects of domestic political institutions. In light of the recent interest expressed in the effects of regime type on various aspects of foreign policy (e.g., Doyle 1986; Farber and Gowa 1995; Russett 1993; Siverson and Emmons 1991), a systematic analysis of the links between regime type and trade policy seems long overdue. In this paper, we examine whether groups of democracies are better able to liberalize trade than are groups of autocracies or groups comprised of both democracies and autocracies. By providing one of the first assessments of whether patterns of commercial protection depend on regime type, our study sheds light on whether there is a "democratic difference" in foreign economic policy. It also should contribute to a broader understanding of the political economy of trade policy. As one recent analysis concludes, "Theoretical and empirical work relating [domestic] institutional contexts to trade policy outcomes is in its infancy but should be a promising area of research" (Rodrik 1995, 1485). We seek to help fill this important gap in the literature. To begin, we develop a formal model that emphasizes the role of domestic legislatures in democracies. Central to this model is that the chief executive (i.e., prime minister or president) in a democracy needs the approval of a legislative majority to enact her preferred trade policies. Absent such approval, the legislature’s preferred policy becomes law. We argue that legislative "ratification" of commercial policy occurs in both parliamentary and presidential systems. In parliamentary systems, ratification is often ex ante; the potential prime minister must negotiate an acceptable trade policy with her legislative majority prior to taking office. This internal bargain is enforced afterward because the majority can call for a vote of (no) confidence if the prime minister strays from the accepted position in international economic negotiations. In presidential systems, major changes in trade policy usually require legislative ratification ex post. By contrast, autocracies vest their chief executives with much more authority than democracies. Autocrats do not need a legislative majority to pass their trade policy initiatives; either a legislature does not exist or it rubber-stamps executive proposals. We claim that this institutional difference contributes to the tendency for pairs of democratic countries to agree upon lower trade barriers than pairs comprised of a democracy and an autocracy (i.e., mixed pairs). We further demonstrate that whether pairs of autocracies liberalize trade more extensively than either pairs of democracies or mixed pairs depends on the preferences of the political decision makers involved, as well as on institutional differences between their regime types. Hence, we can make no prediction about the trade barriers of autocratic pairs relative to mixed pairs or democratic pairs without conditioning on preferences. Our focus differs sharply from that of studies addressing how institutional variations among countries having the same regime type affect trade policy. Crucial to the following analysis is that the institutional feature distinguishing democracies from autocracies is the existence of a popularly-elected legislature with the capacity to constrain a country’s chief executive. We assume that legislatures are more protectionist than executives in democracies and show that, even so, pairs of democracies are more likely to liberalize commerce than mixed pairs. Our focus, then, is on differences across regime types, not on differences within them. The results of an empirical analysis based on the period from 1960 to 1990 are consistent with these predictions. Holding constant various economic and political factors, democratic dyads tend to trade more freely than dyads composed of a democracy and an autocracy. Furthermore, no significant difference exists in the openness of commerce within autocratic pairs and within democratic pairs. THE MODEL The Actors and their Preferences A key difference between an autocracy and a democracy is that the latter is vested with a legislature that has de jure and de facto ratification power over the chief executive’s proposals. Because no such institution exists or has such powers in an autocracy, it is modeled as a unitary actor, having only an executive who is effectively a dictator and is labeled A. In contrast, a democracy is characterized by two actors: an executive who may be the president or prime minister, P, and a legislature represented by the median legislator, C. In what follows, we consider the case where two countries enter into negotiations to reduce trade barriers between them. The labels A, P, and C refer to the "home" country in these negotiations, whereas A, P, and C refer to the "foreign" country. In the home country, each of the aforementioned actors tries to maximize its utility by obtaining the greatest possible level of electoral support. The support that actors accrue depends on the trade policy chosen. Every actor wants to enact the level of trade barriers most closely resembling that which will maximize its political support (i.e., its ideal point). But each one must strike a balance between the conflicting preferences of its constituents to determine such an ideal level. Actors receive political support from a variety of domestic groups. Some groups prefer the home country to liberalize commerce (e.g., consumers, exporters, users of imported inputs, owners of relatively abundant factors), while others prefer it to raise trade barriers (e.g., import-competing firms, owners of scarce factors). All groups, however, prefer that the foreign country’s trade barriers be as low as possible, an outcome benefiting consumers, exporters, and importcompeting firms, alike. Thus, ceteris paribus, all domestic groups will increase their electoral support for policy makers in the home country if these officials negotiate very low levels of protection abroad. We refer to the home country’s trade barriers as t and those of the foreign country as t. The political support function of each actor, Ui, can be expressed as a simple loss function. Every actor has an ideal level of trade barriers at home and abroad, denoted (ti, ti ), that i would choose if it was not constrained by any of the other actors. All domestic actors prefer the elimination of foreign trade barriers; hence, ti * = 0 for i = P, C, A , and ti = 0 for i = P, C, A. Each actor therefore maximizes its utility by minimizing the difference between the level of protection that it prefers both at home and abroad and the level that actually obtains. For the home country, i’s utility function is Ui(t, t ) = − (t − ti) − t for i = P, A, C. (1) For the foreign country, its actors attempt to maximize Ui(t, t ) = − t − (t − ti) for i = P, A, C. (1) These utility functions imply that the actors’ indifference curves are circles. Their utility decreases symmetrically