The ~Ir!Relevance of Militarized Interstate Disputes for International Trade

Transcription

1 International Studies Quarterly ~2002! 46, The ~Ir!Relevance of Militarized Interstate Disputes for International Trade Quan Li and David Sacko The Pennsylvania State University Do military disputes between two states suppress trade between their firms? Both liberals and realists suggest that conflict occurrence reduces bilateral trade. However, using a rational expectation argument, Morrow ~1999! proposes that conflict occurrence and trade should be uncorrelated statistically. Empirical evidence to date both supports expectations and appears contradictory and inconclusive. We offer a theory that reconciles, encompasses, and extends the competing arguments, explaining the empirical inconsistency. By incorporating rational expectations and uncertainty into the profit calculus of trading firms, the theory identifies the conditions under which various properties of a conflict ~onset, duration, and severity! should and should not reduce bilateral trade ex ante and ex post. We test the ex post effects in two datasets that cover either a wider range of countries or a longer time period than previous quantitative studies. Both an unexpected MID onset and the unexpectedness of a MID onset reduce bilateral trade substantially ex post. Preliminary tests suggest that MID duration and severity also affect bilateral trade ex post. We conclude by discussing the implications of our research. Do militarized disputes between two states suppress trade between their firms? Conventional wisdom from both liberal and realist approaches suggests that military conflict occurrence reduces bilateral trade. However, Morrow s ~1999! rational expectation argument claims that conflict occurrence and trade should be uncorrelated statistically. Empirical evidence to date both supports expectations and appears contradictory and inconclusive. We offer a theory that explains the empirical inconsistency. While our theory centers on the importing and exporting firms, it allows state intervention to impact trade by affecting the profit calculus of the firms, hence integrating liberal and realist conceptions. Further, by incorporating both rational expectation and uncertainty into the profit calculus of trading firms, the theory identifies the conditions under which various properties of a conflict ~onset, duration, and severity! should and should not reduce bilateral trade both ex ante and ex post. Hence, the theory reconciles, encompasses, and extends insights from competing theoretical arguments. We test our theoretical arguments about the ex post effects of conflict using two datasets that cover either a longer time period or a wider range of countries than previous studies in this research program. The two datasets include one based on Angus Maddison s ~1995! GDP data ~about 42,934 dyad-years, based on Authors note: An earlier version of this paper was presented at the annual meeting of the Peace Science Society ~International!, Ann Arbor, MI, We thank Scott Bennett, Stuart Bremer, Bill Dixon, Erik Gartzke, Robert Pahre, two anonymous referees, the editors of ISQ, and participants at the Department of Political Science Speaker Series at the Penn State University for helpful comments. Scott Bennett and Dong-Joon Jo helped with data International Studies Association. Published by Blackwell Publishers, 350 Main Street, Malden, MA 02148, USA, and 108 Cowley Road, Oxford OX4 1JF, UK.

2 12 The (Ir)Relevance of Militarized Interstate Disputes 1,410 dyads, 56 countries from 1870 to 1992! and one postwar sample ~about 81,212 dyad-years, based on 4,340 dyads, 120 countries from 1950 to 1992!. The statistical results produce strong and robust support for our theory. This research has important theoretical and methodological implications for studying the conflict-trade nexus. To explain how conflict affects trade, we need to understand not only how traders assess conflict-associated risks but also the causal mechanisms of the conflict process. The various theoretical arguments are seemingly competing but are in fact complementary. Modeling the conflict-trade interaction should consider the distinction between the ex ante and ex post effects. We shall return to these topics in the conclusion. The paper proceeds as follows. We first review the mixed empirical findings and competing theoretical claims about the effect of conflict on trade. Then, we elaborate a theoretical logic for how interstate militarized disputes suppress bilateral trade, followed by a short overview of other determinants of bilateral trade. Next, we discuss the research design for the empirical tests. Finally, we present statistical results and conclude with a discussion of their implications. Effect of Conflict on Trade: Mixed Findings, Competing Arguments In two pioneering studies, Pollins ~1989a, 1989b! finds that the relative cooperativeness or hostility of political ties between two states affects their bilateral trade flows. His measure of relative cooperation in diplomatic relations is based on the net annual cooperation or conflict of diplomatic events from the Conflict and Peace Data Bank ~COPDAB! data ~Azar, 1980!. Later dyadic studies ~Gowa and Mansfield, 1993; Mansfield and Bronson, 1997; Morrow, Siverson, and Tabares, 1998; Bliss and Russett, 1998; Barbieri and Levy, 1999! examine explicitly how militarized interstate disputes or wars alone impact bilateral trade, but they reach different conclusions. Some researchers find that war reduces bilateral trade in general ~Mansfield and Bronson, 1997!, or only within certain years or dyads ~Gowa and Mansfield, 1993; Barbieri and Levy, 1999!. Others find that militarized interstate disputes ~MIDs! have no statistically significant impact on bilateral trade ~Morrow, Siverson, and Tabares, 1998, 1999!, or that they decrease trade only for certain years but not for a pool of dyads over time ~Bliss and Russett, 1998!. The empirical contradiction appears to correspond with several competing theoretical arguments. Consistent with the liberal approach and focusing on importers, Pollins ~1989a, 1989b! argues that both economic and political factors enter the utility function of the trading firms with theoretically coequal status. Due to traditional security concerns, importers desire to reward friends, punish adversaries, and minimize risk. Hence, they tend to choose sellers from friendly nations. A contrasting realist approach is state centric, focusing on state intervention in international trade ~Gowa, 1989; Gowa and Mansfield, 1993!. According to this view, international trade increases the wealth of both trading states, providing the governments with an opportunity to turn gains from trade into military capability. The resulting security externality is negative for two ~potential! adversaries and positive for two allies. It follows that each state has an incentive to manipulate its trade pattern, restricting commerce with its enemy while promoting that with its friend. Pollins micro-economic actor, liberal approach and Gowa and Mansfield s state-centric realist approach, while appearing to compete, are in fact complementary. According to Pollins conception, state intervention is implied because it is the state s responsibility to guarantee security for firms. From Gowa and Mansfield s perspective, that firms trade is assumed, because it is the exporting and importing firms that match supply with demand, clear the market, and create wealth. Despite their difference in conceptualization, both approaches imply the same expectation, that the occurrence of military conflict should reduce trade.

