Do Currency Barriers Solve the Border Effect Puzzle? Evidence from the CFA Franc Zone José de Sousa and Julie Lochard ROSES, University of Paris 1 Abstract: In this paper, we use a gravity model to investigate the extent to which currency barriers explain the border effect puzzle, i.e. the impact of national borders on international trade. We focus on the two monetary unions of the CFA Franc Zone in West and Central Africa. We find that these countries display large border effects, and that currency barriers explain between 17 per cent and 28 per cent of the overall border effect. JEL no. F11, F15, F33 Keywords: Borders; gravity; currency unions

1 Introduction A promising research agenda, measuring the impact of national borders in international trade, has been initiated with the work of McCallum (1995). He was among the first to emphasize the existence of a border effect puzzle, one of the six major puzzles in international macroeconomics (Obstfeld and Rogoff 2000). After controlling for differences in size and distance, McCallum (1995) finds that, in 1988, a Canadian province trades about 20 times more with another province than with an American state. This large home bias1 is puzzling since the two countries are similar in terms of culture, institutions and language. Moreover, despite the implementation of the US-Canada Free Trade Agreement in 1989, the border effect persisted (Helliwell 1996; Hillberry 1998). Besides, other regions of the world were found to experience similar border effects (Helliwell 1998; Wei 1996). Remark: We would like to thank an anonymous referee as well as Ansgar Belke, C´eline Carr`ere, G´erard Duchˆene, Joˇze Damijan, Anne-C´elia Disdier, Patrick Guillaumont, Thierry Mayer, Daniel Mirza, Jacques M´elitz, Volker Nitsch, Caroline Vincensini and seminar participants at the University of Paris 1, the University of Clermont-Ferrand, the University of Ljubljana, the CESifo 2003 Venice Summer Institute and the ETSG 2003 Conference for helpful comments and suggestions. Please address correspondence to Julie Lochard, ROSES, University of Paris 1 Panth´eon Sorbonne, 106–112 Bd de l’hôpital, 75647 Paris Cedex 13; e-mail: [email protected]. 1 Subsequently, the terms “home bias” and “border effect” will be used synonymously. © 2005 Kiel Institute for World Economics

DOI: 10.1007/s10290-005-0037-5

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

423

The literature provides various explanations of the border effect puzzle: misspecifications, formal and informal trade barriers as well as national preferences. Another potential explanation stems from the existence of various currencies. Since a nation is by definition a monetary union2 , crossing the border implies the use of another currency, which may therefore entail transaction costs, such as currency conversion costs, hedging costs due to exchange-rate variability, and in-house costs of maintaining separate foreign currency expertise.3 These costs impede international trade and favour intra-national flows. In this paper, we intend to measure to what extent the existence of separate currencies explains the border effect puzzle. Related papers argued that taking into account exchange rate variability helps to explain the magnitude of the border effect (Parsley and Wei 2001). However, the evidence appears rather mixed.4 But after all, sharing a common currency is a much more durable commitment than a fixed exchange rate regime and this one “might yield only some of the economic benefit of a monetary union, while implying most of the costs” (Gros and Thygesen 1998: 264). In a study that compares the effect of these two monetary regimes on international trade, Rose (2000) supports this view and finds that a fixed exchange rate has a negligible impact on trade, while a monetary union increases trade by a factor of three. Consequently, compared to exchange rate variability, the use of different currencies may explain a larger part of the home bias. To deal with this issue, we use the theoretical gravity model developed by Anderson and van Wincoop (2003). Then, we redefine the trade cost factor to account for both the border effect and the currency union effect. This model is helpful since it can be adapted for empