Agricultural market integration in the OECD: A gravity-border effect approach

Available online at www.sciencedirect.com

Food Policy 33 (2008) 165–175 www.elsevier.com/locate/foodpol

Agricultural market integration in the OECD: A gravity-border effect approach Alessandro Olper *, Valentina Raimondi Dipartimento di Economia e Politica Agraria, Agroalimentare e Ambientale, Universita` degli Studi di Milano, via Celoria, 2 – 20133 Milano, Italy Received 27 July 2006; received in revised form 20 March 2007; accepted 26 June 2007

Abstract This paper uses the border effect estimate from a gravity model to assess the level of agricultural market trade integration among 22 OECD countries for the 1994–2003 period. Empirical analysis confirms that the use of a gravity equation derived from theory, in the estimation of border effect, matters. A representative estimate of the border effect shows that crossing a national border within the OECD induces an average trade-reduction effect of a factor 13. This average value masks differences that are quite substantial in market integration, with value for intra-EU trade being higher while that for trade between the Central and Eastern European Countries (CEECs) is lower. The data show a process of strong integration in all the country-trade combinations involving CEECs. However, quite surprisingly, the intra-CEEC and OECD-CEEC integration processes are almost twice as strong as those in the EU-CEEC combination. Finally, the equivalent tariffs implied by the estimated border effects are not implausible compared to the actual range of direct protection measures. Ó 2007 Elsevier Ltd. All rights reserved. Keywords: Trade integration; Agriculture; Border effect; Gravity; OECDs

Introduction The process of trade liberalization implied by WTO negotiations, as well as by unilateral initiatives such as Everything But Arms (EBA) of the European Union (EU), has increased the demand for studies aimed at a better understanding of market access issues. Our paper contributes to this literature by analyzing the process and the levels of agricultural market integration among OECD countries. We depart from recent literature in that we use an indirect estimation approach. Specifically, we estimate the (inverse) level of trade integration within the OECDs using the gravity-border effect methodology. The use of an indirect measure is due to the difficulties encountered in analyzing and estimating protection and market access issues by direct measurements. For example, *

Corresponding author. Tel.: +39 02 50316481; fax: +39 02 50316486. E-mail addresses: [email protected] (A. Olper), valentina. [email protected] (V. Raimondi). 0306-9192/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.foodpol.2007.06.003

a look at the literature on EU agricultural protection reveals a spread of estimates, ranging from the 40% of Messerlin (2001) to the 10% of Gallezot (2003). While these differences can be explained by the data used and the assumptions made in calculation (see Bureau and Salvatici, 2004), the evidence associated with direct protection measures remains questionable (Fontagne´ et al., 2005). Firstly, average tariff figures mask a reality based on numerous tariff peaks. Secondly, it is quite difficult to include the complex system of preferential agreements developed by many rich countries (notably the EU) in the estimation of average ad valorem tariffs. Moreover, zero tariffs and zero quotas do not necessarily mean free access, due to others measures at the border, such as sanitary, phytosanitary and technical regulations, as well as other impediments unrelated to trade policy, such as information related costs and ‘home bias’ in preferences. Given these problems, the literature now gives consideration to the possibility of using a complementary and indirect measure based on recent developments in gravity

166

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175

literature. Indeed this ‘old’ methodology has recently been renewed in two important and related respects: the border effect approach and a more careful theoretical foundation. The border effect approach, initiated by McCallum (1995) comparing intra-national trade between Canadian provinces with international trade between Canada and the US, subsequently has stimulated a large amount of research. The underlying idea is to measure the (inverse) level of trade integration between two countries, comparing their bilateral trade with respect to the trade flows taking place within their own borders. The estimated border effect shows how much trade within countries is above international trade, due to cross-border measures such as tariffs, non-tariff barriers and all other factors that might impede trade. Thus, the border effect captures all trade impediments related to the existence of national borders. This could be a considerable advantage as it is quite hard to measure the majority of such impediments directly. For instance, consider the lack of reliable statistics on technical, sanitary and phytosanitary barriers, so pervasive in agri-food markets, and the conceptual difficulty in estimating their trade (and welfare) effect. By looking at the overall picture based on an indirect estimation approach these problems can be overcome.1 The McCallum (1995) estimate of border effect was extraordinarily high, indicating that Canadian cross provincial trade was 22 times – or 2200%! – greater than Canada–US cross-border trade in 1988. The underlying reason for this impressive number was explained only recently by Anderson and van Wincoop (2003), showing that the McCallum finding was a combination of two key distortions: an asymmetry effect of the border on countries of different size, and a miss-specification of the traditional gravity equation with respect to what the gravity theory tells us. In this paper, we apply the gravity model developed by Anderson and van Wincoop (2003) to agricultural bilateral trade flows between 22 OECD countries, from 1994 through 2003. The empirical analysis tries to address two main questions: First, does a theory based gravity equation do a good job in the estimation of border effect in agricultural markets? Second, is the border effect in agriculture strongly affected by trade policies or not? Our analysis is linked to the literature that used gravitytype models to analyze different features of agricultural trade costs. Recent examples in this direction deal with the effect of food safety standards on bilateral trade (see Otsuki and Wilson, 2001; Otsuki et al., 2001), the impact of distance on US agricultural exports (Wang et al., 1 Another advantage of using the border effect approach, recently stressed by Mayer and Zignago (2005), is that it accounts for the fact that most internal demand is met by domestic producers, not foreign. Thus, an ideal protection index from the point of view of foreign firms needs a benchmark based on the best possible market access situation, that is the one faced by national producers on the home market.

2000), the effect of exchange rate uncertainty (Cho et al., 2002) and of tariffs and non-tariff barriers on agricultural trade (Haveman and Thursby, 2002). However, till now, few studies have applied the border effect estimate to agricultural trade flow and, most importantly, rarely have they explicitly taken into account the important contribution of Anderson and van Wincoop (2003). Our focus on agricultural trade is motivated by the fact that while several papers analyzing national border effects in manufacturing sector (see Anderson and van Wincoop, 2004, for a survey) exist, the only paper published on this topic in agriculture is, to the best of our knowledge, that of Furtan and van Melle (2004) who estimate the border effects of the US, Canada and Mexico using a standard gravity model. Furthermore, agricultural products are characterized by high protection levels, complex tariff structures, low transportability and strong ‘home bias’ in preferences, all factors that induce large border effects. Finally, agriculture is presently at the center of the WTO trade talks; thus, a close examination of the level and trends in border effect, in an endeavor to better understand the degree to which trade policies matter, would offer new insight into this literature. The paper is organized as follows: Section ‘‘The derivation of the gravity equation” summarizes the structure of the model and derives the gravity equation. Section ‘‘Data and measures” describes the data sources and variables used in the empirical model. Section ‘‘Result” is devoted to the presentation of our empirical results. The final section discusses the main implications and our conclusions. The derivation of the gravity equation Our main goal is the estimation of a bilateral trade model with a gravity specification close to theory. Anderson and van Wincoop (2004, 2003) recently showed that the correct derivation of the gravity equation from theory is crucially important to the validity of empirical results, and this is especially true in the case of border effect estimation (see Feenstra, 2002). Our derivation of the gravity equation summarizes Anderson and van Wincoop (2004). We outline only the key features of the model here, for the full theoretical setup the reader is directed to the above papers that review recent gravity literature. Theoretical framework A gravity-like equation can be derived from a variety of different trade models2. Indeed, Anderson and van Wincoop (2004) (hereafter AvW) argue that trade models, where the allocation of trade across countries can be analyzed separately from the allocation of production and con2 The first theoretical derivation of a gravity model is due to Anderson (1979). Deardoff (1998) derived gravity equations from the Hesckscher– Ohlin model, Bergstrand (1989) from models with monopolistic competition, Eaton and Kortum (2002) from Ricardian models.

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175

sumption within countries, give a gravity-like structure. The AvW model includes identical constant-elasticity-ofsubstitution (CES) regional preferences over goods differentiated by country of origin, and ‘iceberg’ trade costs, namely the bilateral trade costs, tij, are proportional to the quantity of trade.3 Let X kij be the exports from i to j of sector k. Under this set-up, AvW shows that the model yields the following compact trade pattern characterization between exporter i and importer j !1rk k k k C Y t j ij i X kij ¼ k ð1Þ Q Y P kj ki where ð

k Y

Þ

1rk

i

ðP kj Þ

¼

X

tkij

!1rk

C kj

P kj Yk !1rk X tkij Y ki ¼ Qk Yk i i

ð2Þ

j

1rk

ð3Þ

and Yk is world output for sector k; Y ki ðC kj Þ are theQexpork ter (importer) output (consumption) for sector k; i and k P j are inward and outward price indices called ‘multilateral resistance factors’ because they are positively affected by all the bilateral resistance terms, tkij ; finally, rk is the substitution elasticity of the importing country j. As in the traditional gravity equation, trade depends positively on the size of each country and negatively on the trade barriers. However, the key implication of Eq. (1) is that bilateral trade depends on relative trade barriers, namely the bilateral barrier tkij divided Qk by the product of their ‘multilateral resistance indices’ i and P kj . Thus, the gravity equation suggests that trade between two regions depends, after controlling for size, on the bilateral barrier between them, in relation to the average trade barriers that both regions face with all their trading partners. The next step before deriving an estimable equation is the specification of the unobservable trade cost factor tij. Following other authors, trade costs can be separated into transport costs, dij, and a border component, bij, both conveniently expressed in the loglinear form d

tij ¼ d qij bijij

ð4Þ

where dij is distance between the economic centers of i and j, and bij  1 is the tariff equivalent of all the trade barriers associated with the border. dij is equal to zero if i and j are in the same countries (intra-country-trade) and is equal to 1 3

The ‘iceberg’ trade costs hypothesis implies that the price pij charged by i for export to j – inclusive of transport costs, tij – can be written as pij = pitij, where pi is the local supply price received by producers in country i net of any transport costs. Because tii = 1, and tij = 1 we have that tij units of the product must be shipped to country j in order for one unit to arrive. In other words the amount (tij – 1) ‘melts’ along the way and is the ad-valorem tariff equivalent of trade costs.

167

if they are separate countries (cross-border or international trade). Before concluding this section it is useful to have a better understanding of the significance and current interpretation of the border costs, bij. Specifically, we can highlight three broad, but conceptually different, factors (see Anderson and van Wincoop, 2002; Evans, 2003; Chen, 2004): (i) border costs related to policy barriers such as tariffs and non-tariff barriers (NTB); (ii) border costs unrelated to policy barriers, such as consumer ‘home bias’ in preference, and information related costs; and, finally (iii) substitutability between imports and domestic goods. Focusing on the first two components, the key question is that while trade policies generate rent for government and/or private beneficiaries, factors unrelated to trade policy such as transaction costs or ‘home-bias’ in preferences do not generate rents. This distinction between rent and non-rent border costs have significant and practical implications. In fact, it is important to note that if much of the border effect arises from barriers that do not create rents – think for example of the coordination of safety regulations or preference for home goods – then its welfare impact will be greater than a situation where much of the border effect is due to rent barriers, such as tariffs. However, in the latter situation, the potential role for policy will be more complex because non-rent border costs are probably not so easy to eliminate by policy choices. Empirical specification Replacing the trade cost factor in (1) with (4), then the export volume from i to j can be expressed in log-linear form as (omitting for simplicity the constant and the superscript k) xij ¼ y i þ cj þ ð1  rÞq ln d ij þ ð1  rÞðdij Þ ln bij  lnðPi Þr1  lnðP j Þr1

ð5Þ

where lowercase variables represent the natural logarithms of their uppercase variables. In the estimation of the gravity equation (5) the main problem is to take account Q of the unobservable multilateral resistance factors, i and Pj, implied by the theory. The literature proposes three main, but different, approaches (see Feenstra, 2002): the use of price index such as consumer price index to measure the price effects in the gravity equation, as in Baier and Bergstrand (2001); the use of non-linear least squares to solve the system of Eqs. (1) and (2) as in Anderson and van Wincoop (2003); and, finally, the replacement of multilateral resistance terms with country dummies as in Hummels (2001) and Feenstra (2002). As recently shown by Feenstra (2002), only the last two approaches lead to consistent estimates. However, the former of these is more complex so the use of the fixed effects method is preferable due to its simplicity. Moreover,

168

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175

another important advantage of using a fixed effects specification is to sweep out any other unobservable variables omitted in the trade costs function (4).4 Thus, we will run our key estimations using the fixed effects for source and destination countries. Introducing a constant b0, the border coefficient c = (1–r) lnbij, and the error term eij, we obtain xij ¼ b0 þ hi þ kj þ qð1  rÞ ln d ij þ cðdij Þ þ ð1  rÞeij ð6Þ Q where hi = yi  (1  r) ln i is a fixed effect of the exporting country and kj = cj  (1  r) lnPj is a fixed effect of the importing country. In Eq. (6) the key parameters to be estimated, other than the fixed effects, are the distance coefficient q(1  r), and the border effect coefficient c = (1  r) lnbij. Taking the antilog of the estimated border coefficient [=exp(c)], we have an estimate of the so called border effect, namely how much intra-country trade is above international trade, after controlling for size, transport costs and any other unspecified trade costs. Moreover, it is important to note that the inclusion of country-fixed-effects in (6) removes (only) Qthe cross-section correlation between the unobservable i and Pj terms. However, as our panel of data is also intended to extract information on the evolution in border effect, the timeinvariant fixed effects in (6) do not remove the potential time series correlation bias. Indeed, if in the observed period the determinants of bilateral trade costs vary over Qtime, then these determinants are also in the unobserved i and Pj terms. Thus, in order to handle this potential source of bias, we will also test a specification that includes timevarying country fixed effects, as suggested by, among others, Baldwin and Taglioni (2006).5 Before concluding this section it can be useful to compare the gravity equation (6) with a traditional (or a McCallum) gravity specification like Eq. (7) xij ¼ b0 þ a1 y i þ a2 y j þ q ln d ij þ cðdij Þ þ eij :

ð7Þ

This specification treats the gravity model as an empirical regularity, without an explicit link to a theoretical model with micro-foundations.6 Clearly, the key difference between the specification (7) and (6) is that the former omits the multilateral resistance terms implied by the theory. Thus, the estimate of the border coefficients c would be biased, because the trade costs tij are correlated with the multilateral resistance indices, which are themselves a function of trade barriers (see Eqs. (2) and (3)). Thus, if the objective of the researcher is to use the gravity equation 4 Moreover, it is important to note that the same fixed effects specification emerges from theoretical gravity models based on different assumptions about consumer preferences or technology, suggesting that the above specification is quite general (see Estevadeordal et al., 2003; Helpman et al., 2004). 5 This involve the inclusion of several dummies. Specifically, with N nations and T the number of years, we must include 2NT = 420 dummies! 6 Note that a gravity specification like Eq. (7) can, in any case, be derived from quite different theoretical models, as discussed by Feenstra et al. (2001).

to infer trade cost parameters, then the kind of specification adopted should be of critical importance. Data and measures Our gravity model includes 22 OECD countries: 9 OECD, 9 EU and 4 CEEC.7 Let us consider the total agricultural imports of the 22 OECD countries from all the other 22 OECD countries over the 1994–2003 period. The data set is almost perfectly ‘square’, presenting 4840 theoretical observations, of which 4620 (=22  21  10) are bilateral cross-border observations, and 220 (=22  10) are intracountry trade observations. In the dataset, there are 81 (1.7%) zero bilateral trade flows, thus, the real total non-zero observations drops to 4759. Those agricultural trade data consider 40% of the world agricultural trade flow and 60% of the OECD agricultural imports from the world. The needed data involve, primarily, bilateral trade, production and consumption data, in a comparable industry classification. The original trade data come from the United Nations Comtrade database. We consider the data reported by the importer countries, starting from HS-96 at 6-digits. Then, we use the conversion table from HS-96 to ISIC Rev.3 at 4-digits, to aggregate import data at the 2-digit ISIC level. This conversion allows us a fully comparable classification of trade and agricultural production and consumption data. Indeed, the agricultural output comes from the OECD STAN database that uses ISIC Rev.3 at the 2-digit level. Thus, our agricultural trade and production data consider the following 2-digit ISIC codes: agricultural, hunting and forestry (01-02), plus fishing (05). The consumption data of the importer country are calculated as the difference between total agricultural production and total exports to the world, plus total imports from the world. The empirical implementation of Eq. (6) needs intracountry trade data. However these figures were not available for our country sample. Thus, as is common in this literature, a country’s ‘imports’ from itself are calculated as in Wei (1996). Such imports are defined as the difference between total agricultural production and total exports to the rest of the world.8 The data come from the same, above described sources.

7

The 22 trading countries are: Australia, Canada, Czech Republic, Denmark, France, Germany, Greece, Hungary, Italy, Japan, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal, Slovak Republic, South Korea, Spain, Switzerland, United Kingdom and United States. Excluded from our sample are data for Belgium and Ireland. In the former, the Comtrade database (as Eurostat) collects the data with that of Luxembourg for 1995–1998 years. Instead for Ireland we lack comparable data on production. 8 A limitation of this measure in the agricultural sector is due to the fact that some of the differences between total production and exports might be accounted for by storage and not by internal trade. However, working at the aggregate level, it appears quite difficult, if not impossible, to correct for this potential bias.

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175

Moreover, we need measures of distance between and within countries. We used the intra-national distance recently proposed by CEPII. This distance database has the considerable advantage of making internal distance constructions consistent with international distance calculations. Note that, as is evident from the specification of trade costs (4), and as shown empirically by Head and Mayer (2002), any overestimate of the internal distance relative to the external one will mechanically translate into an overestimate of the border effect. In the CEPII database the calculation is based on the bilateral distances between cities, weighted by the share of the city in the overall population of the country. This procedure is used for both internal and international distances. Finally, we also take into account whether or not two countries share a common border and a common language, adding two different dummies to our specification. Following Helliwell (1997), the two dummies take a value 1 when country i and country j (for i 6¼ j) speak a common language or share a common border (0 otherwise). After the inclusion of these two variables the border effect answers the following question: how much more intensely does a country trade with itself than with other unrelated countries? This will simplify the comparison of our results with previous findings that use this kind of specification of the language and adjacency dummies.9 Results Base model vs. fixed effects Table 1 shows the regression results of different specifications based on the gravity equations (6) and (7), pooled over the 1994–2003 period. Each regression includes a common intercept and a set of yearly fixed effects. We start by estimating a single average border coefficient, c, for all 22 OECD countries, using a border dummy, dij, equal to zero for intra-country trade and equal to 1 for cross-border trade. The first column reports the ordinary least square results (OLS) of a McCallum gravity equation, identical to specification (7). The overall fit of this regression is in line with the usual findings in gravity literature. All the estimated coefficients have their expected sign, and are highly significant (p < 0.01). The importer and exporter consumption and production elasticity, equal to 1.10 and 1.18, respectively, are near to the unitary value predicted by the theory.10 The trade elasticity of distance, around 1.0, is also comparable with the usual findings. Instead the lan-

169

guage and contiguity effects are larger than those obtained on OECD data by Wei (1996) and Helliwell (1997) in the manufacturing industry, but in line with those of Head and Mayer (2000). In the basic specification, the estimated border effect is quite large. A border coefficient of 3.47 means that intra-country trade is, on average, 32.1 (=exp (3.47)) times larger than the cross-border trade in OECD countries. An analogous estimate for agricultural trade in the OECD does not exist. However Furtan and van Melle (2004), using a traditional gravity equation with a specification very close to column 1, found a border effect even stronger for Canada and US agricultural trade. Because trade can affect consumption and production patterns, the specification in column (2) extends the basic gravity model by constraining the coefficients of these terms at unity, shifting the two terms on the left side of the equation. Thus, the dependent variable is now the size-adjusted trade. This modification increases the absolute magnitude of the border coefficients to 3.71. Note that there occurs a reduction in overall fit because the explanatory power of production and consumption no longer contributes to the calculation.11 Finally, column (3) reports a specification identical to column (2) but estimated with Heckman’s two stage procedure, a first-stage probit model and a second-stage OLS model12. The rationale for using this estimation procedure lies in the fact that zero trade flows in the dataset do not occur randomly, but are the outcome of a selection procedure. The significant coefficient on Mills’ ratio in the second stage, displayed in Table 1, indicates the necessity of this adjustment. As a result of this selection correction, the absolute border coefficient increases to 3.84. In columns (4), (5) and (6), we repeat the same battery of regressions by including fixed effects for source and destination countries, to check for unobserved multilateral resistances. The fixed effects (not reported) are both individually, and jointly, highly significant for both source and destination countries. This theoretical modification strongly reduces the absolute magnitude of the border coefficient. Now, crossing a national border inside the OECD reduces trade by a factor of 8.8 (=exp (2.18)), a figure that better fits our intuition and is within the same order of magnitude with previous findings for OECD countries (see Anderson and van Wincoop, 2003; Chen, 2004). Once again, using Heckman’s two stage procedure (see Column (6)) to correct for selection bias, the border effect increases to 13.2 (=exp (2.58)). Finally, to check for the robustness of Heckman’s two stage procedure we re-estimate, in column (7), the gravity

9

Another way of treating these dummies is to follow Wei (1996), that also gives values of 1 for each country’s sales to itself. However, as shown by Helliwell (1997, pp. 174–175) our formulation can be reconciled with that of Wei, simply subtracting the value of language and adjacency coefficients from that of the border. 10 A joint F-test, using the linear restriction that both coefficients are equal to one, is strongly significant.

11 Indeed, the root mean squared error of the regression is little changed from that of column (1). 12 The explanatory variables used in the first stage are all the variables of the model, with the only exception of the language dummy due to collinearity problems. The probit estimated coefficients are similar in signs and significance to the OLS.

170

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175

Table 1 Border effect average between OECD countries Model regression

Fixed effectsa

McCallum (1)

(2)

(3)

(4)

(5)

(6)

1.10 (0.02) 1.18 (0.03) 1.10 (0.03) 1.35 (0.10) 0.41 (0.09)

1

1

1

1

1

1

1

1

1.03 (0.02) 1.43 (0.09) 0.56 (0.08)

0.97 (0.03) 1.35 (0.10) 0.59 (0.08) 7.16 (1.34)

0.79 (0.21) 0.06 (0.19) 1.44 (0.03) 0.12 (0.08) 0.34 (0.08)

1.44 (0.03) 0.12 (0.08) 0.34 (0.08)

1.28 (0.03) 0.04 (0.08) 0.50 (0.07) 12.72 (1.24)

1.06 (0.04) 0.09 (0.07) 0.38 (0.10)

Border coefficient (c)

3.47 (0.12)

3.71 (0.10)

3.84 (0.11)

2.18 (0.17)

2.18 (0.17)

2.58 (0.17)

2.43 (0.11)

Average border effect [exp(c)]

32.1

40.7

46.4

8.9

8.9

13.2

11.4

Adj. R2 # Observed

0.655 4759

0.552 4759

0.556 4759

0.847 4759

0.802 4759

0.808 4759

0.982 4840

ln Cj ln Yi ln distanceij Language Contiguity Mills ratio

(7)

Notes: Columns (1), (2), (4) and (5) show the OLS regressions, columns (3) and (6) the second stage of the Heckman procedure, column (7) the Poisson pseudo-maximum likelihood estimation method (see text). Dependent variable: lnXij in regressions (1) and (4); ln(Xij/CjYi) in regressions (2), (3), (5), and (6); Xij in regression (7). Each regression includes a common intercept and a set of yearly fixed effects. Heteroskedasticity robust standard errors are in parentheses. a Included fixed effect for source and destination countries.

equation with the zero-trade pairs, using the Poisson pseudo-maximum likelihood method (PPML) recently suggested by Santos Silva and Tenreyro (2006). These authors argue that an OLS estimation of the gravity equation is inconsistent because of heteroskedasticity, and bias as it omits zero trade flows. Thus, PPML may represents a ‘natural way’ to deal with zero in trade flows. Comparing the PPML with the OLS results for the positive-trade sub-sample (columns (7) and (5) of Table 1) shows that distance elasticity is substantially larger under OLS, in line with the Santos Silva and Tenreyro (2006) findings. On the contrary, no significant differences in contiguity and language effect are detected. Moreover, a similar decrease in distance effect can be shown in the OLS estimation when it is corrected for selection bias, as reported in column (6). More importantly for our purpose is the fact that both the Heckman and the PPML techniques induce a comparable increase in the border coefficient value, from 2.18 for the simple OLS to 2.58 and 2.43 for Heckman and PPML, respectively. Thus, notwithstanding the potential advantage of using the PPML method to estimate the gravity-like model, for our sample and specification no substantial differences were detected in the estimated coefficients, especially when the OLS was corrected for selection bias. Overall, these results confirm that using a gravity equation derived from theory in the estimation of border effect matters. Indeed, an a-theoretic gravity equation strongly inflates the border effect estimate, that suffers an omitted variable bias, as discussed in Section ‘‘Empirical specification”. Moreover, border effect in OECD agricultural mar-

kets appears higher than those estimated for manufacturing trade, probably because agricultural products may be affected by additional factors such as perishability, higher protection and the Community agricultural policy among EU countries. Controlling for country specific effects also affects the other coefficients of the equation. Specifically, we can see a strong reduction in the language coefficient, an increase in the distance elasticity, and a switch in the sign of the exporter production elasticity that loses substantial significance. Note that with fixed effects included, the importerconsumption and exporter-production coefficients explain only the time dimension in bilateral trade. Indeed cross country (size) differences are now captured by country specific effects. Thus, the variation in bilateral trade during the observed period seems exclusively affected by demand pull, while supply push appears virtually nil. Border effect across different country groups Since bilateral trade barriers vary across countries, the average border effect shown above can mask substantial differences across country groups. For instance, zero tariff and zero non-tariff barriers in the EU internal market should imply a significantly higher level of integration for intra-EU countries than for the OECD average. Moreover, a similar pattern can be expected for trade between EU and other OECD countries, on the grounds of their historical trade integration. On the contrary, trade combinations involving CEEC countries would be the less integrated in our sample as these countries only recently joined the

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175

WTO. However, potential differences can also be expected due to the preferential trade regime within CEECs under CEFTA, and their preferential arrangement with EU countries. In order to check for these differences, let us break down the 22 OECD sample into three ‘natural’ country groups: 9 EU, 9 OECD (non EU) and 4 CEEC countries. Thus, we estimate six border coefficients, cm, by means of six different border dummies, dmij , one for each of the following country group combinations: intra-EU, intra-OECD (non-EU) and intra-CEEC trade, and their bilateral combinations EU-OECD, EU-CEEC, and OECD-CEEC, respectively. In this case Eq. (6) will become: X xij ¼ b0 þ hi þ kj þ qð1  rÞ ln d ij þ cm ðdmij Þ þ ð1  rÞeij : m

ð8Þ The results are shown in Table 2, comparing once again the differences between the base model vs. fixed effects specification. The estimated coefficients confirm our a priori expectation, showing substantial differences in the degree of integration. For example, the intra-EU trade has, in all the specifications, a lower (absolute) border coefficient, while the OECD-CEEC trade combination has the highest. However, some inconsistency in the estimated border coefficients can be seen in the base model (columns 1 and 2). For example, the estimated (absolute) border coefficient of intra-CEEC trade is close to (column 1) or lower than (column 2) the border coefficients of EU-OECD and EUCEEC trade, respectively. These results appear at odds with a reality characterized by the CEEC preferential access to the EU market and, especially, the historical higher level of integration between the EU and OECD countries.

171

A very interesting question to understand is what happens if we include the country fixed effects suggested by the theory. This is shown in columns 3 and 4 of Table 2. First, all the border coefficients but one (intra-CEEC) decrease substantially. Second, this reduction is particularly large for intra-OECD and EU-OECD trade, reconciling the inconsistency of the previous findings with common belief. These last results appear to reinforce the importance of using a gravity fixed effects specification suggested by the theory in the estimation of border effect, because the traditional gravity equation not only inflates the border effect estimate, but also tends to produce inconsistency in the estimated level of integration between different trading partners. In the fixed effects specification the level of agricultural trade integration among EU countries is in the same order of magnitude as the previous findings applied to total trade. For example, our results suggest that crossing a national border inside the EU reduces trade by a factor of 6.8 (=exp (1.92)), a figure not so far from the value 4.3 found by Chen (2004) for total manufacture trade in 1996, using a fixed effects specification. It is not surprising to find that the level of EU integration between EU countries is unmatched in the other trade combinations considered here. With a border effect of 9.8, trade between the OECD countries alone seems to have comparatively easier access than intra-EU trade, followed with a factor of 13.1 by EU-OECD trade. Finally, as expected, trade between OECD and CEEC, with a factor of 146, appears to be the most impeded in our sample. Finally, as discussed in Section ‘‘Empirical specification”, in order to account for the potential bias induced by the multilateral resistance terms variation over time

Table 2 Border effects in the EU, OECD, and CEEC countries Model

McCallum (1)

(2)

(3)

(4)

(5)

ln Ci ln Yj ln distanceij Language Contiguity Mills ratio

0.90 (0.03) 0.97 (0.03) 0.90 (0.04) 1.14 (0.09) 0.38 (0.08) 2.06 (1.69)

1 1 0.95 (0.04) 1.14 (0.09) 0.29 (0.08) 0.29 (1.39)

0.72 (0.20) 0.17 (0.18) 1.15 (0.03) 0.05 (0.07) 0.19 (0.07) 8.29 (1.11)

1 1 1.16 (0.03) 0.04 (0.08) 0.18 (0.07) 7.73 (1.11)

1 1 1.14 (0.03) 0.05 (0.08) 0.20 (0.07) 8.74 (1.17)

Border coefficients EU OECD CEECs EU M OECD EU M CEECs OECD M CEECs

3.02 3.74 4.05 4.03 4.60 5.64

2.92 3.60 3.78 3.89 4.43 5.47

1.94 2.30 4.53 2.60 4.16 4.99

1.92 2.28 4.51 2.57 4.14 4.98

1.95 (0.16) 2 32 (0.21) 4.54 (0.18) 2.61 (0.18) 4.17 (0.14) 5.00 (0.18)

Adj. R2 # Observed

0.702 4759

(0.12) (0.18) (0.17) (0.16) (0.13) (0.17)

Fixed effects

0.617 4759

(0.12) (0.18) (0.15) (0.16) (0.12) (0.16)

0.872 4759

(0.15) (0.20) (0.17) (0.17) (0.13) (0.17)

0.835 4759

(0.15) (0.20) (0.17) (0.17) (0.13) (0.17)

0.829 4759

Notes: All columns report the second stage regression results of the Heckman procedure. Columns from (1) to (4) includes a common intercept and a set of yearly fixed effects; column (3) and (4) also include fixed effects for source and destination countries; differently column (5) includes a common intercept and time-varying fixed effects for source and destination countries. Dependent variable: lnXij in regressions (1) and (3); ln(Xij/CjYi) in regressions (2), (4) and (5). Heteroskedasticity robust standard errors are in parentheses.

172

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175 250

17

OECD-CEEC CEEC EU-CEEC EU-OECD OECD EU

16 200

15 14

150

13 12

100

11 10

50

9 0

8 1994

1995

1996

1997

1998

1999

2000

2001

2002

1994

2003

Fig. 1. Time variation in OECD average Border effect. Notes: The figure is based on results of a regression identical to column (6) Table 1, with added interaction between the border indicator and time dummies (see text).

(see Baldwin and Taglioni, 2006), column (5) reports regression with time-varying nation dummies for importer and exporter countries. As is clear from the comparison of the results in columns (5) and (4), this theoretical modification leaves the estimated border coefficients essentially unchanged, suggesting that time series correlation bias does not affect our results at all.

1995

1996

1997

1998

1999

2000

2001

2002

2003

Fig. 2a. Time variation in Border effect across different country group combinations. Notes: The figure is based on results of a regression identical to column (4) Table 2, with added interactions between the border indicators and time dummies (see text).

18

EU-OECD OECD EU

16 14 12 10 8 6 4

Time variation in border effects

2 0

Gravity can only measure trade barriers relative to some benchmark. For instance, in our framework we are comparing trade barriers between countries to barriers within countries. This can be problematic since different countries could have different barriers for internal trade. Moreover, the results could also be sensitive to the measurement of distance among and within countries, as shown by Head and Mayer (2002). However, none of these issues affect the time variation in border effect, that is therefore informative of the evolution in integration and market access. Figs. 1, 2a, 2b analyze the time variation in border effects over the observed period. Following Mayer and Zignago (2005), these figures are obtained through regressions identical to columns (6) and (4) of Tables 1, 2 respectively, with added interaction terms between (each) border indicator with yearly dummies13. This procedure tends to smooth out the evolution of border effects compared to year-by-year estimates, which are more sensitive to outliers. At the OECD aggregate level (Fig. 1), the average border effect decreases from a factor 15 in the 1994 to a factor 12 in 2003. However, after a slight increase in 1996, the border effect shows consistent reduction, especially in the period 1996–1999, after which it essentially appears flat or with few variations until 2003. In general, it is quite difficult to draw strong conclusions from this pattern due to the short time period involved. However, it appears suggestive that the increase in the average integration tends to 13 We also test whether transportation costs change over time, interacting distance with yearly dummies. However, the yearly distance estimate show no trend: the elasticity of trade with respect to distance is 1.27 in 2003, and this is not significantly different form the level in 1994, equal to 1.26.

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

Fig. 2b. Time variation in Border effect across different country group combinations. Notes: The figure is based on results of a regression identical to column (4) Table 2, with added interactions between the border indicators and time dummies (see text).

coincide exactly with the implementation period of the GATT-WTO agricultural agreement. Fig. 2 breaks down the average border effects of our country groups (Fig. 2a) and gives details for the EU, OECD and EU-OECD (Fig. 2b). Overall, the evolution in the border effects for each country combination shows a generalized tendency to decrease, with few exceptions. The increase in the degree of integration changes somewhat across the different country groups, and appears more pronounced in trade combinations that involve CEEC as partners. For instance, the border effect decreases from a factor of 122 to one of 64 for intra-CEEC trade. However, less pronounced is the border effect reduction of EU-CEEC trade, moving from a factor of 69 to a factor of 50. This pattern of trade integration highlights a CEEC preference to trading domestically in line with the CEFTA free trade agreement signed in 1992. Moreover, it also suggests that the process of trade liberalization between the EU and CEEC countries has not yet produced the expected results at the agricultural level. Note, however, that this is consistent with the treatment of many agricultural goods considered ‘sensitive’ products in the process of EU-CEEC trade liberalization, as a consequence of the Community Agricultural Policy. The border effect evolution of intra-EU trade shows a significant and constant reduction in the observed period,

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175

passing from a factor of 8.3 to a factor of 5.6. Also the EUOECD integration level increases significantly until 1999, after which it tends to decrease somewhat. On the contrary, the integration level among the OECD (non-EU) countries increases slightly over the 1994–1998 period, then tends to decrease substantially. Several reasons could be at the root of these patterns. For example, from 1999 onward, the beginning of the European Monetary Union (EMU) could be a principal candidate to explain the significant intra-EU integration process (see Baldwin et al., 2005). Indeed, there is evidence that uncertainty in the exchange rate negatively affects agricultural trade (see Cho et al., 2002). On the other hand, the contrasting pattern between intra-EU and EU-OECD trade after 1999, is not inconsistence with the interpretation based on WTO. Indeed, agricultural agreement could affect market access between EU and OECD but not, at least directly, market access within EU countries.14

Tariff equivalent of border barriers Given the ‘structural’ derivation of our gravity equation, in this final section we will use the estimated border coefficients discussed above to measure their implied ad valorem equivalent (AVE). To this end, remember that our estimated border coefficients are equal to c = (1  r)lnbij (see Eq. (6)), where bij is 1 plus the tariff equivalent of all the trade barriers associated with the national border, and r > 1 is the import substitution elasticity. Thus, the AVE associated with the border could be computed from the following relation: AVE = exp[c/(1  r)]  1. The main problem in converting the border coefficients to their respective AVE is the choice of the import substitution elasticity. Indeed, there is no good guidance on the correct value of r. Moreover, there is no reason to believe that the elasticity of substitution between domestic production and imports from different countries is equal across countries. For example, it is quite hard to believe that the elasticity of France with respect to imports from Italy is equal to that with respect to imports from South Korea or Poland. However, keeping in mind this oversimplification, and following treatment that is quite standard in the literature, we estimated the AVE of border costs using (only) three different elasticity values, spanning from 3 to 7, with a central and preferred value of 5.15 The results from 14

However, note that this potential interpretation contrasts with the important findings of Rose, who did not find any strong GATT-WTO effect on either trade flow or trade policy (see Rose, 2006). 15 We start from the GTAP elasticity value for agricultural products, that typically range from 1.5 to 6. However, several recent papers have shown that GTAP elasticity is probably significantly low (see, Hummels, 2001; Head and Ries, 2001). Specifically, Hummels (2001) elasticity estimates for agri-food products span from about 3 to 8, with an average value of around 4.5. Thus, because we are working with an aggregate at the 2 digit level and with ‘homogeneous’ (agricultural) goods, a range of elasticity from 3 to 7 appears quite reasonable.

173

this exercise are reported in Table 3, borrowing the presentation approach first proposed by Evans (2003). The first two columns of Table 3 report the border effect averaged over 2000–2003, and representative bilateral estimates of tariff rate based on Market Access Map database (see footnote at the bottom of Table 3 for calculation details). The next columns report, for the three different elasticity values, the total implied AVE of the border effects, and the corresponding unexplained AVE, namely the portion of AVE not explained by tariff barriers. Let us focus the discussion on an intermediate value of elasticity of 5. The all-inclusive AVE of the border effects ranges from 56% for intra-EU trade to 236.8% for OECD-CEEC trade. These values are much higher than the results of previous studies based on OECD trade in manufactured goods, ranging from 26% to 116% (see Table 7 of Anderson and van Wincoop, 2004)16. Though these differences are in line with the higher level of protection that typically characterizes agricultural vs. manufactured goods in OECD countries, the implied AVE of the border effects appears far higher than recognized trade barriers, suggesting that transaction costs and non-distortionary border barriers are substantial. Note that this conclusion holds true especially for trade combinations involving the less developed CEEC, where the unexplained part of the AVE, for the same elasticity value, is always greater than 140%. On the other hand, for trade involving intra-EU, OECD (non-EU) and EU-OECD countries, the unexplained AVE, ranging from 56% to 70%, appears more reasonable, and of the same order of magnitude as recent estimates of the tariff equivalent of NTB. For example, de Frahan and Vancauteren (2006) found a tariff equivalent of the costs of intraEU food regulations of about 180% in the nineties. Moreover, Messerlin (2001) showed that the tariff equivalents of NTB for the EU in 1999 was around 100% for many agricultural products. Finally, Kee et al. (2004) showed that a theoretical consistent estimate of the tariff equivalent of NTB in agriculture is equal to 33% for the EU and 17% for the US. Thus, our figures are not so far out when compared to the actual range of direct (overall) protection measures. Obviously, as it is clear from the figures reported in Table 3, the results are sensitive to the value of substitution elasticity. Increasing (or decreasing) the substitution elasticity between home and foreign goods significantly decreases (increases) the estimated AVE implied by border effects. The intuition behind this is quite simple: the greater the elasticity, the smaller the necessary domestic-foreign price gaps, induced by protection, to have consumers switch to domestic products. Summing up, the exercise reported in this section suggests the following main considerations, conditional to

16 Note, however, that these studies never include CEEC countries in the OECD sample.

174

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175

Table 3 Border effects, implied AVEs and bilateral tariffs (2000–2003) Country group

EU(intra) OECD (non-EU) EU-OECD EU-CEEC CEEC OECD-CEEC

Border effect average 2000–2003

5.9 10.0 12.6 50.9 77.2 130.5

Weightedbilateral tariff (%)

0.0 19.1 18.3 22.4 27.1 25.5

Import substitution elasticity and AVE (%) Elasticity = 3

Elasticity = 5

Elasticity = 7

Total implied AVE

Unexplained AVE

Total implied AVE

Unexplained AVE

Total implied AVE

Unexplained AVE

143.4 215.8 254.4 612.9 776.0 1034.6

143.4 196.7 236.1 590.5 748.9 1009.1

56.0 77.7 88.2 167.0 196.0 236.8

56.0 58.6 69.9 144.6 168.9 211.3

34.5 46.7 52.5 92.5 106.1 124.7

34.5 27.6 34.2 70.1 79.0 99.2

Notes: The first column shows the average 2000–2003 border effects estimated from a regression identical to specification 4 (Table 2), with interaction term between each border dummy and yearly dummies. The second column reports average bilateral tariffs measured as follows. Market Access Map (www.CEPII.org) provides, for 2001, the country ‘bilateral’ weighted tariffs for all but one country (Norway) covered by our analysis. Starting from these figures we created a single tariff for each country group to reflect the average bilateral tariff protectionist of each combination, weighting each country (or group) average tariff by the respective shares of their exports in agricultural bilateral trade. Thus, for instance, if most trade flows from the EU to CEEC countries, then the average CEEC tariff receives a greater weight in the average bilateral tariff. The next columns show the total implied AVEs of border effects and the corresponding unexplained AVEs for three different elasticity values (see text).

the (unknown) true value of elasticity. First, the policyunrelated component of border effect is important. Second, this component tends to decrease with the level of country development, showing that for intermediate and moderately high levels of substitution elasticity the AVE implied by the border effects does not appear to be so far from the actual estimate of the (overall) market access restriction faced by rich countries. Concluding remarks This paper uses a gravity model to investigate the level of trade integration among different OECD country ‘blocks’ through the border effect approach, that is to say, through all the factors that contribute to a country’s internal trade volume deviation from the gravity model prediction. The analysis strongly confirms that a proper derivation of the gravity equation from theory matters in estimating border effects. In fact, the traditional gravity equation not only inflates the border effect estimate, but also produces potential inconsistency in the estimated level of integration among different trading partners. Overall, these results are in line with the most recent literature on gravity models. A representative estimate of the average border effect in OECD markets, after controlling for economic size, transport costs, and any other non specified trade costs, lies in the range from 8.9 to 13.2. This means that crossing a national border into the OECD induces a trade-reduction effect of the same order of magnitude. However, this average border effect masks substantial differences across country groups. Not surprisingly, the level of integration between EU countries, with an average border effect of 6.8, is unmatched by other trade combinations. Instead, trade between OECD and CEEC, with a border effect of 146, appears the most impeded in our sample. The magni-

tude of these border costs are higher than those estimated at the manufacturing level. Thus, while factors such as the perishability of agricultural products and the existence of CAP among EU countries could partially explain these border differences, a deeper explanation of such costs could be an important topic for further research. In the observed period the level of integration showed a general tendency to increase, and this was particularly strong for trade combinations involving CEEC as a partner. Quite surprisingly, the intra-CEEC integration process in agriculture was twice as strong as that in the EU-CEEC trade. Thus, the process of trade liberalization between EU and CEEC has not yet produced the expected results at the agricultural level. Moreover, the time pattern of trade integration is not inconsistent with the GATT-WTO agreement. Finally, given the structural derivation of our gravity equation, and a reasonable assumption of elasticity of substitution, we calculated the overall ad valorem equivalent tariffs implied by estimated border effects, and then compared them with actual tariff and NTB measurements. The overall picture emerging from this exercise suggests that the components of border effects not related to trade policy are important. However, this situation strongly decreases with the level of country development, showing that for intermediate and high elasticity values, the ad valorem equivalent of border effects are not so far from the actual estimate of trade policy restriction faced by rich countries. References Anderson, J.E., 1979. A theoretical foundation for the gravity equation. American Economic Review 69 (1), 106. Anderson, J.E., van Wincoop, E., 2002. Borders, trade and welfare. In: Collins, S., Rodrik, D. (Eds.), Brookings Trade Forum 2001. The Brooking Institution, Washington, pp. 207–244. Anderson, J.E., van Wincoop, E., 2003. Gravity with gravitas: a solution to the border puzzle. American Economic Review 93, 170–192.

A. Olper, V. Raimondi / Food Policy 33 (2008) 165–175 Anderson, J.E., van Wincoop, E., 2004. Trade costs. Journal of Economic Literature XLII (3), 691–751. Baier, S., Bergstrand, J.H., 2001. The growth of world trade: tariffs, transport costs, and income similarity. Journal of International Economics 53, 1–27. Baldwin, R., Taglioni, D. 2006. Gravity for dummies and dummies for gravity equations. National Bureau of Economic Research, Working Paper 12516/September. Baldwin, R. Skudelny, F., Taglioni, D., 2005. The effects of Euro. Evidence from sectoral data. European Central Bank, Working Paper Series, no. 446/ February 2005. Bergstrand, J.H., 1989. The generalized gravity equation, monopolistic competition and the factor-proportion theory in international trade. Review of Economics Statistics 71 (1), 143–153. Bureau, J.C., Salvatici, L., 2004. WTO negotiation on market access: What we know, what we don’t and what we should. In: Anania, G., Bhoman, M., Carter, C., McCalla, A. (Eds.), Agricultural Policy Reform and the WTO. Where Are We Heading? Edward Elgar, Cheltenam, UK. Chen, N., 2004. Intra-national versus international trade in the European Union: Why do national border matter? Journal of International Economics 63 (1), 93–118. Cho, G., Sheldon, I.M., McCorriston, S., 2002. Exchange rate uncertainty and agricultural trade. American Journal of Agricultural Economics 84 (4), 931–942. Deardoff, A.V., 1998. Determinants of bilateral trade: does gravity work in a Neo-classical world? In: Frankel, J.A. (Ed.), The Regionalization of the World Economy. VCH, Berlin, pp. 7–32. de Frahan, B.H., Vancauteren, M., 2006. Harmonization of food regulations and trade in the single market: evidence from disaggregated data, In: Paper Presented at the IATRC Summer Symposium ‘Food Regulation and Trade: Institutional Framework, Concepts of Analysis and Empirical Evidence’, Bonn, Germany, May 28–30. Eaton, G., Kortum, S., 2002. Technology, geography and trade. Econometrica 70 (5), 1241–1279. Estevadeordal, E., Frantz, B., Taylor, A.M., 2003. The rise and fall of the world trade, 1870–1939. Quarterly Journal of Economics CXVIII (2), 359–407. Evans, C.L., 2003. The economic significance of national border effects. American Economic Review 93 (4), 1291–1312. Feenstra, R., 2002. Border effects and gravity equation: consistent methods for estimation. Scottish Journal of Political Economy 49 (5), 491–506. Feenstra, R., Markusen, J., Rose, A., 2001. Using the gravity equation to differentiate among alternative theories of trade. Canadian Journal of Economies 34 (2), 430–447. Fontagne´, L., Mayer, T., Zignago, S., 2005. Trade in the triad: how easy is the access to large markets? Working Paper 2004-04, Centre d’Etudes Prospectives et d’Informations Internationales (CEPII). Paris: CEPII. Furtan, W.H., van Melle, B.M., 2004. Canada agricultural trade in North America: Do national borders matter? Review of Agricultural Economics 26 (3), 317–331.

175

Gallezot, J., 2003. Real access to the EU’s agricultural market. Mimeo, UMR Economie publique INRA-INAPG, Institute National de la Recherche Agronomique, Paris. Haveman, J., Thursby, J.G., 2002. The impact of tariff and non-tariff barrier to trade in agricultural commodities: a disaggregated approach. Mimeo, Purdue University. Head, K., Mayer, T., 2000. Non-Europe: the magnitude and causes of Marlet fragmentation in Europe. Weltwirtschafliches Archiv 136 (2), 285–314. Head, K., Mayer, T., 2002. Illusory border effects: distance mismeasurement inflates estimates of home bias in trade. Working Paper, Centre for Economic Policy Research, CEPR, London. Head, K., Ries, J., 2001. Increasing return vs. national product differentiation as an explaination for the pattern of US-Canada trade. American Economic Review 91 (4), 848–876. Helliwell, J.F., 1997. National borders, trade and migration. Pacific Economic Review 2 (3), 165–185. Helpman, H., Melitz, M., Rubinstain, Y., 2004. Trading partners and trading volumes, unpublished manuscript, Harvard University. Hummels, D., 2001. Toward a geography of trade costs, GTAP Working Paper 17, Purdue University. Kee, H.L., Nicita A., Olarreaga, M., 2004. Ad-Valorem Equivalents of Non-Tariff Barriers. Working Paper. Development Research Group, The World Bank. WB, Washington, DC. Mayer, T., Zignago, S., 2005. Market access in global and regional trade. Working Paper 2005-02. Centre d’Etudes Prospectives et d’Informations Internationales (CEPII). CEPII, Paris. McCallum, J., 1995. National Borders matter: Canada–U.S. regional trade patterns. American Economic Review 85 (3), 615–623. Messerlin, P., 2001. Measuring the costs of protection in Europe: European commercial policy in the 2000s. Institute of International Economics, Washington, DC. Otsuki, T., Wilson, J.S., 2001. Global trade and food safety: winners and losers in a fragmented system, Working Paper 2689, Development Research Group. The World Bank, Washington, DC. Otsuki, T., Wilson, J.S., Sewadeh, M., 2001. What price precaution? European harmonisation of aflatoxin regulations and African groundnut exports. European Review of Agricultural Economics 28, 263–284. Rose, A.K., 2006. The effect of membership in the GATT-WTO on trade: Where do we stand? Working Paper, University of California, Berkeley. Santos Silva, J.M.C., Tenreyro, S., 2006. The log of gravity. The Review of Economics and Statistics 88 (4), 641–658. Wang, Z., Coyle, W.T., Gehlhar, M., Vollrath, T., 2000. The impact of distance on U.S. agricultural exports. An econometric analysis. In: Coyle W., Ballanger N. (Eds.) Technological Changes in the Transportation Sector – Effects on U.S Food and Agricultural trade: A Proceedings. USDA, pp. 71–83. Wei, S-J., 1996. Intra-national versus international trade: how stubborn are nations in global integration? Working Paper 5531. Cambridge, MA: National Bureau of Economics Research.