PLEASE SCROLL DOWN FOR ARTICLE
This article was downloaded by: [National Kaohsiung University]
On: 11 May 2010
Access details: Access Details: [subscription number 788880709]
Publisher Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
Applied Economics
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713684000
Estimating the staged effects of regional economic integration on trade
volumes
I. -H. Cheng a;Y. -Y. Tsai a
a Department of Applied Economics, National University of Kaohsiung, Kaohsiung 811, Taiwan, ROC
Online publication date: 12 May 2007
To cite this Article Cheng, I. -H. andTsai, Y. -Y.(2008) 'Estimating the staged effects of regional economic integration on trade volumes', Applied Economics, 40: 3, 383 — 393
To link to this Article: DOI: 10.1080/00036840600606252 URL: http://dx.doi.org/10.1080/00036840600606252
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial or systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.
Estimating the staged effects of
regional economic integration on
trade volumes
I.-H. Cheng* and Y.-Y. Tsai
Department of Applied Economics, National University of Kaohsiung, 700 Kaohsiung University Road, Nan-Tzu, Kaohsiung 811, Taiwan, ROC
This article compares various specifications of the gravity model of trade, which are commonly used in the literature. It shows that most of them can be regarded as nested versions of a general specification in which the heterogeneity of bilateral country-pair fixed effects is accounted for. This article constructs a modified gravity model in which both the conventional gravity variables and price-effect variable are included. It intends to discuss the trend of the effects of the enlargement and enhancement of regional blocs, particularly taking account of the staged effects to investigate the effects of regional economic agreements on trade volumes. The differences of the effects across regional blocs are investigated.
I. Introduction
Testing the effects on trade of regional economic integration (REI) has attracted great attention in the wake of the growing preferential trading agreements over the past decade.1 Much of the early empirical studies employed the residual imputation models to compare the ex-post income elasticity of demand for all commodities between intra-area and extra-area trade, and, thus, provided explanations for the trade effects of REIs.2 An alternative approach generally
used is estimating trade flows by a dummy variable, which identifies REI partners with standard gravity-type models.3 More specific, while examining the effects on trade, we may construct a REI dummy variable that takes on the value of one if both trading partners interact in the same trading bloc and zero otherwise.
In recent years, the gravity model of international trade flows has been used widely as a baseline model for estimating the impact of a variety of policy issues, including regional trading groups, currency unions,
*Corresponding author. E-mail: icheng@nuk.edu.tw
1
See Bhalla and Bhalla (1997) and Whalley (1998) for a comprehensive survey.
2
For instance, a residual imputation type of model simply compares what actually happens with what would have happened in the absence of economic integration for a post-integration year, and attributes the difference between the actual and ‘constructed’ imports to integration. See Mayes (1978) for a survey of earlier empirical results.
3
The gravity model of international trade was developed independently by Tinbergen (1962) and Po¨yho¨nen (1963). In its basic form, the amount of trade between two countries is assumed to be increasing in their sizes, as measured by their national incomes, and decreasing in the cost of transport between them, as measured by the distance between their economic centres. Following this work, Linnemann (1966) included population as an additional measure of country size. This dummy approach to estimate the REI effects was introduced by Linnemann (1966).
Applied EconomicsISSN 0003–6846 print/ISSN 1466–4283 onlineß 2008 Taylor & Francis 383 http://www.tandf.co.uk/journals
DOI: 10.1080/00036840600606252
political blocs, international migration, patent rights and various trade distortions.4 These events and policies are essentially modelled as deviations from the volume of trade predicted by the baseline gravity model, and are captured by dummy variables, in particular, in the case of regional integration. Although the use of the gravity model has been perceived as an empirical success, more analyses on its econometric properties remain to be explored since its empirical power is simply stated on the basis of goodness of fit.5
This article studies the implications of different empirical approaches undertaken for various estimation results. We conduct alternative empirical estimation, namely the standard gravity-type and heterogeneous trading pair models (HTP), on bilat-eral trade flows. We show that standard methods of gravity-type may over-estimate the elasticity of bilateral trade with respect to national incomes and, thus, probably yield different signs for the influence of the two countries’ populations. We then construct a modified gravity model that incorporates both conventional gravity variables and a price-effect variable, using the modified REI dummies distin-guished by timing, i.e. it treats a single regional grouping which has experienced different degrees of integration as different schemes of REI. The greatest benefits of using those dummies are that by doing so it is able to identify the staged effects of regional economic agreements on trade volumes.
The article is organized as follows. Section II sets out the standard and modified gravity models and the tests which we may be interested in. Section III contains the typical pooled cross-section (PCS) and the modified gravity models, and provides the empirical analysis. Concluding remarks are provided in Section IV.
II. The Statistical Framework
We address in this section the issue of estimation for international bilateral trade flows. Based upon Cheng and Wall (2005), we introduce a HTP model (henceforth) to estimate the staged effects of regional economic agreements. Before proceeding into the HTP model, we shall briefly discuss implications for
estimation of bilateral trade flows of various forms of gravity models.
Essentially, the regression model set out to quantity trade flows is one taking the cross-section form for time t, namely
xijt¼tþb0tzijtþ"ijt, t ¼1, . . . , T ð1Þ
where xijtis the observation on export from country i
to country j in year t, z0
ijt¼ ½zitzjt. . . is the 1 k row
vector of gravity variables and the dummy variables we use when testing the effects of membership in a trading bloc, and "ijt the disturbance term be
normally distributed with zero mean and constant variance for all observations, i.e. "ijtINð0, t2Þ.
We also assume that the disturbances are pairwise uncorrelated, i.e. Eð"ijt, "ij0tÞ ¼0 and Eð"ijt, "ijt1Þ ¼0.
At time t, btis the k 1 regression coefficient vector
excluding the constant term (t), that is, b0t¼
½1t2t kt. Should the classical disturbance-term
assumptions retain, we could then estimate separate cross-section regressions by the ordinary least squares (OLS) procedure for individual time periods.
Clearly, pooling cross-section and time-series data provides increased number of data points, and the use of panel data can allow researchers to classify economic effects that could not be identified in either cross-section or time-series setting alone. However, the major problem of the use of panel data is to decide on questions like whether pooling is appropriate and to what degree one may pool the data. A simple way to deal with the problem of structural change would be to allow for different period-specific intercepts in the regression equations, while stacking only the gravity variables and the other dummies in the restricted equation, i.e. the parameter vector is the same for all t, bt¼b. Thus, the
regression model would be
xijt¼tþb0zijtþ"ijt, t ¼1, . . . , T ð2Þ
where we assume b0¼ ½
1t2 k and "ijt
INð0, 2Þ.
In the light of rapid changes in the size of
economies and population, Bayoumi and
Eichengreen (1997) take an alternative approach of differencing procedure while studying the trade patterns. Suppose that the bilateral relationship between a trading pair remains unchanged over 4
See Aitken (1973), Mayes (1978) and Nilsson (2002) for analysis of integration schemes in western Europe, Bayoumi and Eichengreen (1997), Frankel et al. (1995, 1998), Egger (2004), and Endoh (1999, 2005) for different regional blocs and alternative trade agreements; Bergstrand (1985, 1989) for imperfect competition; Bikker (1987) for substitution between trade flows; Thursby and Thursby (1987), and Frankel and Wei (1998) for currency volatility; Mansfield and Bronson (1997) for political-military alliances; Karemera et al. (2000) for international migration to North America.
5
See Sanso et al. (1993) for an examination of the predictive power of various specifications of the augmented gravity model. Also see Oguledo and MacPhee (1994) for a survey of pre-1990 empirical results.
time, they explain international trade flows by relating to changes in the gravity-type of variables the changes in the volume of trade over time. The first-difference form of gravity model is, therefore, given by
xijt¼tþd0zijtþijt, t ¼2, . . . , T ð3Þ
where xijt¼xijtxijt1, z0ijt¼ ½zitzjt. . . ¼
½zitzit1zjtzjt1. . ., and the parameter vector is
the same for all t, i.e. d1¼d2¼ ¼dt ¼d.
Even though data pooling may be appropriate in studying trade flows, the fact that error term variances in the regression equation are unequal across observations suggests a difficulty in handling the cross-sectional data. In order to focus on the effects of heteroscedasticity, we assume a serial uncorrelated error process. Hence, a more general form of the gravity model with pairwise heterogeneity is
xijt¼tþijþb0zijtþ"ijt, t ¼1, . . . , T ð4Þ
where tis the period-specific effect for time t, and ij
which is taken to be constant over time is the specific country-pair effect between the trading partners. Equation 4 is a two-way fixed-effects model in which the independent variables are assumed to be correlated with ij. In fact, this model is a classical
regression model, which can be estimated using OLS to provide consistent (when N ! 1) and efficient coefficient estimates of the slopes (See Baltagi, 1997). It becomes clear that, in Section III, we enrich this model by considering two scenarios, i.e. ij6¼ji
(the direction of trade flows matters) and ij¼ji
(the direction of trade does not matter), respectively. Another type of the gravity model with trader’s heterogeneity, primarily suggested by Po¨yho¨nen (1963) and formally developed by Ma´tya´s (1997), is a three-way fixed-effects model of the form
xijt ¼tþiþjþb0zijtþ"ijt, t ¼1, . . . , T ð5Þ
where iand jare the effects of the exporting and the
destination countries, respectively, and the other variables and parameters are defined above. In model (5) the independent variables are assumed to be correlated with i and j, and as usual the error
process is assumed to be serially uncorrelated. Thus, this model is also a classical regression model, which can be estimated using OLS. Note that in this model iis only consistent as T ! 1 not N ! 1.
Figure 1 summarizes the gravity-type models set out above and stipulates the tests in which one may be interested. It is clear that the general models can be divided into three groups, each with the arrow sign indicating the hypothesis to be tested. Before formally presenting the estimation results, we remark three
important messages emerge from Fig. 1. First, the general case where we allow for different intercepts and coefficients is model (1). One may consider a hypothesis of a constant b but different ’s, i.e. H1.
In this case, the appropriate procedure is to estimate the regression equation with and without restrictions, models (2) and (1), respectively, then apply the likelihood-ratio test. Next, one may want to test whether the intercepts as well as the slope coefficients are equal. The null hypothesis (H) can also be tested by using a likelihood ratio test. To do the test, we need to estimate model (1) and its restricted version (Model 1R in Fig. 1).
Second, the second type of the gravity model takes the form of first differencing. In this article we will not focus much attention on model (3), and the results of the comparisons between models (3) and (4) will be provided to readers on request. Another general approach to testing trade patterns is based on the use of trading-pair or country specific dummies, i.e. two types of model (4) and model (5). As I have argued above, one reason why we may adopt this approach is that heteroscedasticity is usually present in cross-section studies. As illustrated in Fig. 1, model (5) is a specific form of model (4) in which the implicit restriction ij¼ji¼iþj is imposed. Also note
that model (2) is a restricted version of model (5) in which i¼j¼0. It is interesting to ask how model
(5) differs from model (4) in the study of bilateral trade. The comparison will be represented below.
III. Data and Estimation Results Data
The central issue in this section is to investigate the trend of the effects of the enlargement and enhance-ment of regional blocs with different models of gravity type, identify which specification is preferred, and further discuss the staged effects of REI. To this aim, we estimate the regression models that have been commonly used in the literature, namely, Equation 2, Equations 4 and 5. The explained variable is the values of bilateral trade. The explaining variables are real GDP, population, the great circle distance between geographical centres of two countries, adjacency dummy, and language dummy. In each model, we also introduce a variable characterizing the real exchange rate to account for the price effect on trade volumes.
To highlight the effects of regional blocs, we introduce for each regional bloc a modified regional dummy variable by treating one regional grouping
with varying levels of intended integration (which characterizes the different schemes of regional inte-gration agreements). More specific, we employ a number of pairs of REI dummy variables, including
the European Economic Community (EEC),
European Free Trade Agreement (EFTA),
European Union (EU), Canada-US Free Trade
Agreement (CUSFTA), North American FTA
(NAFTA), Latin American FTA (LAFTA), and the South American Common Market (MERCUSOR). Descriptions of the data and their sources are provided in Table 1.
The data set we have constructed is a balanced panel with 30 107 observations for countries at different levels of economic development and perfor-mance over the period 1981–1997 (1771 country-pairs
The general models
GROUP A Model (1) All coefficients variable
xijt= at + bb′t zijt+eijt
GROUP B Model (3) First-difference equation
∆xijt= gt+ d′ ∆zijt+ uijt
GROUP C Model (4)
Pairwise and period intercepts
xijt= at + aij + b′zijt+ eijt
H2:aij= ai+ aj
Model (2) Slope coefficients constant,
intercept variable.
xijt=at + b′zijt+eijt
Model (5) Trader and period intercepts
xijt= at + ai+aj + b′zijt+ eijt
H:at = a,bt = b b and st2 =s 2.
H1:bt= b b and s t2=s2
Model (1R) All coefficients constant.
xijt= a + b′zijt+ eijt
Fig. 1. Alternative gravity models and hypotheses
Table 1. Variables and data sources
Real Exports (Xijt), measured in millions of US dollars, from the IMF World Trade Flows, 1980–1997
Real Gross Domestic Product (Yijt), in millions of US dollars at market prices, from the World Bank’s World Development
Indicators (WDI) 2003 CD-ROM
Population (Nijt), in thousands of inhabitants, from the World Bank’s WDI 2003 CD-ROM
Real Exchange Rate (XRijt), the PPP conversion factor to official exchange rate ratio, from the World Bank’s WDI 2003
CD-ROM
Distance (Dij), expressed in kilometers, is the great circle distance between geographic centres, from the CIA’s The World
Factbook2002
Contiguity Dummy (Cij), set to be one if two trading partners share a border, from the CIA’s The World Factbook 2002
Language Dummy (LANGij), set to be one if two trading partners share common language, from the CIA’s The World
Factbook2002
Regional Economic Integration (REI) Dummies, set to be 1 if a country joins a REI. For instance, EEC2ijtis the dummy
variable standing for countries i and j in the EEC in time t. Dummy variables EECXijt(EECMijt) are set to be one when
exporting country i (importing country j) is a EEC member in year t
in each year). We included observations of nonzero trade between countries listed in all of the relevant World Bank World Development Reports during these years, but excluded countries that were identi-fied as high-income oil exporters. Neither did we include any observations without countries’ price indexes and/or exchange rates, since we take the price effect into account in this study. The result is a manageable data set that is fairly representative of the literature. Table 2 demonstrates the country list of 44 exporting and 57 importing countries in the data set. Note that for each country the number in the brackets denotes the number of its export destination countries.
Models
For simplicity, we restrict our attention to the comparison between estimation over alternative forms of gravity models, and, we focus, in particular, on the comparison between models (2) and (4). To explain the effects of membership in a common regional bloc on bilateral trading patterns, the typical gravity model (model 2) can be specified as:
ln Xijt¼tþ1ln Yitþ2ln Yjtþ3ln Nit
þ4ln Njtþ5ln XRijtþ6ln Dij
þ7Cijþ8LANGijþ1REI2ijtþ2REIXijt
þ3REIMijtþ"ijt ðPCSÞ
where Xijt is the value of all exports (in $US) from
country i to country j; Yit and Yjt are the two
countries’ GDPs in $US; Nit and Njt are the
two countries’ populations; XRijtis the PPP
conver-sion factor to official exchange rate ratio, Dij is the
distance between their economic centres (their capital cities); Cijis a contiguity dummy that is set to be one
when two countries are contiguous and zero other-wise; LANGijis set to be one when people in the two
countries share common language and zero other-wise; and t denotes the year-specific component of
the intercept. REI2ijtare the dummy variables I use
when capturing the effects caused by REIs, standing for both trading partners in the same regional blocs in time t. Thus, REI2ijtis the dummy variable measuring
the intra-REI effect. In contrast, dummy variables REIXijt(REIMijt) are used to estimate the extra-REI
effects, set to be one when exporting country i (importing country j) is a REI member in year t.6
Table 2. List of 44 exporting and 57 importing countries
Argentina (38) Honduras (0)a Panama (0)a Australia (52) Hong Kong (51) Paraguay (14) Austria (55) Iceland (0)a Peru (29) Brazil (53) India (37) Philippines (37) Cameroon (0)a Indonesia (24) Portugal (43) Canada (56) Ireland (40) Singapore (44) Chile (24) Italy (56) South Africa (26) China (44) Jamaica (12) Spain (56) Colombia (32) Japan (56) Sweden (55) Costa Rica (0)a Kenya (23) Switzerland (55) Cote d’Voire (0)a Korean Republic (48) Thailand (31) Denmark (56) Malaysia (41) Tunisia (0)a Ecuador (0)a Mexico (32) Turkey (18)
Egypt (15) Morocco (32) UK (56) El Salvador (0)a The Netherlands (56) US (56)
Finland (52) New Zealand (36) Uruguay (31) France (56) Nicaragua (0)a Venezuela (0)a Germany (56) Nigeria (13) Zambia (0)a Greece (28) Norway (46) Zimbabwe (0)a Notes: The data set for the estimation of 1771 trading country pairs includes 44 exporting and 57 importing countries over the period 1981–1997, for a total of 30 107 observations. For each country, the number in the brackets denotes the number of its export destinations in our data set.
a
Denotes a country that is a trade destination but not trade origin country in the data set.
6
Endoh (2005) uses similar approach to identify the trade creation and diversion effects of the Global System of Trade Preferences among developing countries. Note that REI2ijtand REIXijt(REIMijt) are respectively defined as ‘intrabloc bias’
and ‘extrabloc openness’ in Frankel and Wei (1998). However, the way of adding REI dummies into regression equation is different from that of Frankel and Wei (1998) in which they include the same set of REI dummies in every year.
Alternatively, we estimate Equation 4 by consider-ing the case where the direction of trade does matter in international trade flows, i.e. ij6¼ji, as:
ln Xijt¼tþijþ1ln Yitþ2ln Yjtþ3ln Nit
þ4ln Njtþ5ln XRijtþ1REI2ijt
þ2REIXijtþ3REIMijtþ"ijt ðHTPÞ
where ijis the specific ‘country-pair’ effect between
the trade partners; and the other variables are defined in Equation PCS. ijand tcan, respectively, account
for the effects of those omitted variables that are cross-sectional specific but remain constant over time, and the effects that are specific to each time-period but are identical for all countries. Since this model controls for the trading-pair heterogeneity, we call it the HTP model hereafter.
Results of the coefficient estimates on gravity variables
In the following, we restrict our attention to the comparison between model (PCS) and model (HTP). The results of estimating the coefficients on gravity variables of the PCS model are reported in the first column of Table 3.7 As we have seen, all gravity variables are highly significant statistically. Estimates of this augmented simple-pooling gravity model predict that: (i) a 10% rise in exporter’s GDP should be associated with a 11.1% rise in exports, all else constant; (ii) if an importer’s GDP rises by 10%, exports should rise by 10.5%, holding exporter’s GDP constant; (iii) exports tend to be capital-intensive and exports tend to be income-elastic (luxury goods), as indicated by the usual negative signs on the population coefficients; (iv) when exporter’s real exchange rate depreciates 10% measured in the real term, it should be associated with a 2.8% rise in exports, all else constant; (v) an exporting country will send 80.8% less of exports to a market that is twice as distant as another otherwise-identical market; and (vi) exports would be higher by 36.1% than the other trade flows if trading partners share common border, while exports would be higher by 58.3% if they use the same language.
The second column of Table 3 reports the estimation results for the HTP model in which we assume ij6¼ji and estimate 1771 country-pair
intercepts in 17 years. As shown, all gravity variables are significant statistically. According to the results for the HTP model predict that: (i) a 10% rise in exporter GDP should be associated with a 4.5% rise
in exports, all else constant; (ii) if an importer’s GDP rises by 10%, exports should rise by 8.5%, holding exporter’s GDP constant; (iii) exports tend to be labour-intensive and income-inelastic goods (normal nonluxury goods), as indicated by the positive signs on the population coefficients; and (iv) when exporter’s real exchange rate depreciates 10% mea-sured in the real term, it should be associated with a 2.4% rise in exports, all else constant. As we expect, a depreciation of the exporter’s currency (a larger XRijt) reflects a decrease in the value of exports and
thus increases the trade flow from exporter i to importer j, i.e. this index is expected to have a positive coefficient.
Notice that controlling for trading-pair heteroge-neity not only lowers the estimated elasticity of trade with respect to GDPs but also reverses the signs on the countries’ populations. Applying the same tech-nique, the strong bias in the residuals that found with the PCS and the Ma´tya´s three-way fixed-effects (MFE) models does not occur in the HTP model. This is clear from Column Three in Table 1, which reports the estimation results for the case when the direction of trade does not matter in international trade flows, i.e. treating ij6¼ji and estimating 1140
country-pair intercepts.
To further investigate the effects of time-invariant variables in the HTP model, we apply the techniques introduced by Cheng and Wall (2005). More specific, denote by ^ij the estimates of the 1140 country-pair
effects and incorporate into independent variables the logarithm of distance and the contiguity and language dummies, we obtain
^
ij¼ 0:821
ð0:094Þ 0:881 ln Dð0:011Þ ijþ0:137Cð0:047Þijþ0:509Lð0:023Þij
The numbers in parentheses are White-corrected SEs, and the R2¼0:239. These results suggest that the signs of the coefficients on distance, the logarithm of distance and the contiguity, and language dummies are identical to those in the PCS results, and all three variables are statistically significant determinants of the country-pair effects. Note, however, that the contiguity estimate is quite different from that obtained from the PCS model, indicating that the effect of contiguity on trade is overstated. Notice that Columns Four and Five in Table 3 present the estimation results with and without time-invariant variables, respectively, for the three-way panel data gravity model (Equation 5) in which the fixed time, exporter and importer effects are controlling for. 7
We do not describe the coefficient estimates on year dummies, neither do the results for the misspecification tests for normality, heteroscedasticity and linearity. The detailed estimates are available upon request.
Table 3. Results from the gravity estimation of regional blocs, 1981–1997 PCS HTP (pair ij 6¼ ji ) HTP (pair ij ¼ ji ) MFE MFE Model Simple pooled Two-way fixed-effect Two-way fixed-effect Three-way fixed-effect Three-way fixed-effect Dependent variable: log of exports Origin GDP 1.109** (0.011) 0.446** (0.036) 0.791** (0.020) 0.409** (0.057) 0.426** (0.051) Destination GDP 1.045** (0.009) 0.849** (0.033) 0.637** (0.021) 1.060** (0.011) 1.053** (0.009) Origin population 0.282** (0.011) 0.232* (0.133) 0.147* (0.080) 0.100 (0.283) 0.138 (0.254) Destination population 0.191** (0.010) 0.284* (0.159) 0.230** (0.079) 0.229** (0.011) 0.207** (0.012) Real exchange rate 0.281** (0.056) 0.241** (0.061) 0.201** (0.042) 0.346** (0.068) 0.374** (0.061) Distance 0.808** (0.013) 0.923** (0.012) Contiguity 0.361** (0.048) 0.320** (0.047) Common language 0.583** (0.024) 0.251** (0.046) Constant 1.205** (0.149) EEC2 0.195** (0.064) 0.436** (0.091) 0.280** (0.089) 1.082** (0.066) 0.238** (0.061) EEC X 0.138** (0.014) 0.142** (0.037) 0.184** (0.033) 0.201** (0.011) 0.033 (0.053) EEC M 0.052* (0.030) 0.018 (0.044) 0.084** (0.035) 0.080** (0.032) 0.001 (0.029) EFTA2 0.501** (0.011) 0.175* (0.090) 0.159 (0.096) 1.454** (0.073) 0.324** (0.067) EFTA X 0.487** (0.027) 0.159** (0.034) 0.159** (0.032) 0.058 (0.055) 0.107 (0.049) EFTA M 0.407** (0.031) 0.081** (0.040) 0.065* (0.033) 0.264** (0.032) 0.398** (0.029) EU2 0.231** (0.011) 0.592** (0.092) 0.444** (0.091) 1.341** (0.079) 0.117 (0.073) EU X 0.491** (0.034) 0.213** (0.040) 0.243** (0.034) 0.306** (0.291) 0.152 (0.057) EU M 0.337** (0.038) 0.157** (0.047) 0.049 (0.037) 0.222** (0.040) 0.392** (0.036) CUSFTA2 0.349 (0.011) 0.155 (0.346) 0.143** (0.370) 2.286** (0.473) 0.293 (0.426) CUSFTA X 0.556** (0.066) 0.134** (0.049) 0.194** (0.050) 0.191** (0.081) 0.131* (0.073) CUSFTA M 0.407** (0.076) 0.046 (0.056) 0.088 (0.055) 0.414** (0.078) 0.259** (0.070) NAFTA2 0.882** (0.308) 0.430** (0.214) 0.446* (0.236) 2.511** (0.315) 0.657** (0.284) NAFTA X 0.443** (0.067) 0.183** (0.051) 0.175** (0.050) 0.306** (0.084) 0.187** (0.074) NAFTA M 0.066 (0.086) 0.336** (0.077) 0.330** (0.076) 0.561** (0.110) 0.019 (0.099) LAFTA2 0.732** (0.074) 0.297** (0.105) 0.390** (0.111) 2.568** (0.073) 0.925** (0.063) LAFTA X 0.277** (0.031) 0.227** (0.064) 0.082 (0.057) 0.932** (0.092) 0.378** (0.083) LAFTA M 0.662** (0.032) 0.361** (0.062) 0.472** (0.057) 1.166** (0.033) 0.702** (0.298) MERCOSUR2 1.630** (0.265) 0.384* (0.198) 0.461** (0.212) 4.347** (0.270) 1.757** (0.245) MERCOSUR X 0.682** (0.078) 0.469** (0.080) 0.305** (0.074) 1.121** (0.126) 0.631** (0.113) MERCOSUR M 0.395** (0.082) 0.247* (0.077) 0.150** (0.074) 0.810** (0.084) 0.402** (0.075) # Parameters 46 1812 1180 129 132 Log-likelihood 53 390.8 37 972.7 40 314.6 54 147.3 50 868.3 Adjusted R 2 0.7050 0.8941 0.8762 0.6898 0.7553 Akaike Info. Crt. 3.550 2.643 2.757 3.606 3.388 Schwarz Baye. Crt. 3.562 3.143 3.082 3.641 3.424 Notes : SEs are in parentheses. The estimates on year dummies are omitted here. ** Denotes significance at the 5% level. * Denotes significance at the 10% level.
Furthermore, a comparison of the results con-tained in Columns One and Two in Table 3 suggests a clear distinction between the PCS model (which assumes trading-pair homogeneity) and the HTP model (which allows for trading-pair heterogeneity). Table 3 also briefly summarizes the number of parameters, the maximised value of log-likelihood, and the information criteria for the models we estimate. In the table, the main interests lie in the information criteria, which are useful for comparing the fit of specifications. Usually, we judge empirical power using the adjusted R2, which suggests that estimating the HTP model by treating pairijand pairji
as different cases is preferred to using the alternative specifications. Additionally, if we are concerned that the adjusted R2 does not penalize the addition of right-hand-side variables heavily enough, then we may take two commonly used approaches, namely the Akaike’s information criterion and Schwartz Bayesian criterion, to provide further information (See Pindyck and Rubinfeld, 1998). The result still convinces us that using the HTP model, which presents the lowest information criterion is the favourite.
Single and staged effects of regional economic agreements on trade
As indicated above, this article contributes to the literature by introducing trading-pair heterogeneity in the gravity model with some modified REI dummies and comparing the estimation results to previous studies with different specifications. Based on two reasons, we use modified RIE dummies to explore the static trade effects in the enlargement and enhance-ment of integration schemes. First, the aggregate share of intra- and inter-EC and American trade accounts for nearly two thirds of international trade. Second, the dummy variables indicate the two-stage integration schemes in the European Community and those in America, respectively. We note, however, that it can never be exhaustive to introduce dummy variables characterizing trading partners either for members of a preferential trading arrangement (or political alliance), or for sharing common culture (or languages).
As noted above, dummy variables REI2ijt which
stand for both trading partners in the same regional blocs in time t are used to measure the intra-REI effect on bilateral trade; while dummy variables REIXijt (REIMijt) denoting that exporting country i
(importing country j) is a REI member in year t are used to estimate the extra-REI effect. In general, the intra-REI effect is expected to be positive, if the elimination of protective tariffs and quantitative
trading barriers induces integrating economies to specialize within the regional bloc following their comparative advantage, i.e. the coefficients on REI2ijt
are expected to be positive.
In contrast, there are a variety of arguments about the extra-REI effect: the signs of REIXijtand REIMijt
are ambiguous. In fact, the adoption of a regional trading arrangement may have a positive effect on member countries’ exports to outsiders. This can arise when, for example, (i) more competition in a larger economy which increases the efficiency of produc-tion, leading to increased exports; (ii) stronger power of the negotiation against outsiders, which enables worldwide trading agreements to be more in favour of their own interests; (iii) inefficient trading arrange-ments among themselves, which gives incentives to divert to countries outside the bloc. On the other hand, the implementation of regional trading agree-ments could give rise to a negative effect on imports. Notable reasons include (i) trade within the group increases subsequent to the comparative advantage resulting from the removal of trade barriers between members; (ii) the larger trading blocs can raise tariffs and/or nontariff barriers against outsiders, in parti-cular, in the presence of the interest groups attempt-ing to influence the trade policies.
The REI dummies are used as follows. EEC and EFTA are effective during 1981–1993 in the data set; while the EU dummy is set to be one after the European Economic Area comes into effect in 1994. CUSFTA is effective after Canada-US FTA enters into force in 1989–1993, while NAFTA goes into effect in 1994. LAFTA is effective during 1981–1994 in the data set, while the MERCOSUR dummy is set to be one when their members reach a compromise agreement and the CU becomes effective in 1995. There are three REI dummies for each bloc. For instance, EEC2ijt is set to be one if country i and
country j are both parties to the EEC in year t, and zero otherwise. EECxijt equals one when only the
exporting country i is acting in EEC in year t, and zero otherwise. EECMijt takes the value of one for
imports by any EEC member country from any outsider in year t, and zero otherwise.
Based on the above selected models, we assess and discuss the extent to which a REI can affect on trade volume by examining the results from the HTP model. First, focusing on the REI dummies for the European blocs (including EEC, EFTA and EU) in the second column (in which the pair-wise hetero-geneity is controlled), we see that the coefficient on EEC2 is positive and significant statistically, i.e. their intra-REI effects are significant. This suggests the trade pattern within-EEC exhibiting a sign of trade creation. More specific, the estimation results imply
that the EEC countries trade among themselves 43.6% more than an average trading pair outside the bloc during the period. Looking further into the estimates for EECX and EECM, the increases have
been accompanied by decreased levels of trade with the rest of the world. Exports to and imports from a nonmember country are 14.2 and 1.8% lower than gravity-model predictions, respectively. The negative coefficient of EECM may also be interpreted as
that EEC members are less open than an average country, but the effect is insignificant.
Notably, the coefficient on EFTA2 is negative and significant statistically at the 5% level, indicating that the EFTA countries trade about 17.5% less among themselves than an average trading-pair outsider. On the contrary, the coefficients of EFTAXand EFTAM
are positive and significant statistically, suggesting exports to and imports from a country outside the EFTA were 15.9 and 8.1% higher than the model’s prediction. We should note that the coefficients on REI dummies require a different interpretation for the countries which are both the EEC and EFTA members such as Denmark and Portugal (an EEC member since 1986). For instance, even trade between Denmark and Portugal is the same as that between two identical countries inside the trading bloc, the intra-REI effect is calculated by the summation of the
coefficients on EEC and EFTA, i.e. 0.261
(¼0.436 0.175). Also note that as shown in the first and second columns in Table 2, the coefficients on the EFTA dummies from estimating the PCS model vs. the HTP model are quite different. All coefficients on the EFTA dummies in Column One display the same signs and statistical significance with those estimated in Wei and Frankel (1997).8
The coefficient on EU2 is positive and statistically significant, suggesting that EU countries on average trade 59.2% higher than the prediction of the modified gravity model. In terms of the REI staged effect, for EU countries, which were EEC-members trade among member countries increases from 43.6 to 59.2% after the EU integration enhances. For the EU countries, which were EFTA-members, the staged effect demands a careful interpretation. Indeed, for the countries, which are both the EEC and EFTA members such as Denmark and Portugal, trade among member countries increases from 26.1 to 59.2% after the EU comes into effect in 1994. For the countries, which are the EFTA members such as Austria, Finland and Sweden, their trade pattern exhibits strong evidence of increasing bias among members after they join the EU. However, trade with
the EU members, which already participate in the former EEC appears to be higher than that between two identical countries outside the trading bloc from 15.9 to 59.2%.
Canada and US have approximately raised intra-member trade 15.5% higher than normal trade during 1989–1993, but the predicted effect is statis-tically insignificant. Their exports to a country outside the integrated economy are statistically significant 13.4% less than the model’s prediction. The imports from a country outside the CUSFTA are 4.6% less than an average trading-pair. These results indicate the two economies seem to be less open than an average country, but the predicted effect is statistically insignificant.
We then move on to NAFTA. It is clear from the estimates for NAFTA in the second column in Table 3, bilateral trade between the NAFTA coun-tries is higher than the prediction of the gravity model by about 43%. Moreover, their exports to a nonmember country are 18.3% less while their imports from the rest of the world are 33.6% more than a normal trading pair. In terms of the REI staged effect, the creation of NAFTA that builds on CUSFTA has clearly increased trade in the grouping in the short-term by 27.5% (from 15.5 to 43%). On the other hand, their imports from the rest of the world increase by 38.2% (from 4.6 to 33.6%). This suggests that the NAFTA countries are still open to trade than other countries. Note, however, that the NAFTA has only been practised for 4 years in our sample, it is difficult to precisely determine the effects of the NAFTA integration scheme on trade flows.
Finally, we discuss the development of economic integration in Latin America. The LAFTA group appears to show a certain degree of intra-REI trade bias. The coefficient on LAFTA2 is positive and statistically significant at the 5% level, indicating that trade within the group is higher by 29.7% than trade between two random countries outside the group. Moreover, the coefficients of both LAFTAX and
LAFTAM are negative and significant statistically,
22.7 and 36.1%, respectively. This implies that there exists an apparent diversion of trade away from countries outside toward countries inside the LAFTA group.
On the basis of the LAFTA, the MERCOSUR constituted by four South American countries (Argentina, Brazil, Paraguay and Uruguay) also drives for more intensive trade among members. Based upon our estimation, the coefficient on MERCOSUR2is positive and significant statistically 8
They adopt a specification similar to the PCS model in the current article, but estimate the data during the period of 1970–1992.
at the 5% level, suggesting that bilateral trade among the MERCOSUR countries is higher than the prediction of the gravity model by about 38.4%. Moreover, the MERCOSUR’s exports to a non-member country are nearly one half less than a normal trading pair and appear significant statisti-cally at the 5% level. Contrarily, their imports from the rest of the world are higher than the model’s prediction by 33.6%. In terms of the REI staged effect, the enhancement of preferential trade agree-ments has increased trade among member countries from 29.7% in the LAFTA to 38.4% in the MERCOSUR. However, their exports to countries outside the bloc diminish significantly from 22.7 to 46.9%. As far as extra-group bias is concerned, member countries’ imports from the rest of the world surprisingly step up by 60.8% (from 36.1 to 24.7%). This suggests the MERCOSUR countries are quite open to other countries in 1995–1997 even in the presence of an effective trade liberalization program and the expanding trade.
IV. Conclusions
The main focus of this article has been the estimation of the single and the staged effects of regional blocs on bilateral trade flows at the aggregate level using the heterogeneous trading-pair (HTP) model. Based on our results, we conclude the conventional vari-ables such as the exporter’s and importer’s incomes have positive effects on bilateral trade, but the income elasticity of bilateral trade is less than that estimated by using the conventional gravity specification. However, the coefficients of the exporter’s and importer’s population are positive and statistically significant, which is contrary to the conventional wisdom. The price effect, which is indicated by the real exchange rate variable shows the similar result as expected.
In contrast to previous works that employed the gravity model in which bilateral trade flows are a log-linear function of the sizes of trading partners’ incomes, populations and distance between them, we have demonstrated that the standard models suffer from a heterogeneity bias owing to mis-specified or omitted variables. In fact, when the effects of REIs are concerned, the standard way of assessing such trading agreements is to add dummy variables indicating a common grouping into the gravity models. Nevertheless, failure to take account of falling transportation costs, declining multilateral trade barriers in recent decades under GATT and WTO, and specific preference between trading
partners, namely the exclusion of pair-specific or country-specific factors from the standard gravity models, can cause the REI effects to be overstated.
To appropriately explain the REI effects on within-and extra-regional trade, we employed three dummy variables for each regional bloc to discuss three types of regional economic arrangements: North–North (EEC, EU and CUSFTA), North–South (NAFTA) and South–South (LAFTA and MERCOSUR). To further provide an interpretation of the conse-quences of the enlargement and enhancement of REIs, we uses the REI dummies distinguished by timing, i.e. it treats a single regional grouping which has experienced different degrees of integration as different schemes of REI. We have seen a range of differences in the single and staged effects of regional economic arrangements on trade, and provided evidence indicating that some conventional conclu-sions about the effect of REI on trade seem fragile.
Through an exhaustive search of possible models we have learned that specification of heterogeneity is important to measurement of the effects of regional integration agreements on trade volumes. With regard to the choice between the available gravity models, this article follows the model selection criteria commonly used in the literature. As one has been aware, the model controlling for trading-pair heterogeneity (model HTP) is in general preferred to the others.
Acknowledgements
We would like to thank Ron Smith, Howard Wall, Roderick McCrorie, Bih-Jane Liu, Shinn-Juh Lin, Vey Wang, Yen-Huang Chen and seminar partici-pants at the First Annual Conference of the Applied
Economic Research at National Chung-Hsing
University, and the Sixth Annual Conference of the Empirical Economic Research at National University of Kaohsiung, for their helpful comments and suggestions. I-Hui Cheng is also grateful for financial support from the NSC (Taiwan) No. 89-2415-H-309-002 when underlying this article.
References
Aitken, N. D. (1973) The effect of the EEC and EFTA on European trade: a temporal cross-section analysis, American Economic Review, 63, 881–92.
Baltagi, B. H. (1997) Econometric Analysis of Panel Data, John Wiley and Sons, New York.
Bayoumi, T. and Eichengreen, B. (1997) Is regionalism simply a diversion? Evidence from the evolution of the EC and EFTA, in Regionalism versus Multilateral Trade Arrangements (Eds) T. Ito and A. O. Krueger, University of Chicago Press, Chicago.
Bergstrand, J. H. (1985) The gravity equation in interna-tional trade: some microeconomic foundations and empirical evidence, Review of Economics and Statistics, 67, 474–81.
Bergstrand, J. H. (1989) The generalized gravity equation, monopolistic competition, and the factor-proportions: theory of international trade, Review of Economics and Statistics, 71, 143–53.
Bhalla, A. S. and Bhalla, P. (1997) Regional Blocs: Building Blocks or Stumbling Blocks?, Macmillan Press, London.
Bikker, J. A. (1987) An international trade flow model with substitution: an extension of the gravity model, Kyklos, 40, 315–37.
Cheng, I.-H. and Wall, H. J. (2005) Controlling for trading-pair heterogeneity in gravity models of trade and integration, The Federal Reserve Bank of St. Louis, Review, 87, 49–63.
Egger, P. (2004) Estimating regional trading bloc effects with panel data, Review of World Economics (Weltwirtschaftliches Archiv), 140, 151–66.
Endoh, M. (1999) Trade creation and trade diversion in the EEC, the LAFTA and the CMEA: 1960–1994, Applied Economics, 31, 207–16.
Endoh, M. (2005) The effect of the GSTP on trade flow: mission ccomplished?, Applied Economics, 37, 487–96.
Frankel, F., Stein, E. and Wei, S. (1995) Trading blocs and Americas: the natural, the unnatural, and the super-natural, Journal of Development Economics, 47, 61–95.
Frankel, F., Stein, E. and Wei, S. (1998) Continental trading blocs: are they natural or super-natural?, in The Regionalization of the World Economy (Ed.) J. A. Frankel, University of Chicago Press, Chicago.
Frankel, F. and Wei, S. (1998) Regionalization of world trade and currencies, in The Regionalization of the World Economy (Ed.) J. A. Frankel, University of Chicago Press, Chicago.
Karemera, D., Oguledo, V. I. and Davis, B. (2000) A gravity model analysis of international migration to North America, Applied Economics, 32, 1745–55.
Linnemann, H. (1966) An Econometric Study of International Trade Flows, North-Holland Publishing Company, Amsterdam.
Mansfield, E. D. and Bronsor, R. (1997) The political economy of major power, in The Political Economy of Regionalism(Eds) E. D. Mansfield and H. V. Milner, Columbia University Press, New York.
Ma´tya´s, L. (1997) Proper econometric specification of the gravity model, The World Economy, 20, 363–8. Mayes, D. G. (1978) The effects of economic integration on
trade, Journal of Common Market Studies, 57, 1–25. Nilsson, L. (2002) Trading relations: is the roadmap from
Lome´ to Cotonou correct?, Applied Economics, 34, 439–52.
Oguledo, V. I. and MacPhee, C. R. (1994) Gravity model: a reformulation and an application to discriminatory trade arrangements, Applied Economics, 26, 107–20. Pindyck, R. S. and Rubinfeld, D. L. (1998) Econometric
Models and Economic Forecasts, 4th edn, McGraw-Hill, New York.
Po¨yho¨nen, P. (1963) A tentative model for the volume of trade between countries, Weltwirtschaftliches Archive, 90, 93–100.
Sanso, M., Cuairan, R. and Sanz, F. (1993) Bilateral trade flows, the gravity equation, and functional form, Review of Economics and Statistics, 75, 266–75. Tinbergen, J. (1962) Shaping the World Economy:
Suggestions for an International Economic Policy, The Twentieth Century Fund, New York.
Thursby, J. G. and Thursby, M. C. (1987) Bilateral trade flows, the Linder hypothesis and exchange risk, Review of Economics and Statistics, 69, 488–95.
Wei, S. J. and Frankel, J. A. (1997) Open versus closed trading blocs, in Regionalism versus Multilateral Trade Arrangements (Eds) T. Ito and A. O. Krueger, University of Chicago Press, Chicago.
Whalley, J. (1998) Why do countries seek regional trade arrangements, in The Regionalization of the World Economy (Ed.) J. A. Frankel, University of Chicago Press, Chicago.
