WPS4094 Are Lives a Substitute for Livelihoods? Terrorism, Security, and U.S. Bilateral Imports* Daniel Mirza (CREM, University of Rennes 1, France) Thierry Verdier (PSE, France and CEPR, London) Abstract What is the impact of terrorism on trade through higher security at the borders? We set up a theory which shows that the impact goes not only from terrorism to trade; higher trade with a partner might, in turn, increase the probability of terrorism acts and make security measures more costly for total welfare. In order to identify the true impact of terrorism, our theory allows then for a strategy to condition out the latter mechanism. We show in particular how past incidents perpetrated in third countries (anywhere in the world except the origin or targeted country) constitute good exogenous factors for current security measures at the borders. Our tests suggest that terrorist incidents have a small effect on US imports on average, but a much higher effect for those origin countries at the top of the distribution of incidents. In addition, the level of the impact is up to three times higher when the acts result in a relatively high number of victims, the products are sensitive to shipping time, and the size of the partner is small. The paper further shows how terrorism affects the number of business visas delivered by the Unites States, thereby impacting significantly U.S. imports in differentiated products. These results suggest that security to prevent terrorism does matter for trade. Keywords: Terrorism, trade, security JEL codes: F12, F13 World Bank Policy Research Working Paper 4094, December 2006 The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. *We are especially grateful to James Anderson, Matthieu Crozet, Juan Carlos Hallak, David Hummels, Thierry Mayer, John Romalis, Maurice Schiff, Ben Shepherd, Barbara Spencer, and Vanessa Strauss-Kahn for very fruitful discussions on earlier drafts of the paper as well as conference participants at the NBER Summer Institute, ETSG, PSE, CEPII and universities of Paris 1, Rennes 1 and Paris 9 seminars. The authors wish also to thank Anne-Sophie Novel and Fabien Verger for excellent research assistance and Francoise le Gallo for compiling OECD-Flubil trade data. This work is part of a broader project on trade facilitation and development supported through a trust fund of the U.K. Department of International Development. It has further benefited from the funding of the World Bank and a grant from the French Ministry of Education and Research (contract ZVA46). Thierry Verdier and Daniel Mirza acknowledge respectively financial and logistical support from the University of Southampton (UK) and the University of Paris 9 and CEPII (France) where part of this work was completed. 1 Introduction In his 2003 Remarks at the Heritage Foundation, Robert C. Bonner, Commissioner of the Bureau of Customs and Border Protection at the Department of Homeland Security in the US, stated: "We must protect American lives, but we must also protect American livelihoods­our economy. That's why we have twin goals: (1) increasing security and (2) facilitating legitimate trade and travel." After the events of September 11, the US decided to strengthen the security at its borders against transnational terrorism. In April 2004, it signed with the EU a customs cooperation agreement to extend the Container Security Initiative throughout the EU. In this agreement, US customs officers could operate in some ports of the European Union to screen and control all cargos to the US that depart from or transit through the European countries (Archick (2005)). To date, several countries have already implemented these measures and other important ports are expected to comply, in particular after the recent London attacks of July 7, 2005. Although more controversial, the two partner authorities also reached other agreements in air transport by which they increase identity controls over the borders. Hence, airlines have to provide the US authorities with the identities of the passengers from or via the EU before departure while the latter have now to prove their identity via biometric identifiers on their passports. Are security measures against terrorism affecting international trade flows, and by how much? Are American livelihoods really affected by securing their borders from terrorism? Which kind of goods, sectors or trading partners are most affected by these measures? The purpose of this paper is to provide a first step towards responding to these questions. First, it sets up a simple theoretical framework linking trade, security and the probability of terrorism acts. This theory recognizes explicitly the strategic nature of interactions between terrorist organizations and governments of targeted countries. Second, we take our theory to the test to investigate how terrorism incidents are affecting bilateral trade and how in particular, this effect is translating through counter-terrorism security measures at the borders. More precisely, the theory that is developed emphasizes two channels, of different nature, linking trade to security. First, there is the "traditional view" that an increase in security measures could affect transaction costs and thus trade. However, our model also captures the fact that in return, a country that is a big importer from a given economy for any given reason (proximity, big size of exporter, differences in specialization, etc...) tends to reduce its security at its borders towards the latter. The argument is that the related total cost of security can end up being higher than the associated gain in the probability of preventing terrorism attacks. These two forces suggest two important implications: 2 First, trade and terrorism incidents are endogenous to each other. On the one hand, the relation- ship is negative: terrorism via an increase in security reduces trade. But on the other hand, it can be positive: higher trade volumes are more likely to limit security measures which in turn increases ter- rorism activities. We propose therefore a way to condition out the latter effect which would otherwise bias downward (in absolute value) the estimated negative impact of terrorism incidents on bilateral trade. What we do is to consider a particular type of incidents that we believe are the most exogenous to security and bilateral trade. These are past incidents perpetrated by groups originating from a country against the interests of another, but which take place in a third country. As an example, take for instance the case of Al-Qaeda in 1998, whose origin at that time was still attributed to Saudi Arabia1. In that year, Al-Qaeda managed to explode a car bomb next to the U.S. embassy in Dar El-Salam (Tanzania), which hurt nearly 80 people. We make the point here, that with such types of incidents, based in third countries (here, Tanzania), we are capable to identify the exogenous impact of terrorism originating from one country (here, Saudi Arabia) on its bilateral exports towards the targeted country (here, the US). Second, if transnational terrorism induces security reactions directed towards countries from which the incidents emanate, then trade should be affected differently across countries and products. In- deed, one may expect different bilateral trade effects with origin countries of terrorism (from where transnational terrorism comes) than with other countries. As well, some products and sectors may be more sensitive to bilateral security measures (like social network goods or differentiated products a la Rauch) than others for which transaction costs are less discriminatory across countries (standard goods and primary products). An additional route to identify the counter security effect however, is to link terrorism acts to trade through a security measure that we can observe in the data. We investigate empirically all these issues on US bilateral imports. We have chosen this country because it has been the main "target" of transnational terrorism (i.e. one-third of the incidents over the considered period target the U.S) and the one with the largest variation of "origin" partners (more than 95 countries have hit the U.S between 1968 and 2003). We combine, thereby, three datasets on trade, terrorism incidents and business number of visas issued by the U.S. First for the trade data, we use bilateral imports of the United States at the product level (SITC4/5 digits) from the NBER World Trade Data complied by Feenstra and Lipsey (2005). As it provides only values which exceed 100,000$ per year, we have completed it with the OECD-FLUBIL bilateral trade dataset. Disaggregated data are needed here, in particular, in order to be able to capture the differentiated impact of counter-terrorism measures on trade across products, as mentioned above. Besides, as it will be shown next, as both bilateral trade and terrorism activities seem to be correlated with the relative specialization of countries, one needs to condition out this effect in the regressions, which could not have been done by using highly aggregated data. 1It might be argued that it is not the case anymore in the very recent years (say after 2000), where Al-Qaeda is becoming more fragmented, acting as a Multinational with affiliates over the world. 3 Second, we use the ITERATE dataset set-up by Mickolus, Sandler, Murdock and Flemming (2003) which reports transnational terrorist activities. More precisely, ITERATE is an event-based dataset that provides information on the date, country of localization of the attack, its origin and the targeted country. It lists all of the incidents in the world that have been reported in the medias since 1968 onwards. We are mainly interested in those attacks where the US has been the main target, via its representative authorities, its army or its civilians anywhere in the world. Third, in order to prove that at least a part of the impact of terrorism is indeed translating through counter-security measures, we use a dataset from the Department of State reporting the number of business visas issued by the United States to each partner country from 1997 to 2002. We are in interested here in linking the attacks to visas issuance across countries and thereby, to the induced impact on trade in network related products. Our empirical results are then the following. First, past terrorist events against the US, perpetrated by groups from a given country, affect negatively American imports from the latter. The effect is statistically significant but relatively small on average. Indeed a 1% increase in past terrorism activities from a country reduces US bilateral imports by around 0.01%. This negative effect is nonlinear however. The elasticity is higher the riskier is the country of origin in terms of its related frequency of incidents and the number of victims. In particular, a 1% increase of past incidents from countries such as Colombia over the period (or Pakistan and Saudi Arabia in recent years) results in around 0.5 to 1% decrease in their exports to the US. Second, and consistent with our theory, the impact of past terrorism on current US imports is higher when the partner country is small in terms of its GDP size. Besides, the level of the impact more than doubles (and hence reaches more than 1 to 2% in the case of Colombia, or more recently Pakistan) when the acts result in a relatively high number of victims and for products that are sensitive to the time-length of shipping and network-lengths. Third, our paper shows how terrorism incidents affect the number of business visas delivered by the US, thereby impacting significantly bilateral US imports, specifically in differentiated products. This last result therefore suggests a clear channel through which counter terrorist security reactions may affect differentially bilateral trade flows across US trading partners and across products. There is a significant literature on the economic consequences of terrorism attacks, although not focusing on the impact of counter-security measures. In a nice survey, Frey, Luechinger and Stutzer (2004) mention some studies that look at the impact of terrorism on different channels of globalization (tourism, air transport and foreign direct investment). Abadie and Gardeazabal (2005) look also at the impact on the world economy through foreign direct investment. They argue that the risk of terrorism in a country reduces the expected return to investment while increasing its variance. They show that a one standard deviation increase in the risk of terrorism in a country reduces its net 4 FDI position by 5%. Our work however is more closely related to papers investigating the links between international trade in goods and transnational terrorism2. In the aftermath of September 11, the OECD was particularly concerned by the extent to which the world economy would be hit by the observed increase in security surcharges emanating from airlines, maritime transport companies or insurers due to the increase in terrorism threat (OECD (2002a), OECD (2002b)), although without giving a particular estimate of the impact on trade. Walkenhorst and Dihel (2002) use a CGE modelling to assess more analytically that impact on trade and welfare. The authors model the costs from a terrorist attack in the same way as an increase in tariffs with the only exception that the former is not accompanied by an additional revenue for the importing government. Where the transaction costs born from terrorism are uniform across regions, the results show that highly opened regions and industries with high import price-elasticities would bear a non negligible adjustment in trade and welfare losses. Another study by Nitsh and Schumacher (2004) uses a gravity model to assess the impact on trade between