3 Quan Li and David Sacko 13 More recently, Morrow ~1999! develops an innovative third theoretical argument for how conflict affects trade, with a provocative empirical claim that military conflict occurrence and bilateral trade should be uncorrelated statistically. According to Morrow, the quality of political relations between two states is positively correlated with the volume of their bilateral trade. But military conflict negatively correlates with the quality of interstate political relations, acting as a specific realization of the underlying poor interstate relations ~Morrow, 1999:88!. Relying on some measure of the quality of political relations, traders form expectations about the probability of military conflict and the profitability of trade. Where they expect higher risks of disruption and negative expected returns due to potential conflict, they back out of the market. Before conflict actually occurs, trade is already suppressed ex ante, leading to a statistical finding of no relationship ~Morrow, Siverson, and Tabares, 1998, 1999!. 1 Morrow is correct on the role of rational expectations through which military disputes reduce bilateral trade ex ante, but other factors may also work to cause observable ex post effect of conflict on trade. Below, we elaborate a theory that identifies specific conditions under which conflict should and should not affect trade. We also identify where our theory builds upon previous theoretical arguments and discuss why previous studies reach contradictory findings in light of the new theory. A Theory on How Interstate Military Disputes Affect Inter-Firm Trade The theory consists of four basic arguments. First, it is the profit-maximizing exporting and importing firms, not governments, that trade across national borders. Government intervention has to pass through the profit calculus of the trading firms to affect commercial transactions, even where the state forbids trade with another state. Second, as firms anticipate the interstate conflictassociated risks and internalize the related transaction costs, conflict can reduce trade even before its occurrence ~i.e., ex ante effect on expectations!. Third, because firms are not omniscient and cannot fully anticipate all conflict-related risks, firms seeking to maximize profit over more than one period revise continuously both their expectations and behaviors ex post. Where such risks are poorly anticipated, traders revise the expected returns downward in light of the new information a conflict reveals. Fourth, for traders who need to quantify risks in monetary terms, conflict-related risks not only take account of when a conflict occurs but also how long the conflict lasts and how severe it becomes. The onset, duration, and severity of a conflict affect the expectations and behavior of the trading firms both ex ante and ex post. 1. Firms and States: Calculus of Profit from Trade and Interstate Conflict States fight and firms trade. While states fight for survival, firms trade for profit. As long as international commercial exchanges remain profitable, firms will continue to trade internationally. Therefore, for a militarized dispute to suppress trade between two firms residing separately in two belligerent countries, the firms have to suffer a current loss of trade due to the dispute, or the firms expect a decline in the future profitability such that it is no longer profitable to continue to trade with the partner in the other country. The profit of a trading firm is affected by changes in government behavior or other market conditions, and its expected profit by its expectations of future government behavior or other market conditions. Hence, we need to specify explicitly how conflict may lead to 1 According to Morrow ~1999!, military conflict is sometimes statistically significant only because of measurement errors in existing measures of the quality of political relations.

