International Trade, OECD Membership, and Religion

Open economies review 17: 493 508, 2006 c 2006 Springer Science + Business Media, LLC. Manufactured in The Netherlands. International Trade, OECD Membership, and Religion HEEJOON KANG kang@indiana.edu MICHELE FRATIANNI fratiann@indiana.edu Kelley School of Business, Indiana University, Bloomington, Indiana 47405 Key words: gravity equation, OECD membership, religion, trade flows, transaction costs JEL Classification Number: F13 Abstract Transaction costs in trade gravity equation are proxied by the distance that separates two trading partners, under the assumption that the distance elasticity is the same across all trading partners. We show that distance elasticity, however, critically depends on whether trading partners are industrial countries (i.e., members of the OECD) or share same religion. These heterogeneities are both statistically and economically significant. For instance, expected trade flows are the largest when an OECD member trades with a non-member are nonreligious. Expected trade flows fall as much as by 62.9% between two non-religious, non-oecd members. Expected bilateral trade drops by 48.1% when both countries in the pair are OECD members while one is Christian and the other is Islamic. Both religion and OECD membership significantly affect the typical transaction costs implied by the gravity equation. The gravity equation (GE) has been widely used in the literature to explain international bilateral trade flows; see Anderson (1979), Frankel (1997, Ch. 5), Deardorff (1998) and Rose (2000), among others. These flows, adjusted for prices, depend on real income and per capita real income of each country and the distance separating the two trading partners; distance is a proxy of transaction and transportation costs. Other country-specific characteristics entering the equation are common land border, common language, common currency and membership in regional trade agreements (RTAs). The typical GE specification is as follows: ln(y ijt ) = α 0 + α 1 ln(x i x j ) t + α 2 ln(i i I j ) t + α 3 ln(d ij ) + α 4 FEAT ijt + u ijt, (1) where y ijt = bilateral trade flows between country i and country j at time t in real terms, x = real gross domestic product, I = per capita real income, d = distance, and FEAT is a vector of country-specific dummy variables indicating whether the two trading partners share a common currency or belong to the same RTA. FEAT also includes time-invariant factors such as common language, common colonizer, and shared land border as well as

494 KANG AND FRATIANNI year dummy variables. u ijt is an independent and identically distributed (iid) error term. Distance in (1) enters in log terms and the coefficient α 3, therefore, is the distance elasticity of bilateral trade. Distance is a friction: distant countries trade less than close countries, other things held the same. Recently, we have presented firm evidence that a trade GE with homogeneous distance elasticity may be misspecified (Fratianni and Kang, 2006). In particular, we have shown that distance elasticities are heterogeneous in that they differ depending on whether trading partners enjoy a similar degree of economic development proxied by membership in the Organization for Economic Cooperation and Development (OECD) or similar cultural characteristics proxied by a shared religion like Christianity or Islam. Our results were obtained from a large sample covering five-year intervals from 1970 to 1999. In this paper, we extend our previous results in three ways. First, we enlarge our sample period to include post 9/11 data; specifically our sample consists of annual observations on bilateral trade flows for the years 1980 2003. Second, we test for interaction between degree of economic development and religious affinity. We find that the coefficients of OECD membership and religion dummy variables are statistically very significant. When these dummy variables are multiplied by the (log) distance variable, both OECD and religious status generate statistically significant heterogeneous elasticities. Transaction costs, measured by the distance elasticity, differ substantially depending on OECD membership and country religion. In particular, transaction costs are smaller for Islamic countries than Christian countries, regardless of whether they belong to the OECD. Third, distance elasticities are not only statistically but also economically significant on trade flows. The largest expected trade flows occur when an OECD member country trades with a non-member country and if both countries are neither Christian nor Islamic. The expected trade flows are reduced for other OECD and religion configurations. In general, they are lower for non-oecd countries. For example, when one OECD member trades with a non-member and one country is Christian and the other Islamic, the expected trade flows are small, about 54% lower than the same OECD configuration but without religion. The structure of the paper is as follows. Section 1 describes the model, data, and estimation results. Section 2 shows the economic significance of the importance of the OECD memberships and religious status. Section 3 offers brief conclusions. 1. Trade gravity model and heterogeneous distance elasticity The dependent variable in Equation (1) is a real bilateral trade flow measured by averaging the exports reported by both countries. Table 1 reports descriptive statistics from annual data from 1980 to 2003, consisting of 122,402 observations. The dataset and data description are from Feenstra and are detailed in the CIBER (Center for International Business Edu-

INTERNATIONAL TRADE, OECD MEMBERSHIP, AND RELIGION 495 Table 1. Descriptive statistics. Variable Mean Std. Dev. Min Max Real trade flow 215,573 1,775,515 0.3064 104,541,260 Log of real trade flow 8.1436 3.2519 1.1830 18.4651 Log of real GDP 47.6103 2.7035 35.9590 58.1154 Log of real per capita GDP 15.1753 2.0967 8.0351 21.0682 Log of distance 8.2185 0.7663 4.0168 9.4215 0.0543 0.2266 0 1 0.4656 0.4988 0 1 NonOECD 0.4801 0.4996 0 1 * (log of distance) 0.4139 1.7458 0 9.4169 * (log of distance) 3.8858 4.1813 0 9.3993 NonOECD* (log of distance) 3.9188 4.1199 0 9.4215 BothChristian 0.4035 0.4906 0 1 BothIslam 0.0551 0.2282 0 1 OneChristianOne Islam 0.2680 0.4429 0 1 Neitherreligion 0.2734 0.4457 0 1 Both Christian* (log of distance) 3.2887 4.0342 0 9.4169 Both Islam* (log of distance) 0.4201 1.7534 0 9.1495 OneChristianOneIslam* (log of distance) 2.2004 3.6527 0 9.4215 Neither religion* (log of distance) 2.3087 3.7783 0 9.4190 Common currency 0.0088 0.0933 0 1 Common RTA 0.0238 0.1524 0 1 Inter-regional 0.1335 0.3401 0 1 Common border 0.0244 0.1542 0 1 Common language 0.2089 0.4065 0 1 Common colonizer 0.0820 0.2744 0 1 Colonial relationship 0.0205 0.1418 0 1 Note: The unit of real trade flow is thousand of US dollars. cation and Research) webpage at Indiana University. The mean of real trade is about $216 million, with a range spanning from $306 to $105 billion. The natural log of mean distance is 8.2185 with a range spanning from 4.0168 to 9.4215. The log of real GDP and that of real per capita GDP are, respectively, 47.6103 and 15.1753. Other variables are binary variables. The table also lists some descriptive statistics for the product of dummy variables and log of distance. The impact of RTAs on bilateral trade flows is captured by a dummy variable that is coded one if both countries in the pair belong to the same RTAs, otherwise it is zero. We identify 11 RTAs: ANDEAN (Bolivia, Colombia, Ecuador, Peru, and Venezuela), ASEAN (the Philippines, Indonesia, Malaysia,