as changes occur in any direction away from their ideal level of trade barriers at home and abroad. Such utility functions are very common to spatial models (e.g., Enelow and Hinich 1984). Our approach to the making of trade policy is consistent with behavior in both democracies and autocracies. For instance, our approach captures the notion that democratically-elected politicians are motivated by a desire to be reelected, the chances of which grow larger in the face of both rising welfare experienced by voters and increasing campaign contributions (or lobbying activity undertaken) by organized lobbies, such as industry groups. Hence, trade policies that raise consumer surplus and firm profits increase the political support that a politician receives. Consumers are made better off by a rise in their surplus; and because they vote according to their pocketbooks, their support for the government grows. In addition, as firms’ profits rise, they distribute higher profits to consumers (who are also stockholders) and make larger contributions to (or exercise greater influence over) politicians. Note that the government responds to the pressures of both its import-competing and exporting firms. While we assume that the government does not have a direct instrument to benefit the exporting sector (such as a subsidy), it can reduce trade barriers abroad through negotiations and reciprocal concessions, thereby benefiting local exporting industries. In autocracies, the executive also must maintain political support. The extent of the support provided by consumers and firms often depends on economic conditions. For example, an recent analysis noted that, in the wake of Deng Xiaoping’s death, China’s autocratic leaders "are acutely aware that a vibrant, flourishing, and progressing economy is the core of their legitimacy" (New York Times, February 21, 1997, A15). Again, trade policy has a direct influence on consumer surplus and firms’ profits. Greater consumer surplus means higher real incomes, fewer incentives to riot and rebel, and thus more support for autocratic leaders. Similarly, greater firm profits mean higher incomes for at least some groups (the firm owners), implying increased political support for the autocrat. Whether autocrats care more about consumers and the overall state of the economy or the special interests of firms depends on a number of factors which do not concern us here. The trade policy preferences of autocrats relative to those of democratic leaders are of central concern. It is unclear whether, on average, autocrats are more protectionist than democratic heads of state. Freed from the need to finance expensive electoral campaigns, autocrats may be more attentive to the economy’s performance and hence less protectionist than their democratic counterparts. On the other hand, if autocrats depend only on the support of a small group of powerful interests, they may be more willing to extract economic rents (through trade barriers and other means) from those sectors of the economy in which these interests do not operate. We handle this difficult issue by varying the preferences of the autocrat relative to those of the democratic executive and assessing how changes in their relative preferences combined with the institutional differences between these two regime types affect our results. In a democracy, the two main political actors often do not agree on the ideal level of home trade barriers; i.e., tP and tC are unlikely to be identical. The executive and the median legislator may weigh consumer surplus and firm profits differently in their electoral calculus. Legislators, representing smaller constituencies than the chief executive, are more easily captured by special interests than the executive (Mansfield and Busch 1995; Rogowski 1987). We assume that, as such, the median legislator cares more about the profits of import-competing firms than about consumer surplus and holds more protectionist preferences than the democratic executive (Lohmann and O’Halloran 1994; Rosendorff 1996). Thus, we assume that tP < tC and tP* < tC* . 6 All domestic agents prefer that, ceteris paribus, foreign barriers be as low as possible. Ideal values for the domestic trade barriers do vary, however. The median voter theorem is therefore applicable: single-peaked objective functions over a single dimension imply the median voter is decisive (Black 1958). This similarity of preferences allows us to avoid the cycles associated with median voter models in larger dimensions. We can therefore identify the median legislator as that member holding the median value for the level of domestic trade barriers, and the legislature as represented by this median legislator. Each player receives the maximum utility when its ideal trade barrier combination is adopted: maximum feasible utility for each player is U(tP, 0), U (tC, 0), U (tA, 0), U(0, tP*), U (0, tC*), U (0, tA*). The Non-Cooperative Tariff Setting Game: The Trade War In the absence of any agreement on mutual trade policy concessions, each country is free to set trade barriers only in its own country. Each country sets its trade policy simultaneously, taking the actions of the other as given. The Nash equilibrium to this non-cooperative trade barrier-setting game can be considered the "trade war" outcome, which is what we assume occurs when countries do not agree to mutually acceptable trade barrier concessions. Political leaders at home have authority over t. Foreign barriers (t) are controlled by the foreign government. When both countries are autocratic (denoted AA), each executive determines the level of trade barriers in her country. Consider the problem facing the home autocrat, A, who maximizes equation (1) taking t as given. A’s reaction function is independent of t, and hence is vertical at tA. 7 Similarly, A’s reaction function is independent of t and hence horizontal at tA*. The Nash equilibrium is therefore the pair of preferred trade barrier levels (tA, t * A*) chosen by the home and the foreign autocrat. In Figure 1, this point is labeled NAA*. Note that both parties could be made better off if they could negotiate to a point on or closer to the Pareto frontier, i.e., the contract curve tAt * A*.. [Figure 1 about here] When one or both countries are democracies, we assume that the legislature (C or C) has the final say on domestic trade policy. Again, each domestic player’s ideal domestic policy is independent of the actions of the foreign player (and vice versa). Hence, when the home country is a democracy and the foreign one is an autocracy (denoted DA*), the best-response functions call for C to implement tC for any value of t , and for A to play * A * t , irrespective of t. The non-cooperative game between home and abroad has as its equilibrium NDA* = (tC, tA* ), as labeled in Figure 2a, 2b, and 2c. Realized returns to C, P, and A are C(tC, tA* ), P(tC, tA* ), and A(tC, tA* ), which are lower