4 14 The (Ir)Relevance of Militarized Interstate Disputes adverse changes in government restrictions or other market conditions that reduce trade. When two states fight with each other, each wants to increase its own chance of winning. If either government perceives the commercial exchanges between the firms in the two countries as harmful to its military cause, it has incentives to impose trade restrictions. The government may select from a menu of choices, including trade-related current account and capital account restrictions, economic sanctions, embargoes, and other export controls. 2 At minimum, the changes in government restrictions raise the transaction costs of trade between the firms subject to the legal jurisdictions of the two governments. Trading opportunities that would otherwise be profitable may now become unprofitable. Firms have to look for profitable trade elsewhere, such as trading with third countries. At maximum, a government can forbid any interfirm trade with its enemy. This is the only case where, even if bilateral trade is profitable, it may be recorded as zero trade in the national balance of payments. However, firms will still trade by smuggling, as long as the market demand exists such that they can charge an exorbitant risk premium and as long as they can bypass the government ban to deliver the goods and collect the payments. While smuggling exists, such trade does not show up in a country s balance of payments. 3 Even without raising government restrictions, military violence still can disrupt trade. Interstate conflict may interfere with transportation, communication, and the smooth functioning of the market, resulting in delays in the delivery of goods or unexpected damages. Even low-intensity interstate conflict may cause panics in the financial markets and problems for the payments systems. Investors may flee, causing a credit crunch for traders; national currencies may suffer speculative attacks, raising the volatility and reducing the availability of foreign exchange. These occurrences produce impediments to the settlement of contracts. Where the forward currency market is not well developed, the disruption of trade can be particularly serious. These market disruptions quickly add up the transaction costs for traders, vaporizing those trading opportunities of narrow profit margins only. Therefore, inter-firm trade declines relative to its peacetime level where interstate conflict leads to the imposition of government restrictions or market disruptions. This, however, begs the question of what characteristics of military conflict might cause the decline in trade. We argue that in addition to its occurrence, the dispute duration and severity should also correlate positively with government restrictions and market disruptions. The longer a dispute lasts or the more severe it becomes, the more likely the governments will impose trade restrictions, and the more serious market disruptions are. In essence, the onset, duration, and severity of a dispute all induce risks to trade, raise transaction costs, and lower returns. But previous studies that estimated the effect of military conflict simply used a dummy vari- 2 If one has data on government choices along this continuum of restrictions, then these government-related effects of military disputes upon trade can be examined by looking at the effects of dispute-induced restrictions. Studies of sanctions do show this connection ~Hufbauer, Schott, and Elliott, 1990!. But researchers simply do not have data on the whole range of government options. Focusing on one type of government restrictions instead of military disputes is likely to underestimate the effects on trade because there are many other ways through which a government might restrict trade in the event of a military dispute or even an expected conflict. 3 Note that traders do not simply compare profit from trading with one partner with profit at home. Instead, they also take account of other profit opportunities that may exist in the system. War can potentially make legitimate trade with third parties more profitable and more likely. The neutral trade in the Napoleonic Wars ~U.S., Scandinavia! and WWI ~Norway, U.S. until entry! are good examples. While the bilateral trade models do not capture such trade diversification, it is consistent with the notion of profit-maximizing firms in our story. We thank Robert Pahre and an anonymous referee for bringing this to our attention.

5 Quan Li and David Sacko 15 able measuring the presence or absence of a dispute or war. We shall elaborate the effects of dispute duration and severity later. 2. Ex Ante Effect of Conflict: Role of Rational Expectations Government restrictions or market disruptions never have to materialize for inter-firm trade to decline owing to rational expectations. Forward-looking trading firms operate based on the expected profit rate and must hedge against risks. Ex ante, traders assess the probability of imminent government restrictions and other dispute-related market disruptions. When they expect the risks to rise, firms will internalize them by seeking additional insurance in negotiating future trade contracts. The extra cost makes some trade no longer profitable. Hence, even in the absence of actual government restrictions and market disruptions, rising expectations of their future occurrences can cause a decline in future trade. 4 The ex ante aspect of our theory builds on the innovative argument by Morrow ~1999! for why trade and conflict should be uncorrelated statistically, which we reviewed in detail earlier. Though Morrow s argument explains why actual military conflict may appear to have no statistical effect on trade, his theory is limited both empirically and theoretically. His theory is not consistent with the recurrent finding in many studies that conflict does frequently reduce trade. This empirical limitation may be a function of two weaknesses in Morrow s analysis. Rational firms are implicitly assumed to have perfect foresight and care only about the risk of conflict occurrence. Uncertainty and other conflictrelated risks are absent from his theory. Below, we relax these two restrictions. 3. Ex Post Effect of Conflict: Role of Uncertainty In addition to the ex ante effect on expectations, military disputes also generate ex post effects on trade between the firms of two countries in conflict. The ex post effects of conflict rely on three assumptions. First, we assume traders cannot fully anticipate military disputes ex ante. As a result, imperfect forecasting leads to the ex ante failure of firms to diversify or internalize all dispute-related risks completely. Second, we assume firms seek to maximize profit over more than one period. Hence, they continuously update their expectations of future profitability and revise behaviors accordingly. Third, the pre-conflict trade level must be greater than zero; otherwise, a dispute becomes irrelevant to trade. Indeed, all three assumptions are plausible and defensible. The first assumption is consistent with an analytical insight from recent scholarship in international conflict studies; that is, the occurrence of the militarized dispute should be treated as a stochastic process, involving incomplete information and the signaling of resolve at the crisis-bargaining stage. Hence, war must be unexpected by politicians for it to occur ~Fearon, 1995; Gartzke, 1998; Morrow, 1989, 1999!. Since traders do not possess better information than politicians, war is unexpected by them as well. Hence, traders cannot predict the occurrence of military violence perfectly ~even based on a perfect measure of the quality of political relations!. Moreover, the first assumption goes beyond the above insight about conflict onset and probes other characteristics of the conflict process. It implies that traders are concerned about but cannot fully anticipate all conflict-related risks, including those related to its onset, duration, and severity. Like conflict onset, dispute duration and severity can also be caused by failures of interstate bargain- 4 It is worth noting that the expectations always affect the terms of contracts for future trade, but not the current trade because firms are legally bound to honor their commitments in existing contracts.