496 KANG AND FRATIANNI Singapore, Thailand, Vietnam, Laos, Burma, and Cambodia), CARICOM (Antigua and Barbuda, Bahamas, Barbados, Belize, Dominica, Grenada, Guyana, Haiti, Jamaica, St. Kitts and Nevis, St. Lucia, St. Vincent and Grenadines, Suriname, and Trinidad and Tobago), CACM (Costa Rica, El Salvador, Guatemala, Honduras, and Nicaragua), NAFTA (Canada, the United States, and Mexico), MERCOSUR (Argentina, Brazil, Paraguay, and Uruguay), USIS (the United States and Israel), PATCRA (Australia and Papua New Guinea), ANZCERTA (Australia and New Zealand), SPARTECA (Australia, New Zealand, Fiji, Kiribati, Papua New Guinea, Solomon Islands, Tonga, Vanuatu, and Samoa), and EC/EU (Belgium, France, Germany, Italy, Luxembourg, the Netherlands, Denmark, Ireland, the United Kingdom, Greece, Portugal, Spain, Austria, Finland, and Sweden). Common RTA dummy takes the value of one when the trade countries belong to the same RTA and zero otherwise. OECD membership status is based on information available in 2003. We make a distinction among trade between two OECD countries, trade between an OECD country and a non-member country, and trade between two non-member countries. On religion, we make a distinction between Islamic, Christian and non-christian and non-islamic countries. A country is defined Islamic (Christian) if more than 50 per cent of its population are Muslim (Christian); the source on Islamic countries is the CIA Factbook (http:// www.cia.gov/cia/publications/factbook/) and the source on Christian countries is the World Christian Database (http://www.worldchristiandatabase. org/ecd/). As a result, the sample is divided into four different groups: cases in which both countries are Islamic (dummy variable BothIslam), those in which both countries are Christian (BothChristian), those in which one country is Christian and the other is Islamic (OneChristianOneIslam), and cases in which neither country is Christian or Islamic (Neither). We also investigate the interaction effects between OECD membership and religion status; this permits us to contrast, for instance, trade between an OECD Christian country and a non-oecd Islam country. Table 2 shows the breakdown of OECD membership status and religion. For instance, 46.56% of our 122,402 observations involve trade between an OECD member and a non-member country; among those, 48.29%, or 22.48% of the entire sample, involves Christian countries, the most prevalent configuration. The next common case, with a frequency of 16.7% of the entire sample, is when both partners in the pair are non-oecd members and each country is neither Christian nor Islamic. It is infrequent for both partners to be OECD members and not Christian. Among OECD members, only 0.61% are non-religious and 0.42% involve a pair in which one country is Christian and the other is Islamic. Also infrequent (0.55%) is the combination of one country in the pair being an OECD member countries being Islamic. There are no observations of both countries being simultaneously OECD members and Islamic. The first column in Table 3 shows the estimate of our basic trade GE, that is the model without OECD membership, religion and their interactions. We use year dummy variables to control for idiosyncratic fixed time effects but

INTERNATIONAL TRADE, OECD MEMBERSHIP, AND RELIGION 497 Table 2. Frequency of religion by OECD. (5.43%) (46.56%) NonOECD (48.01%) BothChristian countries (40.35%) BothIslam countries (5.51%) OneChristian OneIslam countries (26.80%) NeitherReligion countries (27.34%) 80.93% (4.40%) 0.00% (0.00%) 7.79% (0.42%) 11.28% (0.61%) 48.29% (22.48%) 1.18% (0.55%) 28.99% (13.50%) 21.54% (10.03%) 28.06% (13.47%) 10.33% (4.96%) 26.82% (12.88%) 34.79% (16.70%) Column total 100.00% 100.00% 100.00% Note: Values in parentheses are frequencies in total sample. do not report their coefficients to save space. The overall R 2 is 0.6784 and the coefficient of the log distance is 1.0874. These and other estimates are very similar to those reported in the literature by Rose (2000, 2003) and others. All parameter estimates are significant at the 1% level. Trade flows respond positively to real GDP and to real per capita GDP. The elasticity of real trade flows with respect to real GDP is the sum of 0.9073 plus 0.1339, slightly above one as predicted by complete specialization models of international trade (Deardorff, 1998). As population increases, for a given real income, bilateral trade flows are depressed. In general, common cultural, political, and historical variables such as common land border, common language, or common currency significantly increase trade flows. The estimate of α 3, the elasticity of trade flows with respect to distance, is 1.0874 with a standard error of 0.0080. An increase in distance of one per cent lowers real bilateral trade flows by 1.09%. The next column shows the estimates of the GE in the presence of OECD membership. We have defined three dummy variables:, which takes a value of one if both partners are OECD members and zero otherwise;, which takes a value of one if one partner is an OECD member and the other is not a member and zero otherwise; and NonOECD, which takes a value of one if neither partner is an OECD member and zero otherwise. The three dummy variables alter the intercept term. In addition, each of these three dummy variables interacts with the distance measured in logarithmic terms. The parameter estimates of the expanded model are again statistically significant at the 1% level. There are three different ways to add three mutually exhaustive dummy variables in order to avoid the so-called dummy variable trap. First, three dummy variables can be added without the overall intercept term. Second, two out of the three dummy variables can be added in addition to the overall intercept term. Third, with the overall intercept term, three dummy