two countries which have experienced terrorism attacks. They find that a doubling of terrorism attacks in those countries affect their trade by around 4%. Fratianni and Kang (2006) extends the analysis of Nitsh and Schumacher (2004) to a different time period (1980-1999) and investigate how the terrorist impact on bilateral flows interacts with geographic distance. Using the ITERATE database for the period 1968-1999, Blomberg and Hess (2006) estimate the cost of violence on bilateral trade flows, considering as well other sources of violence like external conflicts, revolutions, and inter-ethnic fighting. They find that a country which has a terrorist accident is associated with a 7.6 % decline in its bilateral trade. While significant, this is less than half the magnitude of the negative impact on trade from external conflicts and inter-ethnic conflicts. All these papers however, do not deal with the impact on bilateral trade of a targeted country whose main interests and citizens have been hit in a foreign country. As it will be shown below, a significant proportion of the incidents targeting rich countries is actually perpetrated either locally (in the country of origin) or in a third location. Besides, although all these studies emphasize the transaction cost impact of terrorism on trade, most of them do not address the possible endogeneity between trade, terrorism and security measures neither in theory nor in the data. As well, though not directly related to terrorism, Anderson and Marcouiller ( (1997), (2002)) focus on the impact of insecurity on trade. In the first paper, insecurity arises endogenously from the choice of agents to allocate their labor between production and predatory activities, the latter hindering international trade at the borders. In the second paper, the authors model alternatively insecurity as a hidden tax on trade. They find that poor institutions in terms of government transparency and commercial legal systems hinder trade at least as much as tariffs. Instead, our paper model together, the probability of terrorism occurrence (i.e. insecurity), the governments' choice of (counter)-security measures and trade. Less trade in such a framework does not directly come from insecurity but from counter-terrorism security measures at the borders. 2In another paper, Mirza and Verdier (2006) provide a recent survey on these issues. 5 The paper is structured in the following way. In section 2, we present the ITERATE dataset and describe some stylized features that will be of interest to investigate the links between transnational terrorism and bilateral trade flows. Section 3 sets then a simple theoretical model of endogenous transnational terrorism and security, embedded into a standard trade model. Section 4 explains the induced empirical strategy to test the impact of terrorism and counter-terrorism measures. Section 5 takes the model to the test and presents the econometric results. Section 6 provides further evidence on the impact of terrorism translating through higher security at US borders. In particular, it investigates one specific (observable) channel of security measures at the border: the allocation of business Visas by US authorities. Finally, section 7 concludes. 2 Transnational Terrorism and the ITERATE Database ITERATE defines terrorism acts as "the use, or threat of use, of anxiety-inducing, extra-normal violence for political purposes by any individual or group, whether acting for or in opposition to es- tablished governmental authority, when such action is intended to influence the attitudes and behavior of a target group wider than the immediate victims and when, through the nationality or foreign ties of its perpetrators, its location, the nature of its institutional or human victims, or the mechanics of its resolution, its ramifications transcend national boundaries". We amend that definition by two additional conditions to qualify an incident as "transnational terrorism". Focusing first on the term terrorism, we follow Omar Malik (2000) from the Royal Institute of International Affairs who claims that only those incidents that are perpetrated against or within liberal states should be qualified as terrorist attacks. A country is said to be liberal when it safeguards human rights in its laws and practices. Qualifying terrorism acts as incidents against non-liberal countries is usually more controversial. To some observers, these actions might be viewed as terrorism but to others, they might rather be qualified as acts of resistance against a totalitarian country. To avoid getting into the controversy, we decided to withdraw the corresponding observations from the dataset. We had access to the Freedom House dataset that rates civil liberties and political rights on a scale varying between 1 and 7 for each country from the 1970s onwards, in order to distinguish between 'liberal' and 'non liberal' countries. As in Helliwell (1994) and Rodrik (1999), we combine the two ratings into an index varying between 0 and 1. The higher it is in a given year and the more 'liberal' the observed country shall be considered. For the purpose of this paper, we retained only those observations where incidents took place within or against a country associated with an index equal or higher than 0.5. Second, Mickolus et al treat some incidents perpetrated by separatist groups like ETA in the basque country, IRA in Northern Ireland or FLNC in Corsica as transnationals, leaving the choice for the users of the dataset to decide whether or not to include them in the data. We define instead a terrorism incident as "transnational" when it is directed by a group that emanates from an internationally recognized nation against or within an internationally recognized other nation and thus withdraw 6 above observations from our study.3. For instance, when the ETA group from Spain perpetrates an incident in Spain, it shall not be considered as 'transnational' and thus shall be withdrawn from the data at hand. However, when the same ETA group attacks a Spanish authority, one of its representations or Spanish civilians within another country, say France, then the observation is kept in the dataset. That is because that type of act has some implications for security measures on the Franco-Spanish borders. At the end, from nearly 12,500 observations in the ITERATE dataset from 1968 to 2003, we end up with around 10,700. We first look at the origin of the incidents and their place of location. Before going into details, one has to be aware that the country of origin might or might not be the country of location of the incidents: we identify each origin by the country of first nationality of the terrorist group while the country of location is the country where the act has been observed in the ITERATE dataset. In order to save space, table 1 ranks the first 60 countries of origin by their number of incidents over the period, although one should be aware that most if not all of the countries in the world have been at the origin of at least one terrorist incident from 1968. The table indicates that these countries have been related to at least 20 incidents each during the period. Besides, it is worth mentioning that one third of total incidents have been perpetrated by unknown groups, to which no origin have been associated.4 The country of origin as it is defined here might not be that of the operations of the group. In general, when the group does not operate in his own country it might be operating in the country of location of the incidents (hereafter, host country)5. Therefore, it is interesting to see what is the proportion of incidents originating from one country but that takes place in another. ITERATE is a place-based dataset. We know exactly where and when each incident has started and ended. In more than 95%, the location of start is the same than that of its end which makes it relatively straightforward to locate the incidents. 6 Figure 1 sketches the distribution of the incidents extracted from the ITERATE database across 3 3It is worthwhile mentioning that we have kept incidents emanating from Palestine as the latter is already recognized as a state by 94 nations around the world. Further, 11 more nations, generally from the OECD, grant Palestine some specific form of diplomatic status. 4As it has been already documented in Sandler and Enders (2004), the number of incidents has decreased dramatically after the nineties compared to the first decade. Although experienced by most of the origin countries, this drop had not been uniform. For instance, although groups from Palestine and Colombia had been very active during the whole period, Lebanese and Iranien group activities had been extremely high only during the eightees and the nineties. In recent years, it has even risen dramatically in some countries like Pakistan, Afghanistan and Saudi Arabia. 5The case of Al-Qaeda is an exception where the country of nationality of the group (presumably Saudi-Arabia) is different from its presumed country of residence ('Headquarters' in Afghanistan or Pakistan) and further different from many countries where its 'affiliates' operate. See Clarke, 2004. That said however, following our definition of country of origin, we have classified Al-Qaeda operations as originating from Saudi-Arabia as in Krueger and Laitin (2003). Now, because of its affiliates, Al-Qaeda could have many countries of origin and this could be problematic to our study that relates the economic impact of terrorism to the pre-identified country of origin of the groups. Now, one can still assume that the authorities threatened by Al-Qaeda consider the islamic world as one country of origin as a whole, against which they must secure the borders. In that case, our study can still predict of how much Al-Qaeda incidents are affecting trade between the targeted country and the muslim countries taken as a whole. 6Where the start location is different from its end however (i.e. Aerial Hijacking), we code the host country as the country where the incident has started. 7 possible locations (Origin, Target country and Third country). The country is coded as target when it is that of the main nationality of the victims. Nearly 80% of the victims are associated with only one nationality over the whole period, which is why one could assign in a relatively confident way only one target country to an incident. It is important to note here that victims, in ITERATE, are defined as "those who are directly affected by the terrorist incident by the loss of property, lives, or liberty". Thus, when a French embassy is hit without casualties in say, an African country, France is then coded as the target country. Besides, the third country represents the country where the action begins albeit different from the origin and target states. From figure 1, we can see that only a small and relatively stable proportion over time (10 to 20%) takes place in the targeted countries. Attacks like those of New York (2001), Madrid (2003) and more recently London (2005) are not representative of most of the incidents. In the earlier period, around 30 to 50% of the incidents took place in third countries but that share declined steadily over the period to reach around 20% of the incidents. This reduction seems to be concomitant with the rise of the share of incidents taking place in origin countries (i.e. where they have been planned and prepared). Hence, at the end of the period, 60 to 80% of the incidents became local. These findings are quite similar to those of Krueger and Laitin (2003) who use the Department of State dataset to assert that, in recent years, perpetrators preferred setting-up actions against "targets from foreign countries [that are] close to home". The reasons are beyond the scope of this paper. However, even if the third country location is decreasing, it is still highly variable and thus should still matter as much as incidents in origin and target countries for detecting the impact on counter-security measures and trade between them. Table 2 ranks the main 50 targeted countries over the period. The US is by far the country that is most hit by terrorism attacks over the period, before France, Israel and Great Britain. Besides, the distribution of incidents across targeted countries does not change much over time. A simple calculation of the coefficient of correlation between the distribution at the beginning (1968-1978) and that at the end of the period (1997-2003) is around 0.96. It is quite simple to guess, however, that some countries like Israel are systematically targeted by a small number of groups related to one particular state (here Palestine)7. Can we say the same for the other most targeted countries? Table 3 presents the top 65 ranking of 'bilateral' incidents (i.e. ranking by origin and target countries) wherever those incidents take place. One can easily see that over one third of the bilateral incidents involve the US as a target country: that is, the distribution of incidents against the US is spread over a big sample of source countries. This is obviously not the case for Israel, France or Great Britain which are associated with at most 3 countries in the top 65. However, because of the bigger variability of incidents against the US, this makes cross-country studies related to the US as a target country easily implementable. In relation to the link between transnational terrorism and bilateral trade flows, two important 7One should be aware here that only incidents perpetrated by Palestine against Israel, but in a third country, or implying victims from a third country, are reported in the ITERATE dataset. Most of the incidents between these two countries are not reported by ITERATE however, because they are considered to be domestic, not transnational, terrorism. 