than the respective optima C(tC, 0), P(tP,0), and A (0, tA* ). Pareto improvements are possible; if C, P, and A can agree to mutually reduce trade barriers (slightly) from (tC, tA* ), all could gain. [Figure 2a, 2b, and 2c about here] Finally, when both countries are democracies (DD), the Nash equilibrium is set by the two legislatures. In this case, failure to find mutual concessions leads to point NDD* = (tC, t * C*) in Figure 3. [Figure 3 about here] In all three cases, the Nash equilibrium to this simultaneous trade barrier-setting game is the defect-defect outcome in the prisoners’ dilemma. If the countries do not cooperate, they will choose a level of trade barriers at home that is ideal, but that results in a level abroad that is higher than they ideally desire. Since both countries make such a choice, they end up at the non-cooperative Nash equilibrium, which is not welfare maximizing. If they coordinated their trade policies, they could both achieve welfare gains. As such, the problem facing political leaders is to reach a pair of trade barrier levels closer to their optima, and this involves per force making mutual concessions. The Negotiation Game: TILI Consider the simplest bargaining game in which we assume that the actors have perfect information. The players’ ideal trade barrier levels at home and abroad are common knowledge, as is the structure of the game. We consider two bargaining structures. In the first game, the home country -by virtue of having the first move -is chosen to make a take-it-or-leave-it (TILI) offer to the foreign country. The latter can either accept and implement this offer or reject it, thereby inducing a reversion to the trade war outcome, i.e., the Nash equilibrium. Then, we switch the identity of the first mover, and consider the same TILI game where the foreign country makes the first move, and bargaining power is reversed. Under either bargaining structure, neither country can dictate the levels of trade barriers that both choose. Both countries know they can gain from making mutual concessions and they know that everyone must be left at least as well off as they would be in the trade war outcome. We establish that the results are robust to this change in the bargaining structure, and we infer that any structure in which bargaining power is distributed in a less extreme fashion between the players will yield similar results. For instance, if the parties agree to split the gains from any mutual reductions in trade barrier levels evenly -an approach consistent with the symmetric Nash Bargaining Solution (Nash 1950) -they will generate outcomes that lie somewhere between that of TILI when the home executive proposes and TILI when the foreign executive proposes. Thus, our results are general to a variety of bargaining structures. In the TILI game, the moves of the players occur in two steps. First, the home executive, whether democratic or not, makes an offer, specifying a mutually reduced level of home and foreign trade barriers. If both countries are autocracies, the foreign leader is asked to agree. If home is a democracy and foreign is an autocracy, then the home executive puts its offer to both the foreign autocrat and the domestic legislature. Either player has veto power, inducing a reversion to the trade-war outcome. If both countries are democracies, the home executive picks an offer, which must be acceptable to the foreign executive and both legislatures. Failure to obtain legislative ratification in the latter two cases leads to the trade war, or Nash equilibrium, either NDA* or NDD*. It is assumed that no player will ever take an action that is dominated: they will not reject an offer that is better than that which will be received in the trade war. The elimination of dominated strategies is equivalent (in this game) to requiring the equilibrium to be subgame perfect. Note that the structure of preferences in the countries is symmetric. That is, the optimal level of trade barriers preferred by A is the same as that preferred by A, and similarly for P and P as well as C and C. We make this assumption because our focus is on difference between regime types, and not on those within each regime type. This assumption, however, has two implications. First, we do not model what happens when, say, a protectionist autocrat faces an autocrat who prefers to liberalize trade, an interesting but different issue than the one we address. While interesting, this involves variations within regime types, not variation between them. Second, the symmetry assumption implies that the distance between the democratic executive and its legislature is the same at home and abroad for all democratic dyads. Again, we make this assumption because we are interested in variations between regime types, not within them. Both our model and our data analysis conducted below treat each type of regime as homogenous. Democracies are distinct from autocracies because they have a legislature that exercises ratification power, regardless of its preferences. The differences among regime types rest on institutional features and not on the policy preferences of decision-makers. Varying the difference in the ideal policies of P and C does not alter the degree of democracy present; it only changes the extent to which government is divided within democracies. Furthermore, whether a government is divided or unified has no bearing on the degree to which a country is democratic. Hence, examining the influence of such divisions (i.e., as the ideal points of P and C diverge or converge) does not inform us about democracies versus autocracies, only about more divided versus less divided democracies. Our model is capable of deriving outcomes when such symmetry of preferences is relaxed. But we have neither the space nor the data in this paper to explore such variations within democracies. Figures 1-3 show the central results. We measure the aggregate level of trade barriers by the sum of the two countries’ levels of trade barriers. The indifference curves for the actors C, C, P, P, A, and A through N have been drawn and are labeled and IC, IC*, IP, IP*, IA, and IA*, respectively. These curves mark the boundary of the set of mutual reductions in trade barriers that are preferred by each of the players (respectively, C, C, P, P or/and A, A) to the non-cooperative Nash equilibrium outcome. The solution to the bargaining must lie in this feasible set, often termed the “win set.” In addition, the contract curve between the executives A (or P) and A (or P) has been drawn. It is a straight line between the executives’ ideal points (since the indifference curves are circles). The contract curve between C and P (or C and P) can be thought of as that part of the horizontal (vertical) axis that lies between tC and tP (t * C* and t * P*). More generally, the contract curve refers to the set of points where the two actors’ indifference curves are tangent; i.e., it