498 KANG AND FRATIANNI Table 3. Variables OLS estimates with OECD variables. Standard base model With OECD and OECD* (log of distance) Intercept 28.3553*** (0.1172) 29.6664*** (0.1398) 2.6270*** (0.1220) 0.8326*** (0.1002) NonOECD 1.7944*** (0.0865) *(log of distance) 0.3075*** (0.0155) *(log of distance) 0.0883*** (0.0123) NonOECD *(log of distance) 0.2192*** (0.0106) Log of real GDP 0.9073*** (0.0024) 0.9148*** (0.0025) Log of real GDP per capita 0.1339*** (0.0031) 0.1345*** (0.0035) Regional dummy 1.0560*** (0.0368) 1.3667*** (0.0388) Interregional dummy 0.2911*** (0.0163) 0.2397*** (0.0164) Common currency dummy 0.3399*** (0.0585) 0.3766*** (0.0585) Log of distance 1.0874*** (0.0080) 0.9758*** (0.0100) Common land border dummy 0.4076*** (0.0377) 0.4219*** (0.0382) Common colonizer after 1945 dummy 0.8129*** (0.0214) 0.8420*** (0.0218) Colonial relationship dummy 1.3313*** (0.0384) 1.3197*** (0.0386) Common language dummy 0.4416*** (0.0142) Year dummies Estimated but not reported here 0.4027*** (0.0144) R 2 0.6784 0.6805 Number of observations 122,402 122,402 Specification Test 1. F-Value : 209.79 Pr > F(4,.) : <0.0001 Specification Test 2. F-Value : 255.11 Pr > F(2,.) : <0.0001 In second column equation, two restrictions are applied: Both OECD + One OECD + Non OECD = 0; Both OECD* log of distance + One OECD* log of distance + Non OECD* log of distance = 0. Notes: (1) OLS standard errors are shown in parentheses. Statistical significance at the 1 per cent level is indicated by ***, at the 5 per cent by ** and the 10 per cent by *. (2) Specification Test 1: Additional variables from first column equation of Table 3 jointly equal 0. (3) Specification Test 2: Variables for elasticity jointly equal 0.

INTERNATIONAL TRADE, OECD MEMBERSHIP, AND RELIGION 499 variables can be added with the restriction that the sum of their coefficients is zero. The three options are equivalent in that one implies and is implied by the other two. We select the third way to make the situation symmetrical. The overall intercept term can be directly compared with the original term without the dummy variables. The coefficients of the three dummy variables can be symmetrically interpreted. For instance, the intercept term would have been 32.2934( 29.6664 2.6270) for, 28.8338( 29.6664 + 0.8326) for, and 27.8720( 29.6664 + 1.7944) for NonOECD. The estimates of 2.6270, 0.8326, and 1.7944 indicate how the coefficients of the three dummy variables deviate from the overall intercept term. Alternatively, had we included the three dummy variables without the restriction and without the overall intercept term, those three estimates would have been 32.2934, 28.8338, and 27.8720, respectively. We test two hypotheses. The first is whether the addition of the dummy variables and their interaction terms is statistically significant. The F statistic of 209.79 and its associated p-value of less than 0.0001 indicate that the addition is very significant; see Table 3. The second is whether distance elasticities are the same for the three dummy interactions. The F statistic of 255.11 and its associated p-value of less than 0.0001 reject the same-elasticity null hypothesis. There indeed exist heterogeneous distance elasticities on the basis of OECD memberships. Heterogeneity survives even after we control for real GDP and real per capita GDP; see Table 3. The overall distance elasticity falls in absolute value from 1.0874 to 0.9758 when OECD memberships are added. More importantly, there is clear heterogeneity in the distance elasticities depending on these memberships: 0.6683( 0.9758+0.3075) for, 1.0641 for, and 1.1950 for NonOECD, and the differences are statistically significant. The smallest transaction and transportation costs occur when both countries in the pair are industrial; the largest when both countries are not industrial. The lower transaction costs, it should be remembered, come from simply being an OECD member after controlling for real GDP, real per capita GDP, and other factors such as having common land borders and common languages. The first column in Table 4 shows estimation results in the presence of religion dummy variables. Four dummy variables are added with the restriction that their coefficients add to zero. BothChristian (BothIslam) takes the value of one when both trading partners are Christian (Islamic) countries and zero otherwise. OneChristian and OneIslam dummy variable takes the value of one if the trading partners have different religions and zero otherwise. The fourth dummy variable, NeitherReligion, takes the value one if each of the trading partners is neither Christian nor Islamic and zero otherwise. The coefficient of the dummy variables is the largest for OneChristian and OneIslam trade partners (1.5075) and it is the smallest when both trading partners are Islamic countries ( 2.7234). The overall distance elasticity declines more than in the OECD membership case, in absolute values: from 1.0866 in the basic model to 0.8854 when religion dummy variables are added. The heterogeneity in the distance elasticities is, however, pronounced. The