8 remarks are worth making at this stage. First, over the period, and in particular before the 1990s, the terrorist groups tend to hit targets that were relatively close to home and/or had big influence on internal policies of origin countries: that is in particular the case of some Latin American countries (Colombia, Puerto Rico, Peru, Cuba, Argentina) vis-`a-vis the US but also that of Algeria and Spain vis-`a-vis France. As proximity and colony (or neo-colony) ties are also known to be factors of trade this could give a rapid idea on why one could find some positive relationship between terrorism activities and bilateral trade if those factors are not correctly accounted for. In recent years however, the groups that were the most active and that have concentrated their attacks on the US in particular, emanated from Pakistan (100 times more between beginning and end of period), Saudi Arabia (50 times more) and Colombia (30 times more). These extremely high figures have to be attenuated though for Saudi Arabia and Pakistan by the fact that the activities of their groups were quasi-null in the beginning of the period (only one attack each in the 1968-1978 period). Thus, only terrorism groups from Colombia seem to have maintained a high intensity of their activities against the US in Latin America while a new set of groups from countries located relatively far from the US have now significantly intensified theirs. As figure 1 has already suggested, note however that these groups have been mostly operating at home. Second, it is also interesting to see that most of the economies at the origin of the bilateral incidents are developing countries that are mainly specialized in agriculture, natural resources and manufacturing employing intensively those resources. Whereas countries like Saudi Arabia, Iran or even Colombia are specialized in oil production and oil related products like plastic (especially Saudi Arabia), Latin American countries in general (including Colombia) exploit intensively some natural resources from agriculture and fishing (Argentina, Cuba, Colombia, Chile, Puerto Rico) to mineral resources (Peru) and mining (Chile). As differences in specialization between developing and developed countries represent another important factor of trade, this is then another reason why one could retrieve a positive relationship between terrorism and bilateral trade if the degree of specialization of countries is not accounted for. 3 A Simple Model of Trade, Terrorism and Security In this section we describe the basic elements of a simple model of trade, terrorism and security. There is one country (the US) labelled 0 and N other countries with whom country 0 is trading. 3.1 Trade Each country produces differentiated goods under increasing returns. The utility of a representative agent in country 0 has a standard Dixit Stiglitz form: j 1 /(1-1/) =N -1/) U0 = njx(1 0j j=0 9 where nj is the number of varieties produced in country j, x0 is country 0 demand for a variety j of country j (all goods produced in j are demanded in the same quantity by symmetry) and > 1 is the elasticity of substitution. In country 0, this helps define an usual consumer price index: j 1 /(1-) =N P0 = 1- 1- njpj T0 j j=0 where pj is the mill price of products made in j and T0 are the usual iceberg trade costs between j country 0 and country j. If one unit of good is exported from country j to country 0 only 1/T0 j units are consumed. Trade costs depend on geographical distance, trade restrictions and will also be assumed to depend on security measures (more on this below). As is well known the value of demand by country 0 from country j is given by pjT0 1- j m0 = njE0 j (1) P0 where E0 is total expenditure of country 0. In each country, the different varieties are produced under monopolistic competition and the entry cost to produce in a monopolistic sector is supposed to be 1 unit of a freely tradable good which is chosen as world numeraire. This good is produced in perfect competition. This in turn fixes the wage rate in country 0 to its labor productivity a which is assumed to be the same across countries and across sectors under perfect and imperfect competition (for simplicity). Given this, standard mark -up conditions from profit maximization by firms give that mill prices in the monopolistic competitive sector are identical and equal to the mark up /(-1) times marginal costs (also equal to 1). As labor is the only factor of production, and agents are each endowed with one unit of labor, total expenditure in country 0 is given by E0 = aL0 where L0 is the number of workers in country 0. On the supply side, free entry implies that nj = aLj/(). In equilibrium, the indirect utility of the representative consumer in country 0 is j 1 /(-1) =N a U0 = U0(T0) = 1- 1 (aLj)T0j () -1 -1 j=0 with T0 the vector {T0 }j j=0,...Nof iceberg costs between country 0 and the rest of the world. As is well known from this simple model, one gets bilateral imports of country 0 from country j as proportional to : m0 = a.LjE0T0 1- -1 j (2) j P0 3.2 Terrorism and Security We assume that there are K N terrorist organizations, each of them being associated to one particular country or having headquarters located in one country. The objective of each of these 10 organizations is to get visibility (which help them capture or enjoy particular political or economic rents) In order to do this, each organization is going to spend resources to commit a terrorist event on country 0. More precisely, we assume that a typical terrorist organization from country j maximizes MaxRj (Rj,Sj)Vj - Rj (3) where (Rj,Sj) is the probability of success of a terrorist act in country 0. It depends positively on the amount of resources Rj invested by the terrorist organization and negatively on the security measures Sj implemented by the government of country 0 against country j is marginal resource cost of the terrorist organization and Vj is the perceived visibility gain enjoyed by the terrorist organization when terrorism is successful. We assume a specific parametric form for the probability of success (Rj,Sj). More precisely, as in Anderson and Marcouiller (1999) we take a simple asymmetric contest success function: Rj (Rj,Sj) = Rj + Sj with the technological parameter > 0 reflecting the relative efficiency of security measures to reduce the occurrence of terrorism. The solution of (3) gives immediately: the reaction curve of terrorist group j SjVj Vj Rj = R(Sj,) = - Sj for Sj = 0 otherwise The government of country 0 is concerned both by the economic welfare of the representative consumer U0(T0) and about the level of security 0 of his citizens against terrorism. To fix ideas, consider that he maximizes W0 = LogU0(T0) + µLog0 where the level of security 0 is a positive function of the probability of non occurence of terrorist acts in country 0: j=K 0 = 0(R,S) = [1 - (Rj,Sj)] j=1 with R = {Rj}j =1,..Kand S = {Sj}j =1,..K are respectively the vector of resources spent by terror- ists organizations and security measures taken by the government of country 0. Security measures Sj against terrorists residing in country j are likely to increase transactions costs on trade flows (security checks, time delays, restrictions on passports of business people, various immigration controls) and we simply pose that T0 = Tj(Sj) with Tj(.) > 0 j We assume that the government of country 0 forms some beliefs on the level of resources under- 11 taken by terrorists from country j to commit a terrorist act in country 0 and given these beliefs (more on this in the appendix), his problem is simply Max{Sj} LogU0(T0) + µ ER Log0(R,S) where ER(.) reflects the expectation operator of government of country 0 on the vector of terrorist resources R. Neglecting constant terms, this problem can be rewritten as: j =N j=K 1 Max{Sj} Log 1- LjT0 + µ ER Log[1 - (Rj,Sj)] - 1 j j=0 j=1 or j =N j=K 1 Sj Max{Sj} Log - Log Lj[T0 (Sj)]1 j + µ ER - 1 Rj + Sj j=0 j=1 with the obvious notation that for a country j which has no terrorist organization residing there Sj = 0and T0 = T0 (0) j j It is easy to see that the first order conditions of this problem can be written as: T0 j 1 1 d m0 j = µ - [ERj(Log(Rj + Sj)] (4) Sj T0 j Sj dSj with LjT01- j m0 = j (5) h=N LhT01- h h=0 The left hand side is simply the marginal distortional cost of imposing security controls and measures. It affects trade flows and, for a given country j is proportional to the level of imports m0 of country j 0 from country j. The right hand side is the marginal gain of security measures on the probability that there is no occurrence of a successful terrorist act in country 0. It is going to depend on the structure of beliefs that the government of country 0 has on the amount of terrorist resources R spent by terrorist organizations against country 0. To fix ideas, we take for each terrorist organization j, that the resource cost can take two values L and H with L < H. Denote then j and j = 1 - j respectively the beliefs government of L H L country 0 has on terrorist organization j having a resource cost j = L and j = H. Then (4) can be rewritten as: L H T0 j 1 Rj Rj m0 L L j = µj + (1 - j ) (6) Sj T0 j Sj Rj + Sj L Sj Rj + Sj H 12 with8 Rj = R(Sj,L) = L SjVj - Sj and Rj = R(Sj,H) = H SjVj - Sj (7) L H The solution of (6), (5) and (7) defines then a Bayesian Nash equilibrium vector in terrorism and security {S,RL ,RH } = {S(L),RL (L),RH (L)} which depends on the vector of beliefs L = {j }j L =1,..K that government 0 has on terrorist organizations. In theory, once such an equilibrium is computed, one may have the values of trade flows of country 0 with the rest of the world. To be a bit more precise, let us consider the case where transactions costs between countries 0 and j take an exponential form: T0 (S) = Tj eSj with > 0 j and that there is a unique terrorist group in one country j. Then (6) and (7) are rewritten as: m0 j 1 1 = - E( ) (8) µ Sj Vj Sj with E( ) = j L L + (1 - j ) L H ). In the appendix we solve for the case with K terrorist 9 organizations and give sufficient conditions for the existence of a unique Bayesian Nash Equilibrium of the terrorist-security game. The case with only one terrorist group located in a particular country j can be easily illustrated graphically with the structure of the equilibrium represented in figure 2. The first quadrant plots the two relationships (8) and (2). Curve (SS) represents equation (8) and is downward sloping. It shows how the level of security measures undertaken by country 0 is reduced when the level of trade flows between country 0 and country j m0 gets larger. Conversely, curve (TT) represents equation (2) and j depicts the fact that the actual level of trade flows depends negatively on security measures. These two relationships therefore describe a two-way interaction between trade flows and security measures. Assuming, as shown in the picture that a stable equilibrium exists, it is described by point E at the intersection of (SS) and (TT). One may as well compute the average probability of non occurrence of a terrorist act: R(Sj,L) R(Sj,H) E(0) = 1 - j L L = E( ) Sj (9) [R(Sj,L) + Sj] + (1 - j )[ R(Sj,H) + Sj] Vj The second quadrant plots the curve (PR) describing how the average probability E(0) of no 8The derivation of (6) comes from T0 j 1 1 d m0j = µ - [ERj(Log(Rj + Sj)] Sj T0 j Sj dSj with ERj(Log(Rj + Sj)] = j Log(Rj + Sj) + (1 - j )Log(Rj + Sj) L L L H 9We assume a configuration of parameters such that Sj < 4Vj/(E( ) to ensure that the SOC are satisfied. 13 terrorism in country 0 varies with the level of security implemented in the country (equation (9)). The equilibrium average probability of no success of terrorism is then provided by point P in figure 2. Several simple comparative statics can be undertaken inthis setting. It is easy to show that a decrease in the expected cost of some terrorist actions E( ) or an increase in the efficiency of authorities 0 have a positive effect on security measures undertaken at the borders (see figure 3). Interestingly, the (TT) curve remains unaffected which ends-up reducing equilibrium trade flows. Besides, equation (8) shows that in return the probability of non occurrence of incidents decreases. On the opposite, an increase say, in Lj total employment, or a decrease in some trade costs T other than security costs, like transport costs, both tend to increase imports m0 . This however, shifts both j (TT) upward and (SS) downward. The effect is a reduction of security measures Sj and a reduction of the probability of non occurrence of incidents E(0) (i.e. increase in the probability of provoking an incident). 