6 16 The (Ir)Relevance of Militarized Interstate Disputes ing over contentious issues, where such bargaining failures are motivated by uncertainty. As is the case of conflict onset, the continuation and intensity of a dispute are also means for governments to demonstrate their resolve and acquire more information about the strategic variables ~the preferences and capability constraints! of the other side. Where states cannot reach an agreement over a salient issue, they are going to fight longer or harder. To the extent that uncertainty and incomplete information are responsible for the lack of peaceful settlement, they may also motivate the length and severity of a dispute. Hence, as with conflict onset, traders cannot perfectly anticipate dispute duration or severity, either. Because military disputes are more or less unexpected, the pre-conflict trade level is unlikely to be zero because of poor political relations alone. Unless poor political relations lead to a government ban, causing zero trade in reported statistics ~as the government does not count smuggling!, the market simply equilibrates to some low but nonzero level of trade where supply equals demand. Then when a dispute occurs, it may result in actual government restrictions or other costs and cause firms to reassess downward their future trade profit. The consequence is a further decline of the already low expected profitability, deterring possible future contracts. 5 As long as traders lack perfect foresight about all conflict-induced risks, their ex ante and ex post expectations of trade returns are not likely to be identical. The development of a dispute keeps on feeding new information into the expectations of traders about possible future government intervention or market disruption. This allows conflict to affect trade ex post, given new information about its onset, duration, and severity. While the ex ante effects of disputes are derived from expectations, the ex post effects may be both actual and expectational. When military violence breaks out, governments have stronger incentives immediately to restrict commercial activities with the enemy state, unless there is no trade between them to curb. Financial markets typically react negatively to the news, also constraining trade. Owing to the actual government restrictions or other market disruptions, transaction costs of trading between firms of adversarial states rise, producing a suppressive effect. In addition, as discussed earlier, even in the absence of actual restrictions, the mere occurrence of a dispute can lead to an upward revision by firms of future risks, eliminating some currently profitable trade. 6 To illustrate our theory better, we offer the following thought experiment. Decompose an actual dispute into a systematic component traders capture by forecasting based on all available information and a stochastic component driven by the degree of uncertainty. The systematic component measures the accuracy of ex ante rational expectations while the stochastic component reflects the amount of new information ex post. Depending on the relative size of the systematic and stochastic components, traders may be surprised more by some disputes and less by others. Conflicts are heterogeneous in terms of the ex post informational content. Intuitively, for a largely unexpected dispute, firms underestimate and fail to insure against most dispute-related risks ex ante. Therefore, a dispute whose properties are more unexpected causes greater unexpected disruptions and upward revisions of firms expectations of future risks, generating larger ex post effects. 5 Of course, when peace resumes and lasts after the conflict, traders may revise downward the expected future risks and trade recovers, as frequently observed empirically. 6 There is no doubt that in the wake of a conflict, governments may attempt to soothe market expectations with promises of no restrictions. But such promises are rarely credible in the eyes of forward-looking firms because of the time inconsistency problem. After firms sign contracts, governments can always renege and impose restrictions ex post over bilateral trade, resorting to the persuasion of national interest to justify policy changes and to coerce compliance. Expecting the possibility, traders will always internalize the risks, garnering higher transaction costs.

7 Quan Li and David Sacko 17 Our theory suggests that the ex ante and ex post effects of conflict on trade are inversely related in size, which is relevant to how the effects of conflict on bilateral trade should be tested. Instead of using the actual dyadic dispute involvement in the statistical model, accurate empirical tests of the impact of militarized disputes should distinguish between the more unexpected and the less unexpected. The less unexpected dispute should be uncorrelated to trade because of the ex ante effect of expectations. However, the more unexpected dispute should have a statistically significant negative effect on trade. Hence, our argument reconciles the mixed empirical findings, that is, the actual conflict sometimes has no appreciable effect on trade ~as the ex ante effect must have been large! while other times it does reduce trade ~as the ex post effect must have been large!. Below, in light of the theory, we elaborate the ex ante and ex post effects of conflict s onset, duration, and severity, respectively, and generate testable hypotheses. 4. Dispute Onset: Ex Ante and Ex Post Effects Table 1 summarizes our theoretical expectations of the ex ante and ex post effects of interstate military dispute onset on bilateral trade. In cell ~1!, because traders have internalized the risks before conflict occurs, bilateral trade is already suppressed ex ante while the ex post effect is small. This accords with Morrow s ~1999! argument that conflict and trade are statistically uncorrelated. In cell ~2!, bilateral trade is suppressed ex ante. But because the dispute does not happen as expected, traders revise upward the expected profitability of future trade with no new suppressive effect. In cell ~3!, bilateral trade suffers no political shock, as expected. In cell ~4!, the ex ante effect is small or nonexistent because of low expectation of imminent conflict. But the ex post effect is large, because the dispute onset causes unexpected government restrictions or market disruptions, hence a downward revision of the expected returns. Indeed, it is these more unexpected dispute onsets that lead to more unexpected government restrictions over