is the set of trade barriers that both actors mutually prefer to any other set of barriers. Note, however, that movements along the contract curve cannot make one actor better off without making the other worse off. In Figure 1, two autocracies must agree on mutual concessions about trade barriers. Home will offer a mutual reduction of barriers to point a on the contract curve: this is the point on the Pareto frontier that leaves foreign indifferent between accepting and rejecting, thus allowing home to earn all the surplus. The level of bilateral trade barriers under TILI is measured by the sum of the home and foreign trade barriers. Aggregate barriers are 0ta + 0ta . Figures 2a-2c present the results of bargaining among mixed pairs, while varying the relative preferences of the autocrat and the democratic executive. In Figure 2a, a democracy seeks to find mutually acceptable trade concessions with a protectionist autocrat. The democratic executive, P, has freer trade inclinations than the autocratic leader, A. P must offer a set of trade barriers inside the area between the two indifference curves, IC and IA*, in order for C and A* to accept it. In addition, P will offer a point on the contract curve between tP and t * A*. The democratic executive, P, optimally wants to choose her ideal point, tP, but is constrained by A . Thus, P must choose the point where A’s indifference curve through the no-agreement point intersects the Pareto frontier, i.e., point m. In Figure 2b, a moderate autocrat seeks mutual trade concessions with a democracy. Now, P’s ideal point is in the win set. P naturally chooses this as the offer (tP, 0), which both the foreign autocrat and the democratic home legislature accept. In Figure 2c, a liberal autocrat attempts to find mutual trade concessions with a democracy. Here, the democratic executive is much more protectionist than the autocrat. The outcome in this case must also lie within the win set created by the legislature’s indifference curve through the Nash trade war equilibrium, IC, and the autocrat’s indifference curve through this point, IA*. The legislature exercises the biggest constraint here. The democratic executive and the autocrat must agree on a point closer to their ideal points within this win set. Since P makes the offer, P chooses μl = tC − tA*, which is accepted by A. In Figure 3, two democracies with symmetric preferences try to find mutually acceptable trade concessions. The executives must choose a set of trade barriers within the win set formed by the indifference curves of the median legislators through the Nash equilibrium, i.e., between IC and IC*. Again, the contract curve between the two executives inside the win set is the line segment dd. P chooses the point on line segment dd closest to its ideal point, tP. The offer is at point d, which is accepted by the foreign executive and both legislatures. Levels of Trade Barriers and Regime Type Can we compare the levels of trade barriers across the three types of regime pairs? Proposition: Aggregate trade barriers are lower between democratic pairs than between mixed pairs (an autocracy and a democracy). A complete proof for both bargaining structures is in the Appendix (Propositions 1 and 2). While we measure the aggregate barriers by the sum of the two countries' trade barriers in the proofs, intuitive use can be made of the length of the ray from the origin to the agreement point. Figure 4 shows all the possible outcomes in a single diagram. When home makes the offer, the bilateral trade barrier level in the democratic pair is measured by the length of the ray is 0d. In the case of the protectionist autocracy in the mixed pair, the ray has length 0m; in the moderate case, the length is 0tP; in the liberal case, the ray has length 0μl. In all three cases, the ray corresponding to the democratic pair (0d) is smaller than that corresponding to the mixed pair (0m, 0tP, or 0μl). [Figure 4 about here] A democracy lowers its trade barriers more when it seeks mutually acceptable concessions with another democracy than when it deals with an autocracy, no matter what the relative preferences of the two leaders. Why is this the case? When both are democracies and home makes the initial proposal, it is the pressure of the foreign legislature that forces the home country to compromise more, or else end up in a trade war. The Nash equilibrium is worse for both countries in the case of two democracies (NDD*) than in the autocracy-democracy case (NDA*), which induces the democratic executives to agree to liberalize trade more than otherwise. The threat of legislative veto in both countries moves the executives to a freer trade equilibrium than in the case of a mixed pair. Paradoxically, then, a protectionist legislature forces democracies to lower their trade barriers more than otherwise. In contrast, whether such barriers will be higher within democratic pairs than within autocratic pairs depends on the preferences of the leaders in the different regimes. When autocrats are more (less) protectionist than democratic executives, mutual reductions in trade barriers will be smaller (greater) between autocracies than between democracies (Proposition 3 in the Appendix). Institutional differences alone do not distinguish between the outcomes; the structure of preferences is also important. The same is true for mixed pairs relative to autocratic pairs. When autocrats are protectionist relative to the democratic legislature, pairs of autocracies will be unable to lower barriers as much as mixed pairs (Proposition 4 in the Appendix). The absence of a legislature that is even more protectionist than the autocrat means the two autocracies do not have to compromise as much as otherwise, thus reinforcing the counterintuitive point that a protectionist legislature helps promote freer trade outcomes. In the Appendix, we also show what happens when the foreign country is able to make the first offer. On the whole, our results from above continue to hold. Under these circumstances, however, the foreign government has greater bargaining power. In each case, the equilibrium outcome now lies closer to the foreign government’s ideal point. In Figure 1, for instance, the outcome is now a, instead of a. With A proposing, she knows that A will accept any point on the contract curve and at or below the intersection of her indifference curve through the Nash trade war equilibrium, IA; thus A * proposes the set of trade barriers closest to her ideal point that A will also accept, i.e., point a. In Figure 2a, the outcome is m; in Figure 2b, it is m. In Figure 2c, the outcome is now λ, since this is the point where the highest indifference curve of the foreign autocrat IA* is tangent to the home legislature's indifference curve. In the case of two democracies, as in Figure 3, P proposes d, which both the home executive and the two legislatures accept. In effect, changing the