4 Estimation Strategy What are the empirical implications of such a model? Clearly, equations (8) and (2) suggest some endogeneity between bilateral trade flows, security and bilateral terrorism. Second, in order to capture only the relationship going from security to trade, exogenous factors that affect only the security curve (SS) are needed, holding constant all variables that affect both curves (i.e. distance, common colony, GDPs, etc...). Equation 8 is a second degree polynomial equation. Solving for security (Sj), one can show that it directly depends on the interaction between expected marginal costs of the terrorist organization and the effectiveness of security measures, that is E( ).0. It is interesting to see then that these measures are affecting the security curve without impacting the trade curve, which makes them very good candidates to identify our effect. Now, we do not observe the degree of efficiency of security measures, neither do we observe the marginal costs of terrorist actions. To this end, we proxy the former, 0, by the frequency of incidents against the US observed in the past: All things held equal, the higher is the number of incidents against the US compared to the total number of world incidents in the last years, the lower is its efficiency to implement security measures that safeguards its citizens and interests over theworld. We also proxy the beliefs of the authorities about the efficiency of terrorist organizations, E( )j by the world share of incidents that originate from country j in the last few years. To be more precise, let n express the total number of incidents, nj those originating from any country j, nUS those that hit the US in whichever location in the world. Assuming T is the time horizon of the authorities and nT nUS is the total number of incidents over that horizon, a proxy of 0 would be: Ft US = t [t...t-T]t . nT nj,t Besides, the proxy of E( )j would be: Fj,t = t [t...t-T] . nT Thus, in the empirical study, these would constitute our first 2 variables of interest. Alternatively, and following the theory, a third variable of interest can be approached by the interaction of these two variables: 14 nj,t nUS US t [t...t-T] t [t...t-T] t jt = . = Fjt.Ft US nT nT This third variable is an indicator of exogenous security against the occurrence of terrorism inci- dents. All three variables are based on past incidents computed from the ITERATE dataset from 1968 until 2002. Past incident-frequencies are defined over 5 years (i.e. the time horizon over which authorities formulate their beliefs is assumed to be 5 years (T = 5))10. We thus basically ask what is the effect of the past 5 years of incidents, on US imports. Further, ITERATE delivers information on the country of location of each incident. This enables us to split terrorism incidents njt between those perpetrated in the country of origin (Origjt hereafter), those located in the targeted country (i.e. in our case, the US) and those located in third countries (Thirdjt). In particular, we expect observations on past incidents in third countries to be the most exogenous to US security at the borders. The reason is that terrorism in third countries should be in return much less affected by trade between the US and the origin country. In contrast, terrorism located in either the US or the origin country could be related directly or indirectly to trade between them. For instance, higher flows from an origin country to the US could reduce security measures at US borders for reasons discussed earlier, thus increasing the probability of incidents to take place inside the US. Besides, an escalation to war between a given state and the US can reduce bilateral trade but might also independently increase terrorism activities inside the former. In either of these cases, the parameter on frequency of past incidents would be biased. This leaves third country incidents much better candidates of exogenous security than all other incidents. Thus we define an alternative indicator of exogenous security based solely on third inci- t [t...t-T]Thirdj.,t dents. Let Fjt(Third) = , be the frequency of past incidents perpetrated in third nT countries, we thus define jt (Third) = Fjt(Third).Ft US US to be an alternative proxy of exogenous security at the US borders. Because they are the most closely linked to our theory, this variable, together with jt will be our main two variables of interest US in the next sections. The dependent variable we study is bilateral US imports. We have chosen to work with data at the product level in order to control for the relative specialization of countries which we already suspect (see section 2) to be correlated with both measures, bilateral trade and terrorism activities. As well, a product level analysis allows us to investigate the differential effects of transnational terrorism and 10We have also considered time horizons of 3, 7 and 10 years. The results qualitatively very similar to a 5 years horizon. They are available upon request 15 bilateral security measures across sectors. Something that has been so far overlooked in other analyses of the effects of transnational terrorism on bilateral trade flows. We extract 1968-2000 bilateral imports of the United States at the product level (SITC4/5 digits) from the NBER World Trade Data complied by Feenstra and Lipsey. The data however, provides only values of flows that exceed 100,000$ per year. This constitutes a potential problem as most origin countries of terrorism are LDCs that export little of many products and too much of a very few set of others where they are really specialized. Thus, neglecting small amounts could result in an over-representation of products of specialization in the dataset, possibly less sensitive to terrorism attacks. This could end up underestimating the impact of terrorism activities on trade. To deal with this problem, we completed the NBER dataset with the FLUBIL trade dataset from the French National Institute (INSEE), reporting flows over 1,000$. FLUBIL is basically an updated version of the OECD dataset on bilateral trade flows where some aggregation check-ups and minor corrections have been undertaken. It also completes the NBER dataset as it runs until 2002. The sources of the rest of the variables that are used (i.e. traditional gravity and control variables), are listed in the appendix of the paper. 5 Econometric Results We want to study a bilateral US imports relation based on the trade equation (1) or its developed version equation (2), where exogenous security measures directly affect transaction costs. Let trans- action costs be expressed as: Tj = Distj. Sj.e( v.dvj) v . Thus, trade costs depend on geographical distance between j exporter and the US border, a set of dummy variables (dv) designating common language and contiguity with the US, and finally security measures at the US borders. Let further S = S(Zk), represent a variable of exogenous security depending on a set of K alternative variables Zk, each representing a measure of past incidents frequency. By approaching labor by the GDP of the importer, the productivity term a by GDP per capita and the number of varieties by GDP of the exporter in equation (2), taking logs and indexing by time (t), the relation to estimate for each good (g) that enter the US market becomes: log(mgjt) = log(USGDPt) + log(GDPjt) + log(GDPcapjt) + (1 - )log(Distj) +(1 - )1Contigj + (1 - )2Com.languagej + kZk,jt - log(Pt ) + g + t + ugjt g (10) k where g and t are good and time fixed effects, ugjt is the residual. The k are expected to be negative: an increase in past incident shares, increases current security measures (to prevent from potential future incidents), which leads to a decrease in US imports. The US GDP has been removed from the equation as its variation is fully captured by the time fixed effect. Also, as we do not observe the price index P, it is not a strong assumption to assume that it is captured by the time and product 16 fixed effects. We have alternatively run within-form equations where each import value of a given product from any given country is expressed as a deviation from its mean value over the period: (log(mgjt)) = log(mgjt)-log(mgj.), where the overline designates the mean over the period. This alternative equation has the advantage to implicitly although fully account for country fixed effects, along with (coun- try*product) specific effects, that capture the degree of specialization of the country in a given prod- uct. Also, by accounting for fixed effects, these within regressions enable to account implicitly for sanctions taken against particular countries like Cuba or Lybia over the period. However, it has the shortcoming to wipe out all time-constant variables. As most of our gravity (distance, contiguity, common language) and other control variables (see below) do not change overtime, we prefer showing mainly the pooled fixed effects regressions. The main within regression results are also shown in the following tables. All gravity and other control variables in the equation are listed and described in the appendix. The k are semi-elasticities as they are coefficients on frequencies (not in logs)11. At each time we find it necessary, we then convert those coefficients into elasticities at median points. It is important to detail however the computation of elasticities when we introduce our main (interaction) variables that proxy security. As noticed the indicator is a product of two frequencies. Its related coefficient, say , represents the semi-elasticity of US imports to the exogenous security indicator and is quite hard to interpret in simple economic terms. A further simple manipulation, however, enables a much better interpretation of the results. Notice that jt varies with both, past incidents share against the US and past incidents share US that originate from j (i.e.jt = FUSt.Fj,t). Yet, one can observe from appendix 2 that most of the US variation in the data comes from the second term. In fact, the first term, FUSt, varies relatively little : one fourth to one half of the total listed incidents in the world hit the United states across the whole period. Thus, for a better interpretation of the results one can simply fix FUSt to equal its average mean 0.35 and then compute the inferred elasticity of US imports to the frequency of past jorigin incidents. One obtains: Fj,t = 0.35..Fj,t m Needless then to say that because of the skewness of the Fj,t distribution (only a small fraction of origin countries account for most of the incidents), only some few export countries to the US should be significantly affected by the incidents. As a matter of fact, the median frequency of incidents perpetrated by an origin country is 1 per thousand and only 1% of the countries are at the origin of more than 5% of world's total incidents over the period (see Appendix 2). Then, for those risky countries, Fj,t is relatively high and thus the corresponding import elasticity m is expected to be significant. 11Needless to note that one main reason why we use frequencies in absolute values not in logs is that because around 50% of the frequencies of incidents have 0 values, see appendix 2 17 Table 4 presents a first set of results. Notice first, that in all the regressions presented the usual variables in the trade literature (GDP, distance, contiguity, common language) appear with the ex- pected signs and magnitudes12. The GDP per capita variable appears insignificant however, partly because it might not be a good proxy for productivity at the product level13. Second, before including our preferred variable of counter-terrorism measures (j ), we begin US our empirical investigation by including a terrorism variable computed at the bilateral level. That is the frequency of incidents, originating from a country j and directly targeting the US computed nUS as Fj US = j . It is somewhat the outcome of the interactive behavior of both terrorists and US n authorities. This variable however, has a serious shortcoming. As it is defined at the bilateral level it is likely to be endogenous to bilateral trade for reasons detailed already in the stylized facts and theory sections. The effect of bilateral incidents appears however, to be negative on bilateral US imports and statistically significant at 10%, with a semi elasticity of 4.3. The induced elasticity computed at the median point is thus around 0.004, an extremely low figure. But because we suspect endogeneity between bilateral trade and bilateral incidents, we define an alternative variable where the bilateral frequency is computed over the past 5 years of observations. Column 2 shows then that the effect of terrorism incidents increases by more than 70% although it does not gain much in significance. In column 3, we show results where we have split those incidents into three categories with respect to their location: those perpetrated against and within the US, those targeting US interests in the origin country of the terrorists and finally, those targeting the US in third countries. It appears that incidents perpetrated within the US, together with incidents in the home country, do not seem to affect significantly US bilateral imports. By a sharp contrast however, incidents perpetrated in third countries appear to affect negatively and very significantly (1%) exports of origin countries to the United States. Now, if computed at median levels, the elasticity is null because the median frequency of incidents perpetrated in third countries is null. But if one believes that the obtained 180 semi- elasticity is representative of the true effect of incidents, perpetrated in whichever location, then the resulting elasticity of incidents at the median point is around 0.18 (i.e: a 1% increase in incidents against the US results in a reduction of their imports of around 0.18%).14 12 The impact of distance is around 2 times smaller than in the rest of the literature but this is due to the nature of the panel where only the US is the importer. In fact, as we are accounting for contiguity in our regression, the distance variable looses most of its variability as all potential exporters are now at relatively comparable distances from the US. 13 We have also run the same type of equations at the aggregate level where we do find a robust positive effect of GDP per capita. Regressions can be provided upon request. 