trade or other more unexpected market disruptions. The unexpected dispute onsets are most likely to have observable effects on trade, as in the following hypothesis. Hypothesis 1. The more unexpected a dispute onset is, the greater it reduces bilateral trade ex post. One possibility exists that may confound the empirical test. Immediately before the occurrence of a military conflict that firms expect, they may speed up settling current contracts; they may also hoard and stockpile the goods whose availability they expect to be most affected. The empirical consequence may be a temporary rise in trade in the year of conflict. Note that this effect may be of Table 1. Ex Ante and Ex Post Effects of MID Onset over Bilateral Trade Expected Dispute Onset Likelihood More Expected More Unexpected Dispute Onset Yes No ~1! Expected Conflict Ex ante suppressive effects: Large Ex post suppressive effects: Small or No ~2! Unexpected Peace Ex ante suppressive effects: Large Ex post suppressive effects: No ~rise in trade! ~4! Unexpected Conflict Ex ante suppressive effects: Small or No Ex post suppressive effects: Large ~3! Expected Peace Ex ante suppressive effects: No Ex post suppressive effects: No

8 18 The (Ir)Relevance of Militarized Interstate Disputes little concern to us since this rise in trade actually works against finding significant negative effects of disputes on trade in the context of yearly observations. In addition, two countervailing effects may reduce the rise in trade. The preconflict stockpiling may be compensated for by the loss of trade due to the outbreak of conflict. Or in the absence of the conflict, the hoarding may be compensated for by traders demand for fewer goods in the later period when they find that the conflict does not materialize. In any event, the effects of stockpiling are likely to be smoothed over time. 5. Dispute Severity: Ex Ante and Ex Post Effects Most quantitative analyses of the impact of conflict on trade use the MIDs dataset from the Correlates of War project ~Gochman and Maoz, 1984; Jones, Bremer, and Singer, 1996!. Militarized interstate disputes, as measured, contain different types of military violence, from threat of force, display of force, use of force, to war. As briefly discussed earlier, we expect that the less severe a dispute, the smaller its suppressive effects on trade. At low levels of conflict, the governments may not feel the need to restrict commerce, or firms may not feel that the risk is high enough to charge a greater risk premium. Conversely, war is much more likely to lead to government restrictions, or simply be more disruptive to the market even in the absence of government restrictions. Aggregating military violence of different degrees of severity into one variable is not helpful for sorting out the actual impact of interstate conflict on inter-firm trade. Indeed, studies that code conflicts of different degrees of severity into a dummy variable MID usually end up finding no impact on trade ~Morrow, Siverson, and Tabares, 1998, 1999! or mixed results ~Bliss and Russett, 1998!. Studies that examine war, however, have found a statistically significant negative impact ~Mansfield and Bronson, 1997! or some mixed results ~Gowa and Mansfield, 1993; Barbieri and Levy, 1999!. Neither approach, however, fully exploits the information in dispute severity, nor do they consider how the effect of dispute severity may vary ex ante and ex post due to rational expectation and uncertainty. Table 2 summarizes our theoretical expectations of the ex ante and ex post effects of dispute severity on bilateral trade. In cell ~1!, because traders expect a severe dispute, the suppressive effect on trade is large ex ante while the ex post effect is small. In cell ~2!, the ex ante effect is also large due to the expectation of a severe dispute, but the unexpectedly mild dispute causes an upward revision of expected returns, causing a possible rise in trade. In cell ~3!, the dispute is mild as expected, causing some ex ante decline but no ex post effect. In cell ~4!, due to the expectation of a mild dispute, the ex ante effect is small or nonexistent. But as the dispute turns out to be unexpectedly severe, it causes unexpected govern- Table 2. Ex Ante and Ex Post Effects of MID Severity over Bilateral Trade Expected Dispute Severity Expectedly Severe (Unexpectedly Mild) Unexpectedly Severe (Expectedly Mild) Dispute Severity Severe ~1! Severity Expectedly Severe Ex ante suppressive effects: Large Ex post suppressive effects: Small or No ~4! Severity Unexpectedly Severe Ex ante suppressive effects: Small Ex post suppressive effects: Large Mild ~2! Severity Unexpectedly Mild Ex ante suppressive effects: Large Ex post suppressive effects: No ~rise in trade! ~3! Severity Expectedly Mild Ex ante suppressive effects: Small Ex post suppressive effects: No

9 Quan Li and David Sacko 19 ment restrictions or market disruptions, generating a large negative ex post effect. This expectation leads to the following hypothesis. Hypothesis 2. The unexpectedly more severe a dispute is, the greater it reduces trade ex post. 6. Dispute Duration: Ex Ante and Ex Post Effects Researchers studying the effect of conflict on bilateral trade flows typically measure dispute as a dummy variable in terms of its presence or absence in a year. Following this operationalization, a dispute lasting only a day is treated as if it has the same impact as one of 360 days. Obviously, the dispute lasting 360 days is more likely to induce government restrictions or other disruptions to interfirm trade between two countries. In fact, many disputes last a long time. While about 37% of the disputes in the MID dataset last for 1 day, about 60% of the disputes are longer than 1 day but shorter than a year, and about 3% of the disputes last longer than 365 days ~ Jones, Bremer, and Singer, 1996!. However, previous studies have not examined the effect of dispute duration on trade, nor have they considered how rational expectations and uncertainty interact with dispute duration to affect trade. Consistent with the theoretical logic espoused earlier, Table 3 summarizes our expectations of the ex ante and ex post effects of dispute duration on bilateral trade. In cell ~1!, traders internalize the risks based on the expectedly long dispute, causing a large ex ante effect and a small ex post effect. In cell ~2!, as firms expect a long dispute, the ex ante effect is large. But because the dispute ends quicker than expected, traders revise upward the expected returns, generating a