proposer moves the outcome along the contract curve closer to the proposer’s ideal point, improving the outcome for the proposing country at the expense of the other country. These outcomes, displayed in Figure 4, show that the length of the ray 0d (aggregate barriers under joint democracy) is less that of the rays 0m (aggregate barriers under the mixed case with a protectionist or a moderate autocrat) and 0λ (aggregate barriers in the mixed case with a liberal autocrat). As pointed out above, the equilibrium outcome of any game in which the two countries divide the concessions equally (which we call the SPLIT game) is always in the middle of this range of the contract curve bounded by the maximum value that each can get when she proposes. For example, in the two democracy case (Figure 1), the SPLIT outcome is in the middle of the portion of the contract curve bounded by d and d. Hence, our results hold over a range of bargaining structures. In sum, regime type is likely to exert an important influence on trade policy. Our model predicts that pairs of democracies should trade more freely than mixed pairs, a result that arises from the tendency for a protectionist legislature to actually enhance the prospects of trade barrier reductions. However, our model is silent on whether commerce will be more open between autocratic pairs than between either democratic or mixed pairs, an issue that hinges on the trade policy preferences of decision-makers. Our model is also silent on the effects of variations within regime types. It does not yield predictions about how institutional differences either among autocracies or among democracies (e.g., whether the government is presidential or parliamentary) affect trade policy, issues that have been treated at length in previous studies at least with regard to democracies (e.g., Mansfield and Busch 1995; Milner and Rosendorff 1996, 1997; Milner 1997; Pahre 1997). Here, we focus solely on the institutional differences between autocracies and democracies. Central to our analysis is the assumption that, unlike autocracies, democracies are marked by a legislature that can effectively constrain the chief executive and that is more protectionist than the executive. Following Lemma 2 in the Appendix, it is clear that given our symmetry assumption in the case of two democracies, aggregate barriers are tP irrespective of the location of tC. Hence, it is a direct consequence of Lemma 2 that variations in the degree of division between the executive and the legislature in the democracies will have no effect on the outcome when two democracies bargain. This means that our claim that joint democracies prefer lower barriers than do mixed pairs is preserved even if the degree of protectionism of the legislatures in the joint democratic case rises. Put more formally as a corollary to Lemma 2 and proved in the Appendix: Corollary 1: Irrespective of which country makes the first offer, as the legislatures become more protectionist, the aggregate level of barriers agreed to by a pair of democracies is unchanged. EMPIRICAL ANALYSIS We now turn to an empirical analysis of the hypotheses derived from the preceding model. Recall that our central hypothesis is that trade barriers should be lower between two democracies than between an autocracy and a democracy. It is not possible to test this proposition directly, since reliable data on bilateral trade barriers are not available for many of the countries analyzed here. But data on bilateral trade flows are readily available for most countries and can serve as a proxy for trade policy. Various studies have found that, other things being equal, there is an inverse relationship between the height of trade barriers and the volume of interstate trade (e.g., Harrigan 1993; Leamer 1988; Rodrik 1994, 72; Trefler 1993). We therefore test the propositions stemming from our model by analyzing the effects of regime type on bilateral trade flows and controlling for various other known influences on interstate commerce. To this end, we begin with a gravity model of bilateral trade, which predicts the amount of trade between two countries that should occur in the absence of trade barriers. It includes the national income and population of both trading partners, as well as the geographic distance between them. Existing research indicates that the gravity framework -which can be derived from a broad class of international trade models -has proven quite successful in explaining the flow of interstate commerce (Anderson 1979; Bergstrand 1985, 1989; Deardorff 1998; Eichengreen and Irwin 1995; Feenstra, Markusen, and Rose 1998; Frankel 1993; Frankel, Stein, and Wei 1995; Gowa 1994; Helpman and Krugman 1985; Linnemann 1966; Mansfield and Bronson 1997; Pollins 1989). We extend the gravity model in the following way: log Xij = logβ0 + β1log(GDPi × GDPj) + β2log(POPi × POPj) + β3logDISTij + β4MIXEDij + β5AUTij + β6OTHERij + β7ALLYij + β8PTAij + β9MPij + β10(ALLYij × PTAij) + β11(ALLYij × MPij) + β12(PTAij × MPij) + β13GATTij + β14COLij + β15COMij + β16WARij + β17lagged log Xij + log zij. (2) In equation (2), Xij is the value of exports from state i to state j in year t. Furthermore, GDPi and GDPj are these states’ respective gross domestic products (GDPs) in year t-1, POPi and POPj are their respective populations in t-1, and DISTij is the geographical distance between them. We take the natural logarithm of each of these variables because the gravity model’s underlying functional form is multiplicative (Anderson 1979; Bergstrand 1985, 1989; Deardorff 1998). Entering the trading partners’ national incomes and populations (or, equivalently, their per capita incomes) in product form is common and accords with theories of international trade based on models of imperfect competition (Eichengreen and Irwin 1995; Frankel 1993; Frankel, Stein, and Wei 1995; Helpman and Krugman 1985). These theories indicate that β1 should be positive, and that β2 and β3 should be negative. Ideally, we would use a measure of regime type that directly captures the extent to which countries have a legislature that can constrain the chief executive and affect trade policy choices. Unfortunately, such a measure does not exist, so we rely on a wellknown index of regime type developed by Gurr and his colleagues (Jaggers and Gurr 1995; Gurr et al. 1989). This measure, which has been used repeatedly in studies of international relations (e.g., Gowa and Farber 1995; Mansfield and Snyder 1995; Oneal and Rusett 1997; Russett 1993), emphasizes five institutional factors that distinguish democracies from autocracies: (1) the competitiveness of the process through which a country’s chief executive is selected, (2) the openness of that process, (3) the extent to which institutional constraints exist on a chief executive’s decision-making authority, (4) the competitiveness of political participation within a country, and (5) the degree to which binding