14 Our premise is that US authorities are imposing counter terrorist security measures against origin countries of terrorism. In our theory, this provides a channel through which bilateral US import volumes might affect endogenously (and positively) terrorists activities. Another "security channel" though could alternatively come from security measures implemented by "source" countries' governments. Indeed, if an origin country of terrorism sends large export volumes to the US, then its government may have a larger economic incentives to prevent terrorism against the US. It thus implements counter terrorist and security policies aimed at reducing terrorist attacks against the US interests, on its own territory. One would expect therefore a negative relationship between the volume of US bilateral imports from that "origin country" and the frequency of terrorist incidents against the US in that country. At the same time though, a terrorist organization of the "origin country" would find it relatively easier to hit US interests in a third country (i.e. a location substitution effect). Therefore, by the same token, one would also obtain a positive relationship between the 18 Our theoretical set-up mentions that one good way to capture the efficiency of terrorist organiza- tions that are targeting not only the US but all other countries. Also, another good way to capture the efficiency of US authorities is to consider not only perpetrators from one given country j but perpetrators from all countries together. We thus introduce together into the equation the frequency of incidents originating from a country j (against all targets) and the frequency of incidents against the US (from all countries of origin) as an alternative to the bilateral frequency of incidents variable. Columns 4 and 5 report the results for those variables computed respectively to express current and last 5 years of observations. In magnitude terms, the effects seem to be comparable to those reported earlier in columns 1 and 2. What is important to notice though is that the effects are now much more statistically significant (1%). Finally, our theory mentions that the interaction of terrorist and US authorities efficiencies should reveal even better the impact on security and thereby trade. We thus introduce to the equation the interaction variable , as an alternative security proxy. Namely, this is the product of the share of incidents introduced separately in the latter two regressions. Column 6 shows that the corresponding coefficient is negative and statistically very significant. The inferred elasticity m computed at the median point (1 per thousand of incidents originating from half of the countries) is around 0.0055: this is to say that for half of the export countries in the sample, a doubling of the frequency of incidents appears to be reducing US imports only by 0.55%. Now, although very small on average, that impact could be much more significant for origin countries at the top of the distribution of incidents. Thus, Colombia, a country associated with more than 20% of incidents against the US in some years can then be highly affected as the corresponding elasticity of US imports to past incidents that originate from these countries is respectively around 1 and 1.25. In column 7, we split our interaction variable between incidents perpetrated in own country and incidents perpetrated outside the country. Despite a non significant impact regarding incidents in own country, we obtain a very significant and negative effect of incidents located in a different country. Notice here that the third country estimator is around 5 times higher than all-incidents estimator shown in column 6. Table 5 keeps on using the third country based proxy for exogenous security while introducing progressively all possible controls (column 1 is the benchmark, identical to column 7 in the prior table). As a matter of fact, in order to have a better estimate of the magnitude of the terrorism effect, one needs to control for many other sources that could co-vary independently with terrorism acts on one hand and trade flows on the other. We begin by introducing a set of controls directly related to cross-border security between the US and their partners. In column 2 of table 6 , we include a dummy revealing an occurrence of a Militarized Interstate Dispute between a given country and the US, lagged volume of US imports from the "origin country" and the frequency of terrorist incidents in a third country. As incidents perpetrated in third countries appear to affect negatively and very significantly exports of origin countries to the United States, this suggest that such an alternative "security channel" is not empirically important. As is seen later in section 6.3,, a direct measure of security policies (business visas) of the target country (the US) againsts entry of residents from "terrorist origin countries" is also consistent with our.view. 19 over 10 years of observations as in Glick and Taylor (2005) and Martin, Mayer et Thoenig (2005). The data comes from the Correlates of War project. The sign of the coefficients is negative but not always statistically significant, possibly because we are working on a different panel at the product level. In the next tables we'll see that the impact of war differs across types of products. The inclusion of this measure of cross-border security however, reduces only slightly the magnitude of the coefficient on past terrorism incidents. Second, there are also some reasons to believe that two countries sharing the same types of political and economic institutions on the one hand could also share lower transaction costs and thus make more trade. On the other hand, this institutional proximity could lower the occurrence of terrorism attacks between them. In order to control for this effect, we add a dummy variable constructed from PolityIV dataset that takes on 1 when the polity variable (a grade that measures the degree of good governance) is as high as that of the US15, and 0 otherwise. But the effect, although positive, does not appear to be significant and leaves the variable of terrorism incidents unaffected. We next introduce a series of controls related to insecurity that originate specifically from the exporting country. The objective, here again, is to isolate all the forces that affect both bilateral trade and terrorism incidents. The progressive inclusion of a civil war dummy, a newstate exporter dummy, a proxy of good governance (i.e. polity2 variable in PolityIV, varying from -10 to 10), measures of ethnic or religion fractions (from Alesina et al (2003) dataset), reduce further by a third the magnitude of the coefficient on past frequencies of incidents, although without affecting its high significance in the pooled regression (i.e. estimators reduced from 80 to 57).16. Column 9 introduces almost all of the control variables together17 and shows further that the impact of third countries incidents variable is still significant with a semi-elasticity that reaches 47. Finally, as mentioned earlier, we run a within type regression that accounts for (country*product) dyadic effects in order to account for country specialization. The effect of the terrorism variable based on third countries, appear again with the a magnitude similar to that obtained from the prior regression, if one accounts for standard errors. To sum up, if the true semi-elasticity is say around 40, the inferred elasticity m computed at the median point (1 per thousand of incidents originating from half of the countries) is around 0.015. This is still not a high figure. However, those exporting countries which happen to be at the origin of high terrorism activities over the period like Colombia (more than 20% share of total incidents in some years), tend to be associated with an elasticity of at least 2.8, almost three times as much as that estimated earlier. 15The US grade is 10, the maximum that could be obtained by a ranked country 16Notice however, that most of these variables appear to be statistically insignificant. Religion Fractions in a exporting country seems however to be good for trade with the US. This result is consistent with Alesina et al (2003) findings concerning the role of this variable on various outcomes. 17To avoid multicollinearity, we have removed Ethnic fractions and newstate exporter dummy from the regression 20 6 Terrorism to Reveal Security Although we introduced many controls, we still need to show further that what we are picking is really a specific terrorism effect. Besides, we lack variables describing directly security measures. Thus, although consistent with our story, we could not prove so far empirically that the relationship between terrorism incidents and trade is really due to those measures. This section tries to go further into investigating the relationship between trade and terrorism through the security channel. 6.1 The impact of human victims In order to see first whether we are really capturing a specific terrorism effect, we interact the variable of past incidents shares with the average number of human victims per incident perpetrated by the terrorists of a given country j. Only incidents in third countries are considered here, as we know from the previous section that they seem to pick up most of the exogenous effect of terrorism on security and trade.We expect those incidents with high number of victims to affect even more current security measures and thus bilateral US imports. Table 6, column 1, shows the results for the complete specification. We define incidents as being relatively harmful in terms of casualties when they result in a number of human victims (deaths and injuries) higher than the standard deviation from the average in the sample. The average number of victims in the sample is around 3 by incident while the deviation is around 10. Then, we construct a dummy that takes value 1 when the resulting number of victims passes 13 (i.e. higher than the average+std) and 0 otherwise. The interaction term in column 1 (table 6) is negative but statistically insignificant. The impact of victims becomes statistically significant when their number becomes higher than 5 standard deviations (i.e. more than 50 victims). Column 2 shows indeed that the negative effect on US imports is up to three times higher when the incident is very harmful. The number of victims variable is a specific feature of terrorism and hence is completely consistent with the view that we are really picking up the impact of terrorism on trade. However, we still do not know whether this impact is truly coming from high security measures at the borders or whether it is due to higher insurance costs, or a boycott effect from the US consumers. We develop in what follows a strategy that could help us identify better the security effect. 6.2 Discussing further the security effect hypothesis By taking advantage from trade observed at the bilateral and product level, we take three further routes to analyze whether or not the impact of terrorist incidents are informing on security measures taken at the border. First, recalling our theory, we expect small partners of the US to be much more affected from terrorism than its big partners. The reason is that American citizens' welfare should be more depen- dent on big trading partners which then incite US authorities to limit their security measures towards the latter. In that respect, higher terrorism activities in the past might be more harmful to small 21 partners, but less harmful to big partners. Table 6, column 3, shows indeed that when the GDP of the partner increases the effect of terrorism incidents that originate from the latter decreases on US imports 18 . This size effect does not alter however that of the high-number of victims. This suggest then that the country size effect does matter but for incidents that do not result in a high number of victims. Second, if terrorism increases security controls at the borders then we expect terrorism acts to result in higher time spent at the borders. Thus, time-sensitive products should be much more affected by terrorism than time-insensitive ones. We take advantage from a study by Hummels (2002) where he estimates the average sensitivity of days spent in transport on trade at the SITC2 product level. We classify those products where time-sensitivity of trade is higher than -0.01 (and statistically significant) into a time-sensitive product category and the rest, usually around 0.005, into a time- insensitive categorie19. Table 6 again, shows that indeed time-sensitive products are more sensitive to terrorism acts than the rest. They are even more than 4 times more sensitive when the number of victims per incident is very high. Third, we expect that terrorism against the US affects networks formation between the latter and the country of origin, if terrorism results in lower issuing visas and higher visa controls at the borders. Thus, if security at the border matters, we expect products that ask for networks and where market information is costly (i.e. needs more labor mobility) to be more sensitive to terrorism acts in the past than those products negotiated on global markets where information on prices and quantities is readily available. We thus split the sample by three sets of products classified by Rauch (1996) into products in organized exchange, referenced prices products and differentiated products. Table 6 shows the result for the three subsamples: In the case of organized exchange products, the impact of incidents is insignificant even when they result in a high number of victims. In the case of referenced price products, the impact is as high as for differentiated products (semi elasticity around 52). In the latter case however, when those acts result in a high number of victims, the interaction term shows that the sensitivity to terrorism acts is 5 times higher. This last result is interesting to discuss in the perspective of the alternative boycott explanation of the effect of terrorism on bilateral trade flows. If indeed, a change in US consumers' preferences is the explanation of the negative impact on US imports of terrorist incidents emanating from origin countries, then we should expect this boycott effect to be stronger on standard and referenced goods than on differentiated products, as the first can be more easily substituted towards alternative supply sources. The fact that the impact of terrorist incidents on differentiated products is stronger than on standard products, suggests on the contrary that a change in US consumers preferences is unlikely to be an important explanation of the negative impact of transnational terrorism on US bilateral trade flows The following section goes further in confirming that the security channel is more empirically 18We also find the same qualitative result when we interact the third incidents variable with a dummy that takes on 1 when the size of the country in terms of GDP is higher than the median size country. 