possible rise in trade. In cell ~3!, the dispute is short as expected, causing a small ex ante effect and no ex post effect. In cell ~4!, because traders expect only a short dispute, the ex ante effect is small, but the ex post effect is large, because the dispute lasts longer than expected, causing a higher risk assessment due to unexpected government restrictions and market disruptions. This expectation leads to the following testable hypothesis. Hypothesis 3. The unexpectedly longer a dispute lasts, the greater it reduces bilateral trade ex post. 7. Sample Differences Previous quantitative studies use data that have different and limited temporal and spatial domains, which may be one reason for inconsistent empirical find- Table 3. Ex Ante and Ex Post Effects of MID Duration over Bilateral Trade Expected Dispute Duration Expectedly Long (Unexpectedly Short) Unexpectedly Long (Expectedly Short) Dispute Duration Long ~1! Duration Expectedly Long Ex ante suppressive effects: Large Ex post suppressive effects: Small or No ~4! Duration Unexpectedly Long Ex ante suppressive effects: Small Ex post suppressive effects: Large Short ~2! Duration Unexpectedly Short Ex ante suppressive effects: Large Ex post suppressive effects: No ~rise in trade! ~3! Duration Expectedly Short Ex ante suppressive effects: Small Ex post suppressive effects: No

10 20 The (Ir)Relevance of Militarized Interstate Disputes ings. Although this potential reason is not part of our theory, it is worth discussing in some detail given its relevance to the research question. For example, Mansfield and Bronson ~1997! look at all dyads from 1960 to 1990 ~with every five-year interval! within a pooled time-series cross-sectional design. Morrow, Siverson, and Tabares ~1998, 1999! study major power dyads ~pairs among the U.S., U.K., France, Russia, Germany, Japan, and Italy! from 1907 to 1990 in a pooled design as well. Bliss and Russett ~1998!, on the other hand, base their dyadic analysis upon 14 countries for 28 yearly cross-sections between 1962 and 1989 in addition to the pooled model. Gowa and Mansfield ~1993! limit their analysis to 9 yearly cross-sections between 1905 and 1985 for 7 countries ~U.S., U.K., France, Italy, Germany, Soviet Union, and Japan!. Barbieri and Levy ~1999! apply time-series analysis to 7 dyads for different periods between 1870 and To examine whether the empirical inconsistency in previous studies results from sample differences, we test the following hypothesis, using two datasets that cover ~1! a longer time period and ~2! a wider range of countries. Hypothesis 4. Involvement in a militarized interstate dispute decreases bilateral trade. Other Determinants of Bilateral Trade International trade economists usually employ the gravity model to estimate the normal level of trade between states ~Aitken, 1973; Anderson, 1979; Deardorff, 1997; Frankel, 1998!. The model posits that bilateral trade is positively related to incomes of the two trading countries and negatively related to their geographical distance and the size of their respective domestic markets. Eichengreen and Irwin ~1996! also show that bilateral trade flows exhibit path dependence and urge that we control for the prior level of bilateral trade in the context of a gravity model. Studies of political influences on bilateral trade flows have also led to a number of findings that should be taken into account in empirical analysis. States that share common interests tend to trade more with each other ~Dixon and Moon, 1993; Morrow, Siverson, and Tabares, 1998, 1999!. This is because traders expect less political risk to future trade and transaction costs are lower ~Pollins, 1989b; Morrow, Siverson, and Tabares, 1998, 1999!. Democracies tend to trade more than nondemocratic dyads because of expectations of better property rights protection and the rule of law ~Dixon and Moon, 1993; Bliss and Russett, 1998; Morrow, Siverson, and Tabares, 1998, 1999!. Security concerns affect a state s choice of trade policy because the gains from trade produce security externalities ~Pahre, 1999!. Alliance contributes to the increase in bilateral trade in a bipolar system, but not in a multipolar system ~Gowa, 1989; Gowa and Mansfield, 1993; Mansfield and Bronson, 1997!. Preferential trading arrangements, joint GATT membership, and prior colonial ties help to promote bilateral trade ~Pollins, 1989b; Mansfield and Bronson, 1997!. A number of researchers have found there might be a reciprocal relationship between political conflicts in general and bilateral trade. For example, Polachek ~1980! and Gasiorowski and Polachek ~1982! find that trading relations between states promote interstate cooperation. Oneal and Russett ~1997! and Russett, Oneal, and Davis ~1998! find that militarized interstate disputes are reduced by interstate trade interdependence. With a different measure of interdependence, Katherine Barbieri ~1996! finds evidence that trade increases conflict. Using events data for 16 dyads from 1960 to early 1990s, Reuveny and Kang ~1996! find that the relationship between bilateral conflict0cooperation and bilateral trade tends to be reciprocal. While most previous studies use single-equation models, Kim ~1998! studies the trade-conflict interaction using a simultaneous equations

11 Quan Li and David Sacko 21 model. In this paper, we follow previous single-equation-based studies and address the potential simultaneity bias by lagging the independent variables. As we discuss later, our theory has important implications for modeling conflict-trade simultaneity, suggesting some interesting future research topics. In this paper, we do not include the domestic political determinants of international trade. Voluminous literature exists that examines the influences of domestic interest groups, their coalitional behaviors and the voters in democracy ~e.g., Milner, 1988; Rogowski, 1989; Verdier, 1994!, and state political institutions ~e.g., Rogowski, 1989; Mansfield and Busch, 1995!. In this analysis, we focus on dyadic-level patterns and interactions. Clearly, though, integrating studies at the domestic and dyadic levels is a fruitful avenue for future research. Research Design: Data, Measurement, and Models Data We test the four hypotheses regarding the impact of militarized interstate disputes on bilateral trade flows in two datasets that are either broader in space or longer in time than those in previous studies. The unit of analysis is nondirectional dyad-years. 