rules govern political participation within it. One element of this measure is directly related to our central argument: states are coded as increasingly democratic as the chief executive faces greater institutional constraints, including those emanating from the legislature. But the other components of it are important as well. If political leaders are not elected competitively, then the legislature will be in a much weaker position vis-à-vis the executive. As Shumpeter (1942) observed, unless leaders can lose office, democratic constraints on their behavior -including those imposed by the legislature -are likely to be ineffective. The degree of competitiveness of elections and participation, then, are telling indicators of the extent of constraint that institutions like legislatures can exercise. Since Gurr’s data do not directly tap whether the chief executive needs legislative ratification to enact policies, the following analyses provide only an indirect test of our model. However, we are unaware of any data that more directly measure legislative control over the head of state. Moreover, this index seems entirely appropriate for distinguishing between regimes with no legislature or an ineffective one and regimes with a popularly-elected legislature that can constrain the executive. To measure regime type, Jaggers and Gurr (1995) create an 11-point measure of a state’s democratic characteristics (DEMOC) and an 11-point measure of its autocratic characteristics (AUTOC). They then derive a variable, REG = DEMOC AUTOC, which takes on values ranging from -10 to 10. Initially, we define countries for which REG ≥ 6 as coherent democracies, those for which REG ≤ -6 as coherent autocracies, and all remaining regimes as incoherent. After generating an initial set of estimates, we also assess the robustness of our results to this coding procedure. Relying on these criteria, we code each state in the sample as democratic, autocratic, or incoherent in year t-1. In equation (2), MIXEDij is a dummy variable that equals one if i and j include a democracy and a autocracy and zero otherwise; AUTij equals one if both i and j are autocracies and zero otherwise; and OTHERij equals one if either i or j is an incoherent polity and zero otherwise. The reference category is a pair of democratic countries. Since our model indicates that greater trade liberalization should occur between democratic pairs than between pairs that include a democracy and an autocracy, we expect β4 to be negative. Our model, however, yields no determinate prediction about the sign of either β5 or β6. Besides national income, population, distance, and regime type, it is also important to take account of variables that might be responsible for any observed relationship between regime type and bilateral trade. As we explain in more detail below, our analysis centers on the period after World War II. This era was marked by the existence of two major alliance blocs, one made up of autocracies (the Warsaw Pact) and the other made up almost entirely of democracies (the North Atlantic Treaty Organization). To ensure that our results do not simply reflect Cold War rivalries, we include an array of international political variables in equation (2). For instance, there is evidence that democracies were especially likely to ally with one another during the Cold War (Siverson and Emmons 1991). Furthermore, various studies have found that alliances promote trade, suggesting that alliances might account for any observed effects of regime type on trade flows (Gowa 1994; Mansfield and Bronson 1997). In the same vein, recent research indicates that, since World War II, pairs of democracies have been much more likely to establish preferential trading arrangements (PTAs) than other pairs (Mansfield, Milner, and Rosendorff 1998). And it is widely recognized that such arrangements tend to promote the flow of commerce between member-states (Frankel 1993; Frankel, Stein, and Wei 1995; Gowa 1994; Linnemann 1966; Mansfield and Bronson 1997; Pollins 1989). Thus, we need to control for the effects of alliances and PTAs. To this end, we include ALLYij and PTAij. The former variable equals one if i and j are allied in year t-1 and zero otherwise. The latter variable equals one if i and j belong to the same PTA in t-1, and zero if they do not. In addition to these factors, prior studies have found that states trade more extensively if they are members of the General Agreement on Tariffs and Trade (GATT), if they include a major power, if they have command economies, if they are not involved in a war, and if they had a prior colonial relationship (Gowa 1994; Kleiman 1976; Pollins 1989; Mansfield and Bronson 1997). As such, we include the following dummy variables, each of which is measured in t-1: MPij equals one if either i or j is a major power and zero if neither state is a major power; GATTij equals one if both i and j are parties to the GATT and zero otherwise; COLij equals one if i and j had a colonial relationship that concluded in or before t-1 (and after World War II’s onset) and zero otherwise; COMij equals one if both i and j have command economies and zero if they do not; and WARij equals one if i and j are at war and zero if they are not. 19 By including all of these variables, we control for the most likely international political influences on trade relations during the Cold War. Earlier research indicates that states belonging to the same alliance and the same PTA conduct more trade than countries that are either allies or PTA members, but not both. Furthermore, there is evidence that allies and PTA members trade more extensively if at least one state is a major power (Mansfield and Bronson 1997). As a result, we introduce ALLYij × PTAij, ALLYij × MPij, and PTAij × MPij in equation (2). To account for any temporal dependence in bilateral trade flows, we include an instrument for the lagged value of log Xij. 20 We also account for any country-specific or year-specific effects on trade by including dummy variables for all but one country and for all but one year analyzed here. Finally, log zij is a stochastic error term. Our sample is made up of all pairs of states listed as members of the interstate system by the Correlates of War Project (Singer and Small 1994) in 1960, 1965, 1970, 1975, 1980, 1985, or 1990 for which complete data are available. The independent variables in our model are measured in these years (t-1) and the value of exports from i to j is measured one year hence (in years t, which are 1961,..., 1991) to minimize any potential problems of simultaneity. We do not analyze years prior to 1960 because data on GDP and exports are especially limited for many autocracies until that point. Including the available data for these years could bias our results by systematically under-representing dyads that are not jointly democratic. After pooling these data, ordinary least squares is used to