500 KANG AND FRATIANNI Table 4. Variables OLS estimates with religion and OECD variables. With religion and religion* (log of distance) With OECD, Religion, OECD* (log of distance) and Religion* (log of distance) Intercept 27.8958*** (0.1241) 29.6839*** (0.1511) 3.3592*** (0.1252) 1.0058*** (0.1009) NonOECD 2.3534*** (0.0902) Both OECD *(log of distance) 0.4011*** (0.0158) *(log of distance) NonOECD *(log of distance) BothChristian 0.9001*** (0.0935) BothIslam OneChristian OneIslam 2.7234*** (0.1623) 1.5075*** (0.1154) Neither Religion 0.3158*** (0.1193) BothChristian *(log of distance) 0.1305*** (0.0116) BothIslam *(log of distance) 0.3290*** (0.0210) OneChristian OneIslam* (log of distance) Neither religion *(log of distance) 0.2203*** (0.0142) 0.0218*** (0.0144) Log of real GDP 0.8894*** (0.0024) Log of real GDP per capita 0.1608*** (0.0031) Regional dummy 0.9539*** (0.0377) Interregional dummy 0.3546*** (0.0165) 0.1055*** (0.0123) 0.2957*** (0.0111) 2.1964*** (0.0996) 3.2773*** (0.1634) 1.2535*** (0.1174) 0.1726*** (0.1198) 0.2896*** (0.0123) 0.4026*** (0.0211) 0.1913*** (0.0144) 0.0784*** (0.0144) 0.8949*** (0.0025) 0.1523*** (0.0035) 1.2182*** (0.0388) 0.2922*** (0.0166) (Continued on next page.)

INTERNATIONAL TRADE, OECD MEMBERSHIP, AND RELIGION 501 Table 4. (Continued). Variables With Religion and Religion*(log of distance) With OECD, Religion, OECD*(log of distance) and Religion*(log of distance) Common currency dummy 0.3204*** (0.0576) Log of distance 1.0866*** (0.0098) Common land border dummy 0.4367*** (0.0377) Common colonizer after 1945 dummy 0.6939*** (0.0214) Colonial relationship dummy 1.3974*** (0.0379) Common language dummy 0.4379*** (0.0141) Year dummies Estimated but not reported here 0.3807*** (0.0575) 0.8854*** (0.0121) 0.4369*** (0.0379) 0.7671*** (0.0216) 1.3902*** (0.0379) 0.3895*** (0.0143) R 2 0.6893 0.6927 Specification Test 1. F-Value: 717.41 Pr > F(6,.): <0.0001 Specification Test 2. F-Value: 128.50 Pr > F(3,.): <0.0001 F-Value: 572.51 Pr > F(10,.) : <0.0001 F-Value: 255.32 Pr > F(5,.) : <0.0001 In first column equation, two restrictions are applied: Both Christianity + Both Islam + One Christianity one Islam + Neither Religion = 0; Both Christianity * log of distance + Both Islam * log of distance + One Christianity one islam * log of distance + Neither religion * log of distance = 0. In second column equation, four restrictions are applied: Both OECD + + Non OECD = 0; Both Christianity + Both Islam + One Christianity One Islam + Neither Religion = 0; Both OECD * log of distance + One OECD* log of distance + Non OECD * log of distance = 0; Both Christianity * log of distance + Both Islam * log of distance + One Christianity one Islam * log of distance + neither religion * log of distance = 0. See notes to Table 3. deviation of the distance elasticity for each dummy variable ranges from 0.2203 for OneChristianOneIslam to 0.3290 for BothIslam. Trading costs, measured in terms of distance elasticities, are largest for OneChristianOneIslam with 1.3069( 1.0866 0.2203) and smallest for BothIslam with 0.7576. The other two groups, BothChristian and NeitherReligion, deviate from the overall elasticity by 0.1305 and 0.0218, respectively. Trading costs are largest when one trading partner is Christian country and the other is Islamic. As was the case with OECD membership, these changes take place after we control for real GDP, real per capita GDP, and other factors. The test that religion does not play a role in international trade is rejected with an F value of 717.41 and a p value of less than 0.0001. The null that all the distance elasticities

