THE LOG OF GRAVITY - the University of Mauritius

THE LOG OF GRAVITY 651TABLE 4.—RESULTS OF THE TESTS FOR TYPE OF HETEROSKEDASTICITY(p-VALUES)Test (Null Hypothesis) Exports 0 Full SampleGNR (V[y i x] (x i )) 0.144 0.115Park (OLS is valid) 0.000 0.00027The formula to compute this effect is (e bi 1) 100%, where b i isthe estimated coefficient.28To illustrate the role of remoteness, consider two pairs of countries, (i,j) and (k, l), and assume that the distance between the countries in eachpair is the same (D ij D kl ), but i and j are closer to other countries. In thiscase, the most remote countries, k and l, will tend to trade more betweeneach other because they do not have alternative trading partners. SeeDeardoff (1998).The role of geographical distance as trade deterrent issignificantly larger under OLS; the estimated elasticity is1.17 (s.e. 0.03), whereas the Poisson estimate is 0.78(s.e. 0.06). This lower estimate suggests a smaller role fortransport costs in the determination of trade patterns. Furthermore,Poisson estimates indicate that, after controllingfor bilateral distance, sharing a border does not influencetrade flows, whereas OLS, instead, generates a substantialeffect: It predicts that trade between two contiguous countriesis 37% larger than trade between countries that do notshare a border. 27We control for remoteness to allow for the hypothesis thatlarger distances to all other countries might increase bilateraltrade between two countries. 28 Poisson regressionssupport this hypothesis, whereas OLS estimates suggest thatonly exporter’s remoteness increases bilateral flows betweentwo given countries. Access to water appears to beimportant for trade flows, according to Poisson regressions;the negative coefficients on the land-locked dummies can beinterpreted as an indication that ocean transportation issignificantly cheaper. In contrast, OLS results suggest thatwhether or not the exporter is landlocked does not influencetrade flows, whereas a landlocked importer experienceslower trade. (These asymmetries in the effects of remotenessand access to water for importers and exporters arehard to interpret.) We also explore the role of colonialheritage, obtaining, as before, significant discrepancies:Poisson regressions indicate that colonial ties play no role indetermining trade flows, once a dummy variable for commonlanguage is introduced. OLS regressions, instead, generatea sizeable effect (countries with a common colonialpast trade almost 45% more than other pairs). Language isstatistically and economically significant under both estimationprocedures.Strikingly, in the traditional gravity equation, preferentialtradeagreements play a much smaller—although still substantial—roleaccording to Poisson regressions. OLS estimatessuggest that preferential trade agreements raiseexpected bilateral trade by 63%, whereas Poisson estimatesindicate an average enhancement below 20%.Preferential trade agreements might also cause trade diversion;if this is the case, the coefficient on the tradeagreementdummy will not reflect the net effect of tradeagreements. To account for the possibility of diversion, weinclude an additional dummy, openness, similar to that usedby Frankel (1997). This dummy takes the value 1 wheneverone (or both) of the countries in the pair is part of apreferential trade agreement, and thus it captures the extentof trade between members and nonmembers of a preferentialtrade agreement. The sum of the coefficients on the tradeagreement and the openness dummies gives the net creationeffect of trade agreements. OLS suggests that trade destructioncomes from trade agreements. Still, the net creationeffect is around 40%. In contrast, Poisson regressions provideno significant evidence of trade diversion, although thepoint estimates are of the same order of magnitude underboth methods.Hence, even when allowing for trade diversion effects, onaverage, the Poisson method estimates a smaller effect ofpreferential trade agreements on trade, approximately halfof that indicated by OLS. The contrast in estimates suggeststhat the biases generated by standard regressions can besubstantial, leading to misleading inferences and, perhaps,erroneous policy decisions.We now turn briefly to the results of the other estimationmethods. OLS on ln(1 T ij ) and tobit give very closeestimates for most coefficients. Like OLS, they yield largeestimates for the elasticity of bilateral trade with respect todistance. Unlike OLS, however, they produce insignificantcoefficients for the contiguity dummy. They both generateextremely large and statistically significant coefficients forthe trade-agreement dummy. The first method predicts thattrade between two countries that have signed a trade agreementis on average 266% larger than that between countrieswithout an agreement. The second predicts that trade betweencountries in such agreements is on average 100%larger. NLS tends to generate somewhat different estimates.The elasticity of trade with respect to the exporter’s GDP issignificantly smaller than with OLS, but the correspondingelasticity with respect to importer’s GDP is significantlylarger than with OLS. The estimated distance elasticity issmaller than with OLS and bigger than with Poisson. Likethe other methods, NLS predicts a significant and largeeffect for free-trade agreements.It is noteworthy that all methods, except the PPML, leadto puzzling asymmetries in the elasticities with respect toimporter and exporter characteristics (especially remotenessand access to water).To check the adequacy of the estimated models, weperformed a heteroskedasticity-robust RESET test (Ramsey,1969). This is essentially a test for the correct specificationof the conditional expectation, which is performed bychecking the significance of an additional regressor constructedas (xb) 2 , where b denotes the vector of estimatedparameters. The corresponding p-values are reported at thebottom of table 3. In the OLS regression, the test rejects thehypothesis that the coefficient on the test variable is 0. This