generate estimates of the parameters in equation (2). Tests of their statistical significance are based on White heteroskedasticity-consistent standard errors (White 1980). Estimates of the Parameters Our initial results, which are reported in the first column of Table 1, accord with the central proposition advanced earlier. The estimate of MIXEDij is negative and statistically significant, indicating that trade flows between democratic states tend to be greater than trade flows between democracies and autocracies. Moreover, the predicted difference in trade between democratic pairs and mixed pairs is relatively large: on average, dyads composed of a democracy and an autocracy engage in roughly 15% to 20% less commerce than those composed of two democracies. [Table 1 about here] As noted earlier, our model does not yield determinate predictions about whether trade will be more open within autocratic pairs than within either democratic or mixed pairs. The results in the first column of Table 1 provide little evidence that the flow of commerce between autocracies differs markedly from the flow between democracies, since the estimate of AUTij is not statistically significant. However, autocratic pairs trade much more extensively than mixed pairs. The estimate of AUTij is considerably larger than the estimate of MIXEDij and the difference between them is significant (t = 5.79). Finally, the estimate of OTHERij is negative. Pairs including an incoherent polity conduct significantly less trade than democratic or autocratic dyads, although they engage in significantly more commerce than mixed pairs. To assess the robustness of these results, we conduct a number of additional tests. First, our model emphasizes how domestic legislatures constrain the trade policy choices of executives in democracies. But in countries that have recently undergone transitions to democracy, legislative institutions are often weak and poorly developed (e.g., Haggard and Kaufman 1995; Haggard and Webb 1994). Hence, trade policies set by a pair of mature democracies may differ from those set by a country-pair that includes a nascent democracy with a weakly institutionalized legislature. To address this issue, we include DEMZij in equation (2). It equals one if either i or j experienced a democratic transition between years t-6 and t-1 (i.e., if either state is not a coherent democracy in t-6 and if both states are democratic in t-1) and zero otherwise. For present purposes, AUTij, MIXEDij, and OTHERij are set equal to zero if DEMZij equals one. As shown in the second column of Table 1, pairs including a country that recently experienced a democratic transition conduct significantly less trade than pairs composed of two mature democracies, since the estimate of DEMZij is negative and statistically significant. Furthermore, when DEMZij is included in the analysis, there is even stronger evidence that democratic pairs trade more extensively than mixed pairs. The estimate of MIXEDij continues to be statistically significant and its absolute value is noticeably larger than before. Second, we address the robustness of the preceding results with respect to the coding of democratic and autocratic regimes. Thus far, we have coded states as democratic if REG ≥ 6 and we have coded them as autocratic if REG ≤ -6. These operational definitions have been used in other studies; however, they are clearly somewhat arbitrary and it is important to determine if relaxing them affects our findings. We therefore estimate equation (2) after redefining these thresholds as: (a) 5 and -5, (b) 4 and -4, (c) 3 and -3, (d) 2 and -2, and (e) 1 and -1. Table 2 shows the estimates of MIXEDij, AUTij, and OTHERij for each set of thresholds. (The estimates of the remaining parameters are not presented to conserve space.) These results provide further indication that our earlier findings are quite robust. The estimate of MIXEDij is negative and statistically significant regardless of which set of thresholds is used and its size does not vary much. Furthermore, the estimate of AUTij continues to be positive and there is only one case in which it is statistically significant. The estimate of OTHERij continues to be negative and it is significant in two instances. [Table 2 about here] Besides altering the thresholds that distinguish between regime types when Jaggers and Gurr’s (1995) measure (REG) is used, it is important to examine whether our results differ if a different measure is used. Various alternative measures exist, but very few of them cover the range of countries and years analyzed here. Alvarez et al. (1996), however, code the regime type of most countries in our sample on an annual basis during the period from 1950 to 1990. Unlike Jaggers and Gurr, Alvarez et al. identify each state as either democratic or autocratic in every year. Their measure has certain obvious advantages, since we do not need to establish quantitative thresholds to code regime type. From our standpoint, however, it also has certain drawbacks, especially Alvarez et al.’s (1996, 5) intentionally narrow definition of democracy, which centers solely on the competitiveness of the political process. In their view, democracy exists when there are competitive elections for the head of state and the legislature; otherwise, the regime is autocratic. As noted above, the competitiveness of the political process helps to shape whether the legislature can effectively constrain the chief executive. In contrast to Jaggers and Gurr, however, Alvarez et al. do not directly measure this factor. Nonetheless, it is useful to analyze Alvarez et al.’s data in order to assess the robustness of our initial results. To this end, we re-estimate equation (2) after using their data to code MIXEDij, AUTij, and DEMZij. Note that OTHERij is not included in this analysis, since Alvarez et al. code each state as either a democracy or an autocracy. The results are presented in the third column (omitting DEMZij) and the fourth column (including DEMZij) of Table 1. Clearly, there is considerable agreement between these findings and those generated using Jaggers and Gurr’s data. Particularly important for present purposes is that the estimate of MIXEDij continues to be negative and statistically significant, although its absolute value is somewhat smaller when we rely on Alvarez et al.’s data rather than Jaggers and Gurr’s data. Furthermore, we again find that democratic transitions dampen trade, since the estimate of DEMZij is negative and significant. It is also noteworthy that no statistically significant difference exists in the flow of trade between autocratic pairs, on the one hand, and between democratic pairs, on the other. As in the first two columns in Table 1, the estimate of AUTij is not significant, although it is negative when Alvarez et al.’s data are used and it is positive when Jaggers and Gurr’s data are employed. In sum, then, our results are quite robust with respect to the coding of regime type. Third, it is useful to ensure that our statistical design does not mask any substantial intertemporal variation in the effects of regime type on trade. In a preliminary effort to address this issue, we conduct separate analyses of each decade included in our sample, by pooling the observations for 1960 and 1965, for 1970 and 1975, and for 1980 and 1985, and by analyzing 1990 alone. For each set of observations, we estimate equation (2) using the same operational measure of regime type that was used to generate our original estimates (i.e., states are coded as democratic if REG ≥ 6; they are coded as autocratic if REG ≤ -6). Table 3 presents the estimated coefficients of MIXEDij, AUTij, and OTHERij. (As in Table 2, the remaining coefficient estimates are not reported to conserve space.) In three of the four decades we assess, the estimate of MIXEDij is negative and statistically significant, indicating that the tendency for trade to be more open within democratic pairs than within mixed pairs is relatively stable over time. On the other hand, it is interesting that any tendency for autocratic pairs to trade more extensively than their democratic counterparts has become attenuated over time, an issue that warrants additional attention in future research. [Table 3 about here] Our sample includes all countries for which we were able to obtain data for the variables in equation (2). However, there are various countries for which data on GDP, trade, or both are missing in certain years analyzed here. Since a number of these countries are autocracies that are unlikely to trade extensively due to their severely distorted economies (e.g., Albania, Cambodia, Cuba, Laos, Libya, North Korea, and Sudan), we ran a series of Heckman selection models to ensure that there was no resulting bias in our earlier findings (Greene 1993, 706-14). Based on these models, we found no evidence of such a bias. In addition to those countries that are included in our sample and are missing data for certain years, there are other countries for which we were unable to obtain any economic data. The latter states (e.g., Afghanistan, Bangladesh, Burundi, Liberia, Rwanda, Somalia, Uganda, Yemen) are excluded from our sample altogether. Virtually all of them are autocracies having highly distorted economies. Since such distortions are likely to hamper commerce, there is ample reason to expect that our results understate the actual extent to which democratic dyads trade more freely and extensively than dyads composed of an autocracy and a democracy. Equally, the omission of these countries probably helps to explain why, for example, the estimate of AUTij is positive in Table 1 when Jaggers and Gurr’s measure of regime type is considered. Finally, we address whether our results are robust with respect to the inclusion of various factors that are omitted from equation (2). Of chief importance in this regard are economic growth and the level of economic development. It is possible, for example, that democratic pairs conduct more trade than mixed pairs because democracies experience higher growth than autocracies and growth promotes trade. Similarly, democracies may be more highly developed economically and more highly developed states may engage in more trade than less developed countries. In fact, our results do not accord with either of these possibilities, since while democratic pairs trade more extensively than mixed pairs, so do autocratic pairs. Nonetheless, to further assess the robustness of our results, we include in equation (2) the real per capita GDP of both i and j in year t-1. Then, we replace per capita GDP with various other factors which have been used as measures of economic development in studies of international trade (e.g., Polachek 1980), including the per capita number of highway vehicles, per capita school enrollment, per capita university enrollment, and per capita electrical production. Finally, to examine the effects of growth, we include the change in per capita GDP of both i and j between year t-6 and year t-1. It should be noted that, for some of these variables, complete data are available for only a fraction of the states and years analyzed here, so the estimates generated after including them in equation (2) are not directly comparable to those in Table 1. But our findings indicate that including these variables has no substantive influence on the estimates in Table 1. Most important for present purposes, the estimate of MIXEDij is negative and statistically significant in each of these analyses. Turning to the remaining variables in equation (2), the effects of GDP, population, and distance accord with expectations based on the gravity model. As shown in Table 1, the estimate of log(GDPi × GDPj) is positive and statistically significant, and the estimates of both log(POPi × POPj) and logDISTij are negative and significant. In addition, the estimates of ALLYij, PTAij, MPij, ALLYij × PTAij, and ALLYij × MPij are positive; and the estimate of PTAij × MPij is negative. All of these estimates are statistically significant when Jaggers and Gurr’s data are used and all but the estimate of ALLYij × MPij is significant based on Alvarez et al.’s data. These results indicate that alliances, PTAs, and the presence of a major power each promotes commerce. Moreover, countries that belong to the same alliance and the same PTA conduct more trade than countries that are either allies or parties to the same PTA, but not both. Alliances provide a much greater impetus to trade between states that include a major power than between other states. And PTAs yield marginally more commerce between trading partners that include a major power than between other countries. Our results also indicate that a former colonial relationship heightens bilateral trade flows and that states having command economies trade extensively with each other, since the estimates of COLij and COMij are positive and statistically significant. GATT membership, however, has a weaker effect on bilateral trade flows. The estimate of GATTij is positive, suggesting that trade flows between parties to the GATT tend to be somewhat larger than trade flows between non-member states. But this estimate is relatively small and it is not statistically significant when Jaggers and Gurr’s data on regime type are used. In sum, the results of this empirical analysis support the central prediction stemming from our formal model. Holding constant a large number of economic and political factors -including factors pertaining to the dynamics of the Cold War -pairs of democracies have conducted considerably more trade than mixed pairs since World War II. Furthermore, while our model generates no definitive predictions about autocratic pairs, it is noteworthy that, based on our sample, no systematic difference exists in trade relations between these pairs and democratic dyads.