502 KANG AND FRATIANNI are the same regardless of religion is also rejected with an F value of 128.50 and a p-value of less than 0.0001. In sum, religion, or some factors closely related to religion, does exert a statistically significant impact on international trade. Finally, we combine both OECD and religion dummy variables; see the last column of Table 4. The R 2 value of 0.6927 shows a significant statistical improvement from the R 2 value of 0.6784 of the base model. The hypothesis that both OECD and religion dummy variables are irrelevant for trade flows is rejected with an F-value of 572.51 and a p-value of less than 0.0001. The hypothesis that all OECD distance elasticities and all religion distance elasticities are the same is also rejected with an F-value of 255.32 and a p-value of less than 0.0001. The deviations of OECD and religion dummy variables from the overall intercept term are qualitatively the same when the dummy variables for each group are added separately. The same pattern is true for the distance elasticities. OECD memberships and religion of trading partners are therefore separately important in international trade. That is, the results in the last column of Table 4 with both OECD and religion variables are essentially the exact combination of the second column of Table 3 with OECD variables only and the first column of Table 4 with religion variables only. 2. Economic significance of OECD membership and religion The statistically significant heterogeneous distance elasticities strongly suggest that OECD memberships and religion are relevant in predicting bilateral trade flows. The next question is whether economic significance comes along with statistical relevance. For that we proceed as follows. We use the estimated results to compute the values of expected trade flows for twelve possible configurations: three from different OECD membership groups and four from different religious groups. These expected values are then compared to infer the economic significance of heterogeneous distance elasticities arising from OECD membership and religion. The results are summarized in Table 5; to make the comparison simpler the columns of Table 5 are ranked in terms of the value of expected trade flows, with columns on the left having higher values than columns on the right. The first column of Table 5 shows the results of combining with NeitherReligion; this is our reference group and represents 10.03% of the entire sample. Other than their relevant dummy variables for OECD memberships and for countries religion, the expected trade flows are computed at the sample mean values of other variables in the model. In the computation of the expected trade flows, we have used the parameter estimates from the last column in Table 4 with and religion dummy variables and their interaction terms with distance, as shown by the top portion of Table 5. Using the sample mean values of Table 1, the expected log real trade flows is 8.5676, or $5.258 million.

INTERNATIONAL TRADE, OECD MEMBERSHIP, AND RELIGION 503 Table 5A. Economic significance OECD and religion effect on bilateral trade flow. Variabl e and neither religion (A) and neither religion NonOECD and neither religion Islam Islam Christian Intercept 29.6839 29.6839 29.6839 29.6839 29.6839 29.6839 3.3592 3.3592 1.0058 1.0058 1.0058 NonOECD 2.3534 * * NonOECD* 3.2964 3.2964 0.8671 0.8671 0.8671 2.4302 BothChristian 2.1964 BothIslam 3.2773 3.2773 OneChristian OneIslam Neither religion 0.1726 0.1726 0.1726 Both Christian* 2.3801 BothIslam* One Chr. OneIsl.* Neither religion* 0.6443 0.6443 0.6443 3.3088 3.3088 (Continued on next page.)

504 KANG AND FRATIANNI Table 5A. (Continued). Variable and neither religion (A) and neither religion NonOECD and neither religion Islam Islam Christian Log of real GDP 42.6065 42.6065 42.6065 42.6065 42.6065 42.6065 Log of real GDP 2.3112 2.3112 2.3112 2.3112 2.3112 2.3112 per capita Log of distance 7.2767 7.2767 7.2767 7.2767 7.2767 7.2767 Expected Log of bilateral trade flow Net changes in log of bilateral trade flow of (A) 8.5676 8.3661 8.3520 8.1273 7.9258 7.9122 0.2015 ( 2.4%) 0.2156 ( 2.5%) 0.4403 ( 5.1%) 0.6418 ( 7.5%) 0.6554 ( 7.7%) Expected bilateral trade flow Net changes in expected bilateral trade flow of (A) 5,258 4,299 4,239 3,386 2,768 2,730 960 ( 18.3%) 1,020 ( 19.4%) 1,873 ( 35.6%) 2,491 ( 47.4%) 2,528 ( 48.1%) Net change from mean of real bilateral flow ( 0.45%) ( 0.47%) ( 0.87%) ( 1.16%) ( 1.17%) Case frequency in the sample (Rank out of 12 cases) 10.03% (6) 0.61% (9) 16.70% (2) 0.55% (10) 0.00% (12) 22.48% (1)