19Standard error of the estimates were not provided. Hence, we could not compare statistically the level of estimators with each others. That is why we have chosen the threshold method where 0.01 seemed to be a clear cut between insensitive and sensitive-time products. 22 consistent with the data. 6.3 Terrorism, Visas and US Imports Here, we pursue our investigations by running a series of regressions where we could employ a true variable of bilateral security at the borders but on a much smaller period. We thus assemble data on the number of non-immigrant visa issuances by partner country from 1997 to 2002 (last year of our US imports dataset). These data are provided online by the US Department of State20. We have chosen to work on the number of visas issued for Business (B1) and Business and Leisure (B1-2), assuming that those who come for both Business and Leisure decide to do so primarily for business activities21. Now, the rate of visas issued (i.e. ratio of number of visas to total visas demand) would have been even a better proxy for security, as it informs on the number of visa denied as well by the United States. However, and probably for political reasons, we could not find this information on the Department of State website. We want to investigate whether the impact of terrorism incidents on trade in differentiated (network-related) products is truly transiting through the number of issued visas for Business. Hence, on one side we study the relationship between terrorism incidents and the number of visas issued (this is to be called our empirical model 1, hereafter) and on the other side, we study the link between the visas and trade in differentiated products (model 2, hereafter). Model 1 will also serve as a first stage regression when we run an instrumental variable regression of US imports later on. Table 7 presents the results. The first two columns present two alternative econometric methods (Product/year fixed effects and Within) to explain the business visas issued, using mainly gravity type determinants. We add further both types of terrorism incidents based on origin and third countries. As for US imports, third countries incidents variable appear to affect significantly business visa issuance. However, no evidence is provided for incidents perpetrated on the origin country soil. In return, Columns 3 to 5 investigate the impact of business visas on US imports. We expect the effect to be positive and statistically significant for differentiated products, and no effect for organized exchange products. Column 5 confirms the first intuition: namely a 10% increase in visa issuing increases by almost 5% trade with US in differentiated products. However, the effect of business visas appears to be negatively affecting trade in organized exchange but this effect is not robust across specifications22. Finally, we have also run an instrumental variable regression in model 2 where the number of business visas is instrumented by all variables described in model 1. The chi-squared Anderson statistic presented rejects the exogeneity hypothesis of the number of visas and the instruments pass the over-identification test. The effect appears now to be higher (more positive) whichever the class of 20http://travel.state.gov/visa/frvi/statistics/statistics 1476.html 21Only citizens of countries that are not part of the Visa Waiver Program are included in our analysis. Hence, most of the OECD countries, part of this program, are not included in the panel because their nationals do not need visas in general to enter the US for Business or Leisure for a short stay (under 3 months). 22Results upon request 23 products. In particular, and as expected, the impact of the number of visas on US imports of network related products is now 25% higher than in the fixed effects regression presented earlier, the impact on imports of referenced prices goods becomes now slightly positive and statistically significant and the effect on organized exchange goods appears to be insignificant. We have also considered other alternative types of instruments such a series of frequency of inci- dents in third countries lagged over several years (generally up till 6 years) and bilateral type incidents in third countries (incidents targeting directly the US, but perpetrated in third countries) lagged also over several years. The results remain unchanged. They are not presented here to save space23. From the IV regression above one can then easily compute the impact of incidents on us imports via the number of delivered visas. The elasticity at the median point would be the product of the elasticity of trade to visas and that of visas to incidents shares: = (0.69)(800.0010.35) 0.019. This is comparable to the early figures in the prior sections where the number of visas was not yet introduced into the study. 7 Conclusion In this paper, we have asked what is the impact of security, to prevent terrorism, on bilateral trade. To this end we have set up a theory which shows that the impact goes not only from terrorism to trade. Trade might, in turn, increase the probability of terrorism acts. Our theory however, allows for a strategy to condition out the latter, in order to identify the true impact of terrorism. We have shown in particular, how past incidents located in third countries (anywhere in the world except the origin or the target country) can constitute good instruments of current security measures at the borders of the latter. We have run our tests on US imports. We have shown that past terrorist acts, perpetrated by groups from a given country against the US, affect its exports to the latter. The level of the impact is multiplied by three when the acts result in a relatively 'high' number of victims (ie. higher than a standard deviation from the mean number of victims over the period). To fix ideas, a 1% increase in the frequency of terrorism acts originating from a high-terrorism origin country, say Colombia, against the US, reduces imports from Colombia by 3%. This effect reaches a striking 10% decrease in US imports when terrorism attacks have important victim consequences. But this high figure is rather an exception. Only 1 percent of the countries (i.e. the most risky ones) are associated with significant effects on their exports to the US. For an extreme majority of cases, the elasticity of US imports is very much lower. Further, we expect that security measures at the borders are time costly and thus should affect more time-sensitive products (foreign newspapers, live animals, fresh fruits, etc...). We also know that they could affect international networks and business through limiting the movements of businessmen and the issuing of visas. Thus, products that are sensitive to these features could be also more affected 23They can be asked for upon request 24 by higher security to prevent terrorism. Our results appear to be perfectly consistent with these two views. We have found that the negative impact of terrorism is two to three times higher for products that have these characteristics. Further, using an additional dataset from the department on state on visa issuance from 1997 to 2002, we have shown how terrorism affects the number of business visas delivered by the US, thereby impacting significantly bilateral exports in differentiated products. All these results suggest that security to prevent terrorism does matter for US imports. What can we conclude from these results? As long as US imports come mainly from countries that do not represent a high risk in terms of terrorism acts, the US consumers should not be too much affected by security measures at the borders. However, those few countries at the origin of most of the attacks towards the US could be highly affected, especially those countries for which the US often constitutes a significant market for their export products. Hence, the protection of US lives might be undertaken at the expense of some foreign less developed countries' economies. Our results are consistent with the role played by security measures at the borders. It should be noted however that other elements might as well affect the nexus between trade and transnational terrorism. For instance, changes in the behaviors of insurers (higher rates of insurance prices) or changes in consumer choices (discrimination and embargo) could also affect trade and consequently terrorist attacks. Besides, we assign in this paper each terrorist attack to one particular origin. We know however that this is only partly true in today's changing forms of terrorism where terrorist organizations are increasingly becoming more multinational. Put differently, this paper does not study the indirect impact of terrorism from one country of origin on security measures over other suspected countries, which for instance might host groups from the same 'multinational' organization. One might argue that the indirect impact can be substantial as well. All these issues that arise naturally from our work, deserve to be specifically investigated in future research. 25 References Abadie, A. and J. Gardeazabal (2005). "Terrorism and the world economy". Harvard University Working Paper. Alesina, A., A. Devleeschauwer, W. Easterly, and S. Kurlat (2003). "Fractionalization". Journal of Economic Growth (2), 155­194. Anderson, J. and D. Marcouiller (1997). "Trade and security, i: Anarchy". NBER working paper n. 6223. Anderson, J. and D. Marcouiller (2002). "Insecurity and the pattern of trade: An empirical inves- tigaton". Review of Economics and Statistics 84, 345­352. Archick, K. (2005). "US-EU cooperation against terrorism". CRS Report for Congress. Blomberg, S. and G. Hess (2006). "How much does violence tax trade". Forthcoming in Review of Economics and Statistics. Feenstra, R. and R. Lipsey (2005). "World trade flows: 1962-2000". NBER working papers n.11040. Fratianni, M. and H. Kang (2006). "International terrorism, international trade and borders". Indiana University Mimeo. Frey, B., S. Luechinger, and A. Stutzer (2004). "Calculating tragedy: Assessing the costs of terror- ism". Institute for Empirical Research in Economics, Working Paper 205. Glick, R. and A. Taylor (2005). "Collateral damage: Trade disruption and the economic impact of war". NBER Working paper 11565. Helliwel, J. (1994). "Empirical linkages between democracy and economic growth". British Journal of Political Science XXIV, 225­248. Hummels, D. (2002). "Time as a trade barrier". manuscript. Malik, O. (2000). Enough of the Definition of Terrorism. The Royal Institute of International Affairs. Martin, P., T. Mayer, and M. Thoenig (2005). "Make trade not war?". CEPR Working paper 5218. Mickolus, E. (1982). International Terrorism: Attributes of Terrorist Events, 1968-1977 (ITERATE 2), [Computer file]. (Ann Arbor, Inter-university Consortium for Political and Social Research [producer and distributor] ed.). ICPSR. Mickolus, E., T. Sandler, Murdock, and Flemming (2003). "International terrorism attributes of terrorism events, iterate: 1968-2002". Mirza, D. and T. Verdier (2006). "International trade, security and transnational terrorism: Theory and empirics". CEPR working paper. Nitisch, V. and D. Schumacher (2004). "Terrorism and international trade: an empirical investiga- tion". European Journal of Political Economy 20, 423­433. 26 OECD (2002a). "Economic consequences of terrorism". OECD Economic Outlook 71, 117­140. OECD (2002b). "The impact of terrorist attacks of 11 september 2001 on international trading and transport activity". Working party of the Trade Committee, TD/TC/WP 9. Rauch, J. (1996). "Networks Versus Market in International Trade". NBER working paper 5617. Rodrik, D. (1999). "Democracices pay higher wages". The Quarterly Journal of Economics 114(3), 707­738. Sandler, T. and W. Enders (2004). "An economic perspective on transnational terrorism". European Journal of Political Economy 20, 301­316. Walkenhorst, P. and N. Dihel (2002). "Trade impacts of the terrorist attacks of 11 september 2001: A quantitative assessment". Working paper. 