7 The first dataset ~hereafter the postwar sample! includes about 42,934 dyad-years, based on 4,340 dyads, 120 countries from 1949 to The second dataset ~hereafter the Maddison sample! has about 81,212 dyad-years, based on 1,410 dyads, 56 countries from 1870 to Dependent Variable (DTRADE ij,t ) The dependent variable is the logged total dyadic trade ~DTRADE ij,t!, i.e., the logged sum of exports and imports between two states in a dyad-year. 8 The dyadic trade data ~in millions of current U.S. dollars! are from Katherine Barbieri ~1996, 1998! for all sovereign states within the interstate system, as defined by the Correlates of War ~COW! project for the period Cases are coded as missing if they have missing data for exchange rates, only partial information about dyadic flows or national totals, or unrealistic values ~dyadic trade exceeding total trade of a country in a dyad!. Explanatory Variables The dataset of militarized interstate disputes ~MIDs! assembled by the Correlates of War project ~Gochman and Maoz, 1984; Jones, Bremer, and Singer, 1996! and later refined by Maoz ~1999! is utilized to create the dispute variables necessary to evaluate our hypotheses. A MID is an international interaction involving the threat, display, use of military force, or war. It is explicit, overt, and sanctioned 7 Several previous studies utilize directed dyad year as the unit of analysis because their dependent variable is the bilateral export ~Morrow, Siverson, and Tabares, 1998; Mansfield and Bronson, 1997!. We do not use directed dyad-year as the unit of analysis because our dependent variable is total bilateral trade. Both units of analysis ~and both dependent variables! are frequently used in the economics literature. Each has its advantage. With export being the dependent variable, one may interpret the coefficient of economic variables as affecting import elasticity directly. With total trade as the dependent variable, however, one avoids having duplicate observations for some independent variables. 8 Some dyads have zero bilateral trade values for certain years. Researchers using the gravity model typically handle the log transformation of zero trade by two different methods. See Eichengreen and Irwin ~1996! for a discussion. Some scholars, such as Mansfield and Bronson ~1997:105!, code the zero value as missing. Others, such as Bliss and Russett ~1998:1133!, replace the zero value with some arbitrary small value. In this analysis, we report results with the zero trade coded as missing, as it also follows our theoretical assumption on page 15 that bilateral trade must be non-zero for dyadic conflict to be relevant. We have also estimated the models with zero trade replaced by some arbitrary small value and the statistical results are generally robust.

12 22 The (Ir)Relevance of Militarized Interstate Disputes by the government. Data on the following dispute variables are either directly from or derived based on the dyadic MIDs dataset ~dymid 1.0! from Zeev Maoz ~1999!, who has converted the MIDs dataset to the dyadic format. MID Involvement (MID ij, t-1) MID ij, t 1 is a dummy variable that is coded 1 if two countries are involved in any MID in a dyad-year and zero otherwise. The variable is lagged one year to control for the reciprocal effect of trade on conflict. The variable construction is consistent with some previous studies ~Morrow, Siverson, and Tabares, 1998, 1999; Bliss and Russett, 1998!. It tests the effect of MID involvement on bilateral trade flows in datasets larger than those in previous studies. According to hypothesis 4, it should have a negative sign. Unexpected Effects of Dispute Onset (UNEXMID ij,t 1, MID_CONTROL ij,t 1 ; UNEXMID2 ij,t 1, MID_CONTROL2 ij,t 1 ) We test hypothesis 1 using two alternative measures, one dichotomous capturing the unexpected MID onset and the other continuous capturing the degree of unexpectedness of a MID onset. The two variables are constructed in the following manner; Appendix describes the relevant variables used in their creation. Relying on previous dyadic studies of the determinants of MIDs ~e.g., Bueno de Mesquita, 1981; Bremer, 1992; Maoz and Russett, 1993; Mansfield and Snyder, 1995; Oneal and Russett, 1997; Beck, Katz, and Tucker, 1998; Gartzke, 1998!, we build a forecasting model of the onset of MIDs, where MID_ONSET ij, t 5 f ~ JOINT DEMOCRACY, PREFERENCE SIMILARITY, MIDHISTORY, MAJOR POWER DYAD, GEOGRAPHICAL CONTIGUITY, GEOGRAPHICAL DISTANCE, RELATIVE CAPABILITY, DEMOCRATIZATION, TRADE INTERDEPENDENCE, BILATERAL TRADE FLOWS, ECONOMIC GROWTH, ALLIANCE, ECONOMIC DEVELOPMENT, TEMPORAL DEPENDENCE! ij, t21. Simulating how traders assess the political risk of military violence, we use logit analysis to acquire the predicted probability of the onset of MIDs in two steps. 9 The above fully specified model is estimated separately for the two datasets and the predicted probability of MID onset in each dyad-year is obtained. Next, since several economic variables ~economic growth, trade interdependence, bilateral trade flows, economic development! have substantial missing values, a model excluding these variables is estimated and the predicted probability of MID onset is obtained to replace only the missing values from the first set of MID onset probabilities. Because traders could only use available information to predict conflict, all independent variables are lagged one year. Two new variables are then created as follows. Unexpected MID Onset. If the predicted probability of a MID onset is smaller than 0.20, the onset is defined as unexpected. 10 UNEXMID ij,t 1 is coded 1 when a dyad experiences an unexpected MID onset in a year and zero other- 9 Estimation uses LOGIT procedure in STATA 7.0. Two issues are worth discussion. First, the model only approximates and is not equivalent to the models used by traders. Traders apparently incorporate more contextual information in their assessments, whereas our interest is the underlying causal patterns. We thank one referee for pointing this out. Second, an alternative approach is to use the neural network model ~Beck, King, and Zeng, 2000!. Future research may assess whether our results are sensitive to the use of different forecasting models. Albeit imperfect, the logit, like any other model, generates an approximation of the predicted probability based on available information. 10 Even though the predicted probability cutoff 0.20 is arbitrary, it does what we intend it to do in principle. We also examine the effects by changing the probability cutoff to 0.10 for the group of least expected MIDs and the statistical results remain the same.

13 Quan Li and David Sacko 23 wise. 11 The variable is one year lagged behind the dependent variable in order to avoid potential simultaneity bias and, more important, to measure its effect on future trade. The variable construction represents quite appropriately those disputes belonging to cell ~4! in Table 1, from which large suppressive ex post effects are expected. According to hypothesis 1, UNEXMID ij,t 1 is expected to have a negative sign. Since UNEXMID ij,t 1 captures only surprising MID onsets, the anticipated MID onset and the continuation of any MID are not reflected in this variable and should be controlled for in order to avoid the omitted variable bias. For this purpose, MID_CONTROL ij,t 1 is constructed. It is coded 1 if a dyad experiences an anticipated MID onset or continues to have an ongoing MID in a year, and zero otherwise. Recall our earlier discussion; MID_CONTROL ij, t 1 is likely to have a small or zero negative effect because of the preemptive risk diversification by traders. Unexpectedness of MID Onset. As the dichotomous measure of unexpected MID onset throws away a lot of useful information, an alternative is to measure the unexpectedness of MID onset as a continuous variable. UNEXMID2 ij,t 1 MID onset dummy ~1- predicted probability of MID onset!#. It is lagged one year behind the dependent variable. The variable corresponds to both cells ~1! and ~4! in Table 1, measuring the degree of surprise of a MID onset. According to hypothesis 4, UNEXMID2 ij,t 1 is expected to have a negative sign. MID_CONTROL2 ij, t 1 is a dummy variable coded 1 if a MID continues in a year and zero otherwise. The variable is expected to have a small or zero negative effect. Although they both test hypothesis 1, the unexpected MID onset ~UNEXMID ij, t 1! and the unexpectedness of MID onset ~UNEXMID2 ij,t 1! are somewhat different conceptually. One corresponds to cell ~4! only in Table 1 while the other corresponds to both cells ~1! and ~4!. UNEXMID ij, t 1 captures only unexpected MID onsets above some threshold level while UNEXMID2 ij,t 1 captures the degree of surprise a MID onset brings about. The associated MID control variables are constructed accordingly. Dispute Severity (HOSTILITY ij,t 1 ) and Duration (DURATION ij,t 1 ) Rigorous tests of hypotheses 2 and 3 require that we capture either the unexpected dispute duration and severity or the unexpectedness of dispute duration and severity, as in the case of dispute onset. Owing to space limitation and measurement complexity, we perform two preliminary tests of hypotheses 2 and 3 using the actual MID hostility and duration instead. The tests, though crude, are justifiable because as Tables 2 and 3 show, the ex post effects of the long dispute weakly dominate those of the short one and the ex post effects of the severe dispute weakly dominate those of the mild one, regardless of the degree of uncertainty over duration and severity. We leave more discriminating tests for future research. HOSTILITY ij,t 1 is an ordinal-level measure of the severity of a dyadic MID in a year. It is the highest level of hostility in a dyadic dispute, ranging from 1 to 5 representing no militarized action ~1!, threat to use force ~2!, display of force ~3!, 11 From 1870 to 1992 the number of dyadic MID involvements is 3,424 with 2,260 being observations of MID onsets. Because of missing values in various variables, estimation for the Maddison sample covers 2,118 observations of dyadic MID involvements with 1,420 MID onsets, and estimation for the postwar sample covers 1,934 dyadic MID involvements with 1,340 MID onsets. The number of observations for the unexpected MID onsets ~when UNEX- MID ij,t 1 1! is 1,102 for the Maddison sample, accounting for about 77% of MID onsets and about 52% of MID involvements in the dataset. The number is 1,057 for the postwar sample, accounting for about 79% of MID onsets and about 55% of MID involvements in the dataset.