INTERNATIONAL TRADE, OECD MEMBERSHIP, AND RELIGION 505 Table 5B. Economic significance OECD and religion effect on bilateral trade flow. Variable NonOECD and both Islam and one Christian one Islam Christian NonOECD Christian and one Christian one Islam NonOECD and one Christian one Islam Intercept 29.6839 29.6839 29.6839 29.6839 29.6839 29.6839 3.3592 3.3592 1.0058 Non OECD 2.3534 2.3534 2.3534 * One OECD* Non OECD* 0.8671 3.2964 3.2964 2.4302 2.4302 2.4302 Both Christian 2.1964 2.1964 Both Islam 3.2773 One Christian 1.2535 1.2535 1.2535 One Islam Neither Religion Both Christian * Both Islam * One Chr. OneIsl. * 3.3088 2.3801 2.3801 1.5722 1.5722 1.5722 Neither religion * Log of real GDP 42.6065 42.6065 42.6065 42.6065 42.6065 42.6065 Log of real GDP 2.3112 2.3112 2.3112 2.3112 2.3112 2.3112 per capita Log of distance 7.2767 7.2767 7.2767 7.2767 7.2767 7.2767 (Continude on next page.)

506 KANG AND FRATIANNI Table 5B. (Continued). Variable NonOECD and both Islam and one Christian one Islam Christian NonOECD Christian and one Christian one Islam NonOECD and one Christian one Islam Expected log of bilateral trade flow Net Changes in log of bilateral trade flow of (A) 7.9118 7.7771 7.7107 7.6966 7.5756 7.5616 0.6558 ( 7.7%) 0.7904 ( 9.2%) 0.8569 ( 10.0%) 0.8710 ( 10.2%) 0.9919 ( 11.6%) 1.0060 ( 11.7%) Expected bilateral trade flow Net changes in expected bilateral trade flow of (A) 2,729 2,385 2,232 2,201 1,950 1,923 2,529 ( 48.1%) 2,873 ( 54.6%) 3,026 ( 57.6%) 3,057 ( 58.2%) 3,308 ( 62.9%) 3,335 ( 63.4%) Net change from mean of real bilateral flow ( 1.17%) ( 1.33%) ( 1.40%) ( 1.42%) ( 1.53%) ( 1.55%) Case frequency in the sample (Rank out of 12 cases) 4.96% (7) 13.50% (3) 4.40% (8) 13.47% (4) 0.42% (11) 12.88% (5) Notes: (1) The values are calculated based on the overall mean value of each variable except OECD and religion dummy and their interaction terms with log of distance. The overall mean values are following: log of GDP = 47.6103; log of GDP per capita = 15.1753; log of distance = 8.2185. (2) One OECD and Both Christian religion is the most common case in our sample (see Table 2). (3) Mean of real bilateral trade flow is 215,573.

INTERNATIONAL TRADE, OECD MEMBERSHIP, AND RELIGION 507 The second column of Table 5A shows the combination of and Neither Religion, 0.61% of the entire sample. Its expected log trade flow is 8.3661, 2.4% lower than the expected value of the reference group. But in actual dollars, the expected trade is $4.299 million against $5.258 million of the reference group, a decline of 18.3%. For the combination NonOECD and NeitherReligion (16.7% of the sample), the expected trade value falls by 19.4% (from $5.258 million to $4,239 million) with respect to the reference group and by 1.4 % (from $4.299 million to $4.239) with respect to the combination and NeitherReligion. The 1.4% drop in expected value of trade can be interpreted as the marginal transaction cost of both countries in the pair of not being industrialized. The combination of and BothChristian (4.4% of the sample) yields an expected trade value of $2.232 million, a 57.6% drop relative to the reference group. The transaction cost of being Christian rises by 48.1% (from $4.299 million to $2.232 million) relative to the alternative of being neither Christian nor Islamic. We can compute the marginal transaction cost of being Islamic relative to the alternative of being neither Christian nor Islamic by comparing the combination of NonOECD and NeitherReligion with the combination of NonOECD and BothIslam: expected trade value declines by 35.6% (from $4.239 million to $2.729 million). Being Christian implies higher transaction costs than being Islamic. Finally, we can compute the marginal cost of two countries having different religions relative to the alternative of sharing the same religion. The value of this marginal cost depends on whether the benchmark is a pair of countries sharing Islam or Christianity. The marginal cost of having different religions relative to the alternative of sharing Islam is 29.5% (from $2.729 million to $1.923 million); the marginal cost is 12.6% (from $2.201 million to $1.923 million) if the alternative is sharing Christianity. These costs were obtained by comparing the expected value of trade of the combination NonOECD and OneChristianOneIslam with the combination NonOECD and BothIslam and the combination NonOECD and BothChristian, respectively. The overall conclusion is that industrialization lowers transaction costs, while religion raises them. Both Christianity and Islam impose costs relative to the alternative of no religion, with Islam having an edge over Christianity. Transaction costs rise if countries have different religions relative to the alternative of sharing the same religion. 3. Conclusions As commonly interpreted in the use of the trade equation, bilateral distance represents trading costs. Controlling for real GDP, per capita real GDP, and other characteristics, researchers have assumed that the distance elasticity is the same across all trading partners: the estimated value of this elasticity, 1.0874, is almost universal in the literature. But we have found that

508 KANG AND FRATIANNI distance elasticity is not homogeneous across countries. Both OECD membership and country religion are important determinants of bilateral flows. It was also found that the interaction of those two variables with distance is statistically significant. Expected bilateral flows are largest when an OECD member trades with a non-member and when both countries are not religious in the specific sense of being neither Christian nor Islamic. Pairs of religious industrial countries trade more than pairs of non-religious industrial countries. Religion raises trading costs. The lowest expected value of trade occurs when countries are not industrialized and one country is Christian while the other is Islamic. Transaction costs, measured in terms of the distance elasticity, are smaller for Islamic countries than for Christian countries, regardless of whether they belong to the OECD. Trading costs rise when two countries have different religions relative to the alternative of sharing the same religion. All of these results were obtained after having controlled for real GDP, real per capita income, and other factors such as RTAs. The single most important conclusion of this paper is that diversity of religions, like diversity of cultures, may create differential trading costs. This conclusion aligns with the works of Weber (1930), Landes (1998, pp. 175 181), and Huntington (1993), who have discussed religion as a determinant of economic growth and economic development. Additional studies need to be undertaken to uncover the reasons underlying heterogeneous trading costs. References Anderson, James E. (1979) A Theoretical Foundation for the Gravity Equation. American Economic Review 69:106 115. CIA Factbook, http://www.cia.gov/cia/publications/factbook. Deardorff, Alan V. (1998) Determinants of Bilateral Trade: Does Gravity Work in a Neoclassical World? In Jeffrey A. Frankel (ed.), The Regionalization of the World Economy. Chicago: University of Chicago Press (for NBER). Frankel, Jeffrey A. (1997) Regional Trading Blocs in the World Economic System. Institute for International Economics Washington, DC. Fratianni, Michele and Heejoon Kang (2006) Heterogeneous Distance-Elasticities in Trade Gravity Models. Economics Letters pp. 68 71. Huntington, Samuel P. (1993) The Clash of Civilizations. Foreign Affairs 72:22 50. Landes, David S. (1998) The Wealth and Poverty of Nations: Why Some Countries are so Rich and Some so Poor. W.W. Norton. Rose, Andrew K. (2000) One Money, One Market: The Effects of Common Currency on Trade. Economic Policy 30:9 45. Rose, Andrew K. (2003) Which International Institutions Promote International Trade? Manuscript. University of California at Berkeley. Weber, Max (1930) The Protestant Ethic and the Spirit of Capitalism. Allen & Unwin. World Christian Database, http://www.worldchristiandatabase.org/ecd.