27 Appendix 1: Existence of a Bayesian Nash equilibrium security vector S = (S1,....SN) in the multi-country terrorist case: Transactions costs between countries 0 and j take the exponential form: T0 (S) = Tj eSj with > 0 j Let us denote the following Assumption : 2 E( ) Assumption A : < 1 + for all j [1,K] Vj Then we have the following result : · Under assumption A, there is a unique Bayesian Nash equilibrium of the security-terrorism game between country 0 and the K terrorist organizations. It is characterized by an equilib- rium security vector S = (S1,....SN), and an equilibrium terrorist vector Rj L ,(resp. i[1,K] RjH ) associated to the realization = L (resp. = H) of the terrorist resource cost. i[1,K] Equation (6) rewrites m0 j 1 1 = - E( ) µ Sj Vj Sj with LjT0 1- j Lje (1-)Sj m0 =j = h=N h=N LhT0 1- h Lhe (1-)Sh h=0 h=0 Hence Lje (1-)Sj µ 1 1 = - E( ) for all j [1,K] h=N Sj Vj Sj Lhe (1-)Sh h=0 28 and Sj = 0 for j [K,N] Denote h=N A = Lhe (1-)Sh h=0 and consider the equation Aµ 1 1 Lje (1-)Sj Vj = - E( ) for Sj Sj Vj Sj 2 E( ) [E( )]2 It is easy to see that for < 1 + it generates a unique solution Sj(A). As a matter of Vj fact, the function Aµ 1 E( )1 (S) = Lje (1-)S - - S Vj S (1-) Vj (E( ))2 is continuous and such that (0) = - and ( Vj > 0. By the theorem (E( ))2 ) = Lje of intermediate values there is at least one value Sj(A) which is such that (Sj(A)) = 0. The value is unique because for any S such that (S) = 0 and S Vj , one has (S) > 0. As a matter (E( ))2 of fact Aµ 1 E( ) 1 (S) = Lj(1 - )e (1-)S + - S S Vj 2 S Aµ 1 E( )1 Aµ 1 E( ) 1 = -( - 1) - + - S Vj S S S Vj 2 S Aµ 1 E( ) 1 1 > - - ( - 1) S Vj 2 S S 2 Aµ 1 E( ) 1 E( ) > - - ( - 1) > 0 S Vj 2 S Vj 29 Hence there can only a unique solution of (Sj(A)) = 0. The situation is depicted by a picture identical to figure 2 in the main text. It is easy to see as well that dSj µ 1 E( )1 1 = - - > 0 dA S Vj S - (S) and that limA Vj 0 Sj(A) = 0 and limA j S (A) = (E( ))2 Now we get the equilibrium value of A from the following equation: h=N A = (A) = Lhe (1-)Sh(A) h=0 (A) is decreasing in A (recall that Sh(A) is increasing in A and > 1). In A = 0, it has a positive value and it remains bounded when A goes to infinity, From this (A) - A is strictly decreasing with value (0) > 0 at 0 and value - for A tending to . Hence there is a unique A satisfying A = (A). Once we know A, we can recover the equilibrium security vector S = [Sj(A)]j , the corre- [1,K] sponding equilibrium efforts of terrorism of each group Rj = R(Sj(A,L)) and Rj = R(Sj(A,H)) L H and the probability of non occurrence of a terrorist act in country as L H E(0) = 1 - ii =K L Rj L Rj =1 j + (1 - j ) Rj + Sj L RjH + Sj Trade flows are immediately obtained from 1- LjT0 j Lje (1-)Sj m0 = j = h=N h=N 1- LhT0h Lhe (1-)Sh h=0 h=0 QED. 30 Appendix 2: Bayesian revision of beliefs after past terrorism in a third country. We provide here a simple justification of why beliefs of the government can be correlated to past terrorist actions in third countries. Consider the following timing. At the beginning of the period, a terrorist organization k tries to hit citizen or economic interests of country 0 in the rest of the world but not in country 0 itself. The technology is the same as before, namely in country j = 0, a terrorist organization k maximizes MaxR Rk,Sk Vk - kRk j j j j j k where Rk,Sj is the probability of success of a terrorist act in country j committed by orga- j nization k against country 0. with j Rk,Sj = j Rk j j Rk + Sk j which depends positively on the amount of resources Rk invested by the terrorist organization and j negatively on some specific factor Sk to country j (security measures, environment, political stability links between countries k and j, etc...). k is the marginal resource cost of the terrorist organization j and Vk is the perceived visibility gain that is enjoyed by the terrorist organization when the terrorist act is successful in country j against country 0. The solution of (3) gives immediately: the reaction curve of terrorist group k in country j j j Rk = R(Sj,k) = j SkVk - Sj k and the frequency of terrorist acts by organization k in country j against country 0 is j k = 1 - j kSk j Vk 31 as k can only take two values L and H with L < H, let us denote 0 and 0 = 1 - 0 L H L k k k respectively the initial beliefs that the government of country 0 has on the value of k. Assume also j j that Sk/Vk is iid distributed across countries and follows a density law f(.) Then applying Bayes' law gives us the revised belief of the government of country 0 after having j observed k in country j L -k]2 j L 0 f([1L ) k 1 = k 0 f([1L ) + (1 - 0 )f([1H ) L -k]2 j L -k]2 j k k or the odd ratio can be written as : j 1 - 1L -k]2 k 1 - 0L k f([1H ) = 1L L k 0 k f([1L ) -k]2 j and after the observation of all countries but 0 , one gets in the end: -k]2 j 1 - 1 L j=N k 1 - 0L f([1H ) = k 1 L L k 0 k j=1 f([1L ) -k]2 j To fix ideas, consider the case where Sk/Vk is exponentially distributed f(x) = e-x. Then we j j get j=N 1 1 j2 1 - 1L k 1 - 0L - [1-k] H -L = k e j=1 1L L k 0 k L j It is easy to see immediately that 1 is an increasing function of k (the probability of success of k a terrorist action by organization k in country j) 32 l cal ti oli olus,kc ics/f- tm .h P Mi um for 3: ee.org tmh ers um nsi s.e pap @ 2003)( lo 1978-200 tancs y/ onsorti ng it ziarg/ C le-gal org/ ol ucsd.edu/academ. and mi ac yti me e.s dd/di ar. p/rc rps w Fl w-i ersv ns edu/ (1982) h and francoi (ww rg/data atesofwl arc kc ter-uni and d.edu/i on orre ter s). In al atit Rese org .um .fr/francgraph/b W a.php) al Murdo r.e ww.c ww.stanford. victim ci 1977: r,e bn orldbank.o w cepii cidcm umenc So //w //w w. w. w. w. ter-dat ysical Do 1968- and Sandl ww ww ww ttp:h al ww Barbara w ttp:h Ph - atad nda se es es ri ri an se se of 2003 2002 2002 e e d 2002 2002 2002 Hum esc o ed imt imt . eri p lect 1968- 1968- 1968- no 1968- 1968- 1968- no (resp sour - 1: ion and irs,a and lat me arget duct ix ti bi raey ob- yb roupg data 1980 T cato P mei gener- pro ons L * T pairs and end i 5 and een orist and one eary , w ens n, riest ryt 4/ and riest ryt ryt et terr try ryt y ed try b 001)2 App el of me me Dim Origi and Coun ti Coun SITC lev Coun ti Coun earlyy rv ly y eral Coun coun Coun (onl se coun al and alit y) of tion ion et orld as W een-F( op-lev ors W) na,i and v3.02 tc na es Lipse INSEE t l- De )s (CO je ion Al r,e rst dat Indicat a w 3) fi Resolut zat E ari pro of ject T and Flubi (P Dat Correlates ra hau By (200 A datase ro IV y NBER onali sc try R 2/ d I t P y W orldbank) PI it eel rlet orl ar ol ounc Dataset ITE 1/ trade stra and dataset W men (W CE MID from W P Civil dataset ractiF data Dev Eas Kurlat het on -s is ') a s igu-t Di rac-F ts ort mmo p capit Con C ernancev rgeta 'T im er p Go gious ncideni hs terstate sp. eli In ms ,e R (re ar ateral GDP Engli , n' ables , (MID) W bil P taryli e ari errori D V T US G Distanc ,yti y2)tlio ons Origi' Language Mi pute Stat (P Civil Ethnic ti * 33 se uenci a a freq bi .x Arabi Cuba Cuba Ma Colom of Saudi )s n arey Origi 5 t um las 76 1 672 261 208 261 125 0. 125 0. 173. 0. 0. 0. 0. 0. the Maxim in( tlc ts P 672 209 049 037 022 017 014 007 0. 61. 0. 0. 0. 0. 0. 0. 99th ncideni of tlc P ycne 15 564 026 021 009 009 007 003 0. 18. 0. 0. 0. 0. 0. 0. 95th querF clt P 228 0 0 0 0 0 0 0 d 0. an 5th timsc dev 104 293 013 012 005 005 005 002 Vi 0. 10. 0. 0. 0. 0. 0. 0. Std. on 2 ticss 319 0. 001 001 0 0 0 0 0. 0. 0. Median Stati ev 352 122 006 004 002 002 001 001 Mean 0. 3. 0. 0. 0. 0. 0. 0. iptir sce tsn in of nci-i de inat- inat- j inat- D j er ni ncii p orig orig orig 2: atedcol yx from al ms ix tsn tsn tsn pro S tot US ctiiv ing de de atedcol de S but U t U t end 2 i F. of of ncii j ncii and ncii j try ) F. the ginat tj tj ) App t er of of of S ( coun U )dri el . b ori t ns . t from . from . from rd F. .E igr h S Lab reqF agai Num den reqF ngi reqF ngi reqF tj U 0 O( T( ngi thi F F F tj tj S S t ) ) S j able V )dri t U j U j S U tj t t igr h U j igr )dri ari F N F O( h V T( O( F F T( 34 Table 1: Rankings of Origin Countries across periods All period (1968-2003) 1968-1978 1978-1988 1988-1998 1998-2003 1968-2003 Growth of Origin Total Total incidents Country ranking incidents incidents rank incidents rank incidents rank incidents rank share** UNO* 1 4002 1357 1 1352 1 1051 1 242 1 -38,18% PAL 2 823 409 2 240 2 138 3 36 4 -69,49% COL 3 457 36 12 120 7 146 2 155 2 1392,53% TUR 4 292 46 10 169 4 63 10 14 15 5,50% IRN 5 275 16 27 162 5 90 5 7 22 51,66% LBN 6 236 21 20 178 3 34 17 3 40 -50,48% CUB 7 220 161 3 45 19 10 42 4 30 -91,39% ESP 8 207 31 15 122 6 49 13 5 26 -44,09% GRC 9 207 36 12 85 10 71 9 15 13 44,44% PHL 10 206 20 23 89 9 80 7 17 12 194,65% GBR 11 169 63 7 64 14 34 17 8 19 -55,98% PER 12 164 7 38 78 12 75 8 4 30 98,09% USA 13 162 77 6 72 13 11 38 2 45 -91,00% ARG 14 160 137 4 13 36 9 46 1 55 -97,47% PRI 15 153 91 5 62 15 0 0 -100,00% KUR 16 131 27 27 104 4 0 0,00% FRA 17 130 53 8 60 16 10 42 7 22 -54,22% RFA 18 126 33 14 91 8 2 76 -100,00% SLV 19 119 14 29 79 11 26 20 -100,00% ITA 20 110 52 9 40 20 8 51 10 17 -33,34% SOM 21 95 1 65 1 81 85 6 8 19 2673,21% IRQ 22 86 8 35 36 23 23 21 19 9 723,30% DZA 23 83 8 35 3 68 57 11 15 13 549,97% KOR 24 83 46 18 37 16 0,00% GTM 25 80 21 20 40 20 19 25 -100,00% YUG 26 77 37 11 23 28 12 34 5 26 -53,16% PAK 27 75 5 42 12 40 18 26 40 3 2673,21% JPN 28 69 22 19 32 24 14 30 1 55 -84,24% IND 29 66 17 25 22 29 21 23 6 24 22,35% LBY 30 65 1 65 56 17 8 51 -100,00% EGY 31 63 2 57 13 36 42 14 6 24 939,96% CHL 32 59 4 47 16 34 38 15 1 55 -13,34% IDN 33 57 15 28 4 62 4 64 34 5 685,74% KHM 34 54 1 65 2 71 51 12 -100,00% YEM 35 52 1 81 27 19 24 6 AGO 36 45 3 51 10 41 9 46 23 8 2557,66% PRT 37 45 5 42 38 22 1 89 1 55 -30,67% HND 38 44 30 25 14 30 0,00% NIC 39 41 13 31 18 31 9 46 1 55 -73,33% ISR 40 40 18 24 13 36 8 51 1 55 -80,74% JOR 41 40 5 42 22 29 9 46 4 30 177,32% MEX 42 40 28 16 8 45 2 76 2 45 -75,24% BOL 43 38 21 20 7 48 10 42 -100,00% MOZ 44 36 28 26 8 51 RUS 45 34 17 33 16 28 1 55 SLE 46 34 10 42 24 6 ETH 47 33 11 33 10 41 11 38 1 55 -68,49% SAU 48 33 1 65 1 81 12 34 19 9 6486,38% LKA 49 32 1 65 6 50 21 23 4 30 1286,61% ZWE 50 32 12 32 18 31 1 89 1 55 -71,11% AFG 51 31 1 5 56 13 32 13 16 801,29% ERI 52 27 26 17 1 81 -100,00% URY 53 27 26 17 1 55 -86,67% SDN 54 26 10 41 13 32 3 40 VEN 55 26 14 29 5 56 3 70 4 30 -0,96% BIH 56 25 23 21 2 45 SYR 57 23 1 65 13 36 8 51 1 55 246,65% NGA 58 22 1 65 1 81 2 76 18 11 6139,73% DEU 59 20 17 27 3 40 PAN 60 20 3 51 6 50 11 38 Total 10772 3106 3887 2884 896 *UNO=Unknown Origin ** calculations are based on the relative growth between the share of incidents in the first decade (1968-1978) and that of the last period considered (1998-2003). When the country has not been associated with incidents in first decade, the second decade is taken to compute the related growth rate of incidents. 35 Table 2: Rankings of Targeted Countries across periods All Period (1968-2003) 1968-1978 1978-1988 1988-1998 1998-2003 1968-2003 Targeted Total_incidents Growth share Country Rank (1968-2003) incidents rank incidents rank incidents rank incidents rank of incidents** USA 1 3822 1385 1 1125 1 854 1 458 1 14,60% FRA 2 649 75 6 368 2 180 2 26 4 20,13% ISR 3 647 385 2 140 5 98 7 24 5 -78,40% GBR 4 581 120 3 216 3 170 3 75 2 116,59% TUR 5 310 32 15 126 6 146 4 6 20 -35,02% RUS 6 276 65 7 86 9 115 6 10 12 -46,69% UNO* 7 269 30 16 191 4 44 11 4 24 -53,79% ITA 8 266 39 11 114 7 93 8 20 6 77,71% INT* 9 253 19 20 51 15 133 5 50 3 811,95% RFA 10 212 117 4 95 8 -100,00% ESP 11 218 82 5 62 10 62 9 12 10 -49,29% PAL 12 130 51 9 59 12 20 23 -100,00% JPN 13 123 18 24 46 16 56 10 3 29 -42,24% IND 14 119 34 14 37 19 34 13 14 9 42,69% CHE 15 107 20 19 56 14 22 20 9 14 55,94% IRN 16 106 17 26 60 11 29 14 -100,00% NLD 17 98 35 13 32 23 20 23 11 11 8,91% YUG 18 97 48 10 37 19 10 45 2 34 -85,56% CUB 19 96 56 8 24 29 11 41 5 22 -69,06% UFN 20 91 29 17 19 36 27 15 16 7 91,19% VEN 21 91 14 28 31 25 40 12 6 20 48,52% BEL 22 79 10 34 32 23 22 20 15 8 419,81% EGY 23 72 23 18 31 25 18 26 -100,00% CAN 24 71 13 31 21 34 27 15 10 12 166,57% IRQ 25 70 14 28 43 17 12 38 1 49 -75,25% IRL 26 68 36 12 18 38 11 41 3 29 -71,12% LBY 27 63 59 12 4 75 PRT 28 58 8 40 36 21 10 45 4 24 73,27% NIC 29 57 11 32 33 22 13 37 -100,00% CHL 30 55 19 20 26 28 10 45 -100,00% SWE 31 55 11 32 23 31 18 26 3 29 -5,49% AUT 32 50 10 34 21 34 18 26 1 49 -65,35% COL 33 50 14 28 16 42 12 38 8 16 98,02% MEX 34 50 18 24 16 42 14 34 2 34 -61,50% SAU 35 50 2 60 24 29 23 19 1 49 73,27% KWT 36 49 4 53 38 18 7 56 -100,00% ZAF 37 49 9 38 22 33 14 34 4 24 54,02% GRC 38 43 7 43 16 42 18 26 2 34 -0,99% AUS 39 42 2 60 15 45 17 30 8 16 1286,16% SYR 40 41 10 34 27 27 4 75 -100,00% CHN 41 40 12 50 26 17 2 34 JOR 42 39 8 40 17 40 10 45 4 24 73,27% ARG 43 36 15 27 14 47 5 67 2 34 -53,79% BRA 44 34 6 45 9 54 19 25 -100,00% LBN 45 34 19 20 11 53 4 75 -100,00% NAT 46 33 23 31 8 52 2 34 PHL 47 32 1 78 9 54 14 34 8 16 2672,32% POL 48 32 5 48 8 58 15 33 4 24 177,23% CYP 49 31 2 60 19 36 10 45 -100,00% KOR 50 30 2 60 9 54 17 30 2 34 246,54% Total 10772 3105 3887 2884 896 * INT=International Organizations; UNO=Unknown Targeted country **Calculations are based on the relative growth between the share of incidents in the first decade (1968-1978) and that of the last period considered (1998-2003). When the country has not been associated with incidents in first decade, the second decade is taken to compute the related growth rate of incidents. 36 Table 3: Ranking of incidents by Origin and Target Countries across periods All Period (1968-2003) 1968-1978 1978-1988 1988-1998 1998-2003 1968-2003 Groth share of Origin Target Rank Total incidents incidents rank incidents rank incidents rank incidents rank incidents** UNO USA 1 1591 774 1 392 1 298 1 127 1 -43,14% PAL ISR 2 317 240 2 46 12 25 20 6 19 -91,34% COL USA 3 232 13 35 45 13 54 7 120 2 3098,83% UNO FRA 4 212 19 21 128 2 60 4 5 22 -8,81% UNO ISR 5 192 103 3 51 8 36 10 2 60 -93,27% UNO GBR 6 176 32 11 62 5 58 5 24 4 159,91% PAL USA 7 175 71 6 38 18 48 9 18 6 -12,14% PRI USA 8 142 87 5 55 7 -100,00% PHL USA 9 120 13 35 40 16 57 6 10 9 166,57% UNO INT 10 119 7 62 23 30 65 3 24 4 1088,14% TUR TUR 11 105 17 25 71 3 16 29 1 114 -79,62% UNO RUS 12 103 17 25 32 22 52 8 2 60 -59,23% ARG USA 13 101 91 4 4 160 5 97 1 114 -96,19% GRC USA 14 100 31 12 38 18 22 22 9 10 0,61% ESP FRA 15 97 8 56 66 4 21 23 2 60 -13,36% UNO ESP 16 90 50 7 26 26 10 45 4 28 -72,28% KUR TUR 17 87 10 76 77 2 GBR GBR 18 86 23 18 38 18 21 23 4 28 -39,73% UNO TUR 19 78 14 32 31 23 31 12 2 60 -50,49% PER USA 20 76 6 73 41 14 28 18 1 114 -42,24% UNO UNO 21 76 4 96 56 6 16 29 -100,00% KOR USA 22 74 41 14 33 11 TUR USA 23 73 19 21 19 40 29 15 6 19 9,43% LBN USA 24 69 7 62 47 11 13 35 2 60 -0,99% UNO ITA 25 69 12 42 24 28 29 15 4 28 15,51% CUB USA 26 66 39 9 27 25 -100,00% UNO RFA 27 66 40 8 23 30 3 148 -100,00% IRN USA 28 64 12 42 38 18 11 39 3 44 -13,36% RFA USA 29 60 12 42 48 9 -100,00% SLV USA 30 58 2 153 40 16 16 29 -100,00% CUB CUB 31 56 36 10 9 87 8 57 3 44 -71,12% LBN FRA 32 56 5 81 48 9 3 148 -100,00% COL VEN 33 49 1 219 14 50 29 15 5 22 1632,70% UNO JPN 34 47 21 36 25 20 1 114 UNO PAL 35 46 16 27 22 34 8 57 -100,00% UNO YUG 36 46 16 27 22 34 7 69 1 114 -78,34% GBR IRL 37 45 30 13 10 76 5 97 -100,00% USA RUS 38 45 29 14 16 46 -100,00% CHL USA 39 44 2 153 11 68 30 13 1 114 73,27% ITA USA 40 44 29 14 10 76 1 317 4 28 -52,20% PAK USA 41 44 1 219 7 100 5 97 31 3 10642,75% IRN IRN 42 42 3 121 24 28 15 33 -100,00% UNO UFN 43 41 25 17 2 261 12 37 2 60 -72,28% UNO IRN 44 40 9 52 21 36 10 45 -100,00% ESP ESP 45 39 13 35 6 115 20 25 -100,00% PAL PAL 46 38 16 27 18 43 4 118 -100,00% YUG YUG 47 37 28 16 8 93 1 317 -100,00% UNO EGY 48 36 14 32 13 55 9 52 -100,00% UNO IND 49 36 14 32 10 76 10 45 2 60 -50,49% PAL GBR 50 35 13 35 13 55 7 69 2 60 -46,69% UNO CUB 51 35 18 24 13 55 2 199 2 60 -61,50% UNO SAU 52 35 1 219 14 50 19 26 1 114 246,54% UNO IRQ 53 34 7 62 20 38 6 79 1 114 -50,49% SOM USA 54 33 30 13 3 44 DZA FRA 55 31 1 372 28 18 2 60 UNO NLD 56 31 6 73 13 55 6 79 6 19 246,54% BOL USA 57 30 19 21 4 160 7 69 -100,00% HND USA 58 30 19 40 11 39 GTM USA 59 29 8 56 17 44 4 118 -100,00% IRN FRA 60 29 25 27 4 118 IND IND 61 28 13 35 12 65 3 148 -100,00% LBY LBY 62 28 28 24 SAU USA 63 29 1 0 11 39 17 7 5791,18% UNO BEL 64 28 5 81 10 76 12 37 1 114 -30,69% FRA USA 65 27 15 30 7 100 3 148 2 60 -53,79% Note: UNO=Unknown origin; INT=International Organizations **Calculations are based on the relative growth between the share of incidents in the first decade (1968-1978) and that of the last period considered (1998-2003). When the country has not been associated with incidents in first decade, the second decade is taken to compute the related growth rate of incidents. 37 Table 4: Impact of Terrorism incidents on Log of US imports Variables 1 2 3 4 5 6 7 Constant -1.089*** -0.233*** -0.238*** 0.165 0.772*** -0.227*** -0.216*** [0.156] [0.077] [0.077] [0.179] [0.116] [0.076] [0.078] Log GDP exporter 0.797*** 0.805*** 0.808*** 0.803*** 0.815*** 0.813*** 0.829*** [0.047] [0.047] [0.047] [0.047] [0.048] [0.048] [0.049] Log Weighted Distance -0.465** -0.472** -0.454* -0.485** -0.498** -0.489** -0.523** [0.230] [0.232] [0.230] [0.231] [0.233] [0.233] [0.234] English Common Language 0.380** 0.381** 0.373** 0.389** 0.392** 0.390** 0.437** [0.160] [0.161] [0.162] [0.162] [0.164] [0.163] [0.174] Contiguity 0.994** 0.999*** 1.007** 0.950** 0.936** 0.952** 0.850** [0.384] [0.381] [0.386] [0.388] [0.388] [0.387] [0.385] Log GDP per cap 0.02 0.014 0.013 0.015 0.006 0.009 0.002 [0.063] [0.064] [0.064] [0.063] [0.064] [0.064] [0.064] Frequency of Incidents originating from i against US: _ in current year -4.397* [2.616] _during last 5 years -7.316* [4.235] Frequency of Incidents originating from i against US (during last 5 years) : _ and located in i -3.764 [3.967] _ and located in US -81.545 [128.673] _ and located in third countries -180.106*** [35.838] Frequency of Incidents originating from i -4.470** [1.863] Frequency of Incidents against the US -5.495*** [0.732] (1) : Frequency of Incidents originating from i (during last 5 years) -6.923** [3.181] (2): Frequency of Incidents against the US (during last 5 years) -5.938*** [0.679] (1) * (2): Security proxy -16.327** [8.211] (1) * (2) : Security proxy based on incidents located in i -7.139 [7.529] (1) * (2) Security proxy , based on incidents located in third countries -80.887*** [29.030] Fixed effects: _ product (SITC 5 digits) yes yes yes yes yes yes yes _ year yes yes yes yes yes yes yes Number of Observations 699249 673725 673725 700297 673725 673725 673725 R-squared 0.26 0.26 0.26 0.26 0.26 0.26 0.26 Robust Standard errors provided in brackets with clustering by exporter * significant at 10%; ** significant at 5%; *** significant at 1% 38 01 _ ***323.2 ]552.0[ _ _ _ **986.0- ]213.0[ 802.0 ]536.2[ ***478.43- ]662.31[ 651.0- ]511.0[ 241.0- ]090.0[ 860.0- ]450.0[ **541.0- ]750.0[ 470.0- ]650.0[ 201.0- ]260.0[ *711.0- ]070.0[ 9 **632.0- ]390.0[ ***308.0 ]150.0[ *264.0- ]342.0[ 802.0 ]781.0[ ***251.1 ]483.0[ 240.0- ]880.0[ 102.6- ]828.6[ *907.74- ]075.52[ 571.0- ]511.0[ *961.0- ]290.0[ *401.0- ]160.0[ **451.0- ]560.0[ **521.0- ]250.0[ 750.0- ]750.0[ 650.0- ]070.0[ egaptxend'tnoc 8 stropmISUfogoLnotcapmistnedicni ***812.0- ]970.0[ ***708.0 ]340.0[ **725.0- ]222.0[ *782.0 ]171.0[ ***819.0 ]733.0[ 400.0- ]750.0[ 751.5- ]826.6[ *933.65- ]758.82[ 7 ***513.0- ]380.0[ ***518.0 ]840.0[ **345.0- ]832.0[ ***174.0 ]971.0[ **679.0 ]293.0[ 320.0- ]860.0[ 934.6- ]843.7[ ***413.67- ]423.72[ 6 ***903.0- ]480.0[ ***138.0 ]850.0[ *464.0- ]252.0[ **573.0 ]581.0[ **369.0 ]314.0[ 410.0- ]970.0[ 266.7- ]646.7[ ***925.77- ]603.92[ 5 ***822.0- ]180.0[ ***568.0 ]560.0[ **195.0- ]982.0[ **283.0 ]881.0[ *677.0 ]634.0[ 100.0 ]270.0[ 952.7- ]295.7[ ***199.38- ]125.92[ 4 ***612.0- ]870.0[ ***928.0 ]940.0[ **125.0- ]832.0[ **634.0 ]371.0[ **358.0 ]193.0[ 300.0 ]560.0[ 994.7- ]730.8[ ***329.08- ]150.92[ msirorreTfossentsuboR:5elbaT 3 ***232.0- ]980.0[ ***328.0 ]940.0[ **794.0- ]432.0[ **214.0 ]571.0[ **509.0 ]963.0[ 130.0- ]870.0[ 210.6- ]524.7[ ***492.08- ]675.82[ 2 ***292.0- ]180.0[ ***038.0 ]050.0[ **635.0- ]632.0[ **644.0 ]771.0[ **709.0 ]814.0[ 0 ]760.0[ 963.7- ]054.7[ ***642.87- ]205.72[ 371.0- ]011.0[ *351.0- ]290.0[ 980.0- ]170.0[ *041.0- ]170.0[ *201.0- ]650.0[ 440.0- ]060.0[ 530.0- ]470.0[ 1 ***612.0- ]870.0[ ***928.0 ]940.0[ **325.0- ]432.0[ **734.0 ]471.0[ **058.0 ]583.0[ 200.0 ]460.0[ 931.7- ]925.7[ ***788.08- ]030.92[ selbairaV tnatsnoC retropxe PDGgoL ecnatsiDdethgie WgoL egaugnaLnommoChsilgnE ytiugitnoC pacrep t 1-t 2-t 3-t 4-t 5-t 6-t PDGgoL nodesabyxorpytiruceS:)2(*)1( inidetacolstnedicni nodesab,yxorpytiruceS)2(*)1( seirtnuoc driht nidetacolstnedicni ytirucesnIlaretaliB etupsiDetatsretnIyratiliM 39 11 ]260.0[ *331.0- ]270.0[ ***751.0- ]450.0[ **931.0- ]260.0[ 480.0- ]011.0[ 490.0- ]660.0[ _ 900.0 ]800.0[ _ _ sey sey 371956 61.0 9 ]160.0[ 770.0- ]370.0[ 321.0- ]680.0[ 270.0- ]590.0[ 590.0 ]202.0[ 160.0 ]681.0[ _ 900.0 ]510.0[ _ ***342.0 ]470.0[ sey sey 371956 62.0 8 stropmISUfogoLnotcapmistnedicni ***532.0 ]470.0[ sey sey 527376 72.0 7 101.0- ]170.0[ sey sey 874276 62.0 6 900.0 ]610.0[ sey sey 176956 62.0 5 160.0 ]431.0[ sey sey 589656 62.0 4 520.0 msirorreTfossentsuboR:)d'tnoc(5elbaT ]181.0[ sey sey 527376 62.0 3 361.0 ]202.0[ sey sey 527376 62.0 2 ]160.0[ 970.0- ]870.0[ 380.0- ]480.0[ 520.0- ]790.0[ sey sey 691376 62.0 1 sey sey 527376 62.0 selbairaV 8-t 9-t 01-t SUnahtecnanrevogfognitaremaS ytiruceSlanretniretropxE retropxEnirawliviC etatswenasiretropxE ecnanrevoGetatsfognitar retropxEnisnoitcarfcinhtEfogoL retropxEnisnoitcarfnoigileRfogoL :stceffedexiF )stigid5 raey_ CTIS(tcudorp_ retropxe_ retropxe*tcudorp_ snoitavresbO derauqs-R retropxeybgniretsulchtiwstekcarbnidedivorpsrorredradnatStsuboR %1tatnacifingis***;%5tatnacifingis**;%01tatnacifingis* 40 Table 6: Terrorism and Security Related Effects: Victims, Partner Size, 'Just in Time' and Networks Variables Role of Shipping time Role of Networks Nb. Nb. Exporter Non time Time Organized Referenced Differentiated Victims (1) Victims (2) Size Sensitive sensitive Exchange Prices Products Constant -0.295*** -0.293*** -0.323*** -0.316 -1.041* 1.800*** 0.116 -0.576 [0.080] [0.080] [0.085] [0.353] [0.623] [0.682] [0.396] [0.628] Log GDP exporter 0.805*** 0.804*** 0.809*** 0.736*** 0.934*** 0.276*** 0.524*** 0.868*** [0.044] [0.044] [0.043] [0.030] [0.055] [0.066] [0.033] [0.055] Log Weighted Distance -0.529** -0.530** -0.508** -0.447*** -0.625** -0.322 -0.680*** -0.611** [0.221] [0.220] [0.218] [0.153] [0.270] [0.224] [0.133] [0.285] English Common Language 0.290* 0.290* 0.265 0.323** 0.363* 0.381 0.281* 0.263 [0.172] [0.173] [0.177] [0.127] [0.218] [0.252] [0.154] [0.245] Contiguity 1.010*** 1.017*** 1.059*** 0.819** 0.956** 1.824** 1.304** 1.249*** [0.367] [0.370] [0.375] [0.404] [0.411] [0.792] [0.512] [0.388] Log GDP per cap -0.006 -0.005 -0.006 0.033 0.01 -0.202*** -0.028 -0.054 [0.060] [0.060] [0.059] [0.042] [0.075] [0.077] [0.047] [0.076] (1) * (2) Security proxy, based on incidents located in third countries -46.543* -50.140* -84.532*** -48.863** -55.217* -50.858 -52.697** -51.146* [25.118] [25.573] [30.175] [22.930] [28.940] [32.766] [24.634] [30.659] Security proxy* Number of Victims higher than 1 std -71.298 [55.903] Security proxy * Number of Victims higher than 5 std -166.112** -171.595** -89.614* -235.212** -36.754 43.36 -245.514** [77.344] [73.851] [51.857] [103.924] [45.437] [53.261] [103.774] Security proxy * Partner size 34.599** [17.454] Military interstate dispute: t0 -0.161 -0.195* -0.165 -0.039 -0.332*** 0.233 -0.03 -0.340** [0.111] [0.105] [0.108] [0.118] [0.104] [0.214] [0.145] [0.139] t-1 -0.194** -0.177* -0.141 -0.067 -0.226** 0.019 0.098 -0.261** [0.087] [0.091] [0.095] [0.102] [0.092] [0.169] [0.117] [0.112] t-2 -0.074 -0.103 -0.083 0.018 -0.161** 0.039 0.122 -0.206*** [0.068] [0.064] [0.066] [0.086] [0.066] [0.138] [0.111] [0.063] t-3 -0.157** -0.151** -0.141** -0.027 -0.238*** 0.106 0.064 -0.261*** [0.065] [0.063] [0.066] [0.080] [0.072] [0.149] [0.108] [0.075] t-4 -0.102** -0.121** -0.090* -0.035 -0.169*** 0.045 0.036 -0.192*** [0.048] [0.050] [0.051] [0.069] [0.046] [0.116] [0.087] [0.044] t-5 -0.043 -0.057 -0.047 0.063 -0.164* 0.051 0.127* -0.160** [0.060] [0.056] [0.058] [0.054] [0.084] [0.117] [0.067] [0.068] t-6 -0.069 -0.054 -0.047 0.045 -0.134* 0.11 0.098 -0.138** [0.078] [0.070] [0.073] [0.089] [0.070] [0.160] [0.116] [0.062] t-7 -0.025 -0.04 -0.023 0.048 -0.116* 0.212 0.082 -0.116* [0.064] [0.057] [0.060] [0.062] [0.061] [0.137] [0.099] [0.064] t-8 -0.094 -0.103 -0.09 -0.016 -0.209** -0.018 0.096 -0.175** [0.082] [0.076] [0.078] [0.087] [0.082] [0.157] [0.097] [0.076] t-9 -0.094 -0.112 -0.104 -0.035 -0.227*** 0.022 0.048 -0.178** [0.084] [0.081] [0.082] [0.098] [0.080] [0.214] [0.132] [0.084] t-10 -0.072 -0.065 -0.065 -0.028 -0.167 0.23 0.117 -0.101 [0.094] [0.094] [0.095] [0.112] [0.108] [0.191] [0.130] [0.106] Log of religion fractions 0.243*** 0.242*** 0.232*** 0.196*** 0.274*** 0.06 0.197*** 0.286*** [0.075] [0.075] [0.075] [0.065] [0.087] [0.095] [0.071] [0.092] Fixed effects: _ product (SITC 5 digits) yes yes yes yes yes yes yes yes _ year yes yes yes yes yes yes yes yes Observations 673196 673196 673196 322151 308696 33021 103192 351045 R-squared 0.27 0.27 0.27 0.23 0.35 0.1 0.19 0.3 Robust Standard errors provided in brackets with clustering by exporter * significant at 10%; ** significant at 5%; *** significant at 1% 41 Table 7: Visas, Networks and US imports Model 1: Impact of incidents on business Model 2: Impact of business visas allowance on US imports Visas allowance Organized Referenced Differentiated Organized Referenced Differentiated Exchange Prices Products Exchange Prices Products Product and Within year effects regression Product and year effects Instrumental Variables regression 1 2 3 4 5 6 7 8 Constant 0.842*** 0.057 -0.006 -0.18 -0.564** -0.156 -0.05 -0.687*** [0.119] [0.049] [0.150] [0.133] [0.215] [0.160] [0.135] [0.200] Log GDP exporter 0.807*** 0.257 0.683*** 0.543*** 0.631*** [0.069] [1.503] [0.101] [0.082] [0.146] Log Weighted Distance -1.471*** -0.837*** -0.319 0.515 -0.522** -0.249 0.741*** [0.185] [0.284] [0.231] [0.378] [0.219] [0.183] [0.274] English Common Language 0.925*** 0.138 0.083 -0.465 -0.046 0.041 -0.600* [0.149] [0.286] [0.220] [0.341] [0.298] [0.214] [0.355] Log GDP per cap 0.07 1.072 -0.167* 0.019 -0.182 -0.189* 0.018 -0.181 [0.094] [1.463] [0.096] [0.084] [0.142] [0.110] [0.085] [0.138] (1) * (2) Security proxy, based on incidents located in third countries -88.522* -76.182*** [51.092] [26.398] (1) * (2) Security proxy, based on incidents located in country i 37.594* 9.858 [19.318] [9.234] Log of number of B. visas -0.272** 0.084 0.536*** -0.028 0.134* 0.693*** [0.120] [0.087] [0.130] [0.077] [0.069] [0.155] Control variables Military interstate Disputes Military interstate Disputes Military interstate Disputes (lagged over 10 years), (lagged over 10 years), (lagged over 10 years), Same governance than US Same governance than US, Same governance than US, Civil war, Civil war, Log of religion fractions Log of religion fractions Fixed effects: _ product (SITC 5 digits) yes yes yes yes yes yes yes _ year yes yes yes yes yes yes yes yes _Product*Exporter yes Anderson IV relevance test (Chi2) 49.66 16000 56000 Pvalue [0.000] [0.000] [0.000] Hansen overidentification test 2.95 2.83 4.08 Pvalue [0.399] [0.411] [0.252] Period 1997-2002 1997-2002 1997-2002 1997-2002 1997-2002 1997-2002 1997-2002 1997-2002 Observations 98953 98953 4184 13629 45027 4184 13629 45027 R-squared 0.76 0.13 0.07 0.11 0.29 NB:1/ In VI regressions, Log of number of visas is instrumented by the security proxy variables based on incidents in third countries and orgin countries and the rest of variables in model 1 2/ Log GDP exporter has been moved to left hand side in Instrumental variable regressions as it was multicolinnear to Log of number of visas (VIF related to GDP=105 and VIF related to Log number of visas=99) Robust Standard errors provided in brackets with clustering by exporter * significant at 10%; ** significant at 5%; *** significant at 1% 42 Figure 1: Location of incidents across Origin, Target and Third Countries Share ( in %) 1968 0 10 20 30 40 50 60 70 80 90 100 1970 1972 1974 1976 1978 1980 1982 1984 year 1986 1988 1990 1992 1994 1996 1998 2000 2002 incidentsinTarget incidentsinOrigin Incidentsinthirdcountries 43 mk S mk * T E T Sk* S Sk P E (No Occur.) Figure 2 44 mk S mk *T E E' T Sk* S Sk P P' Figure 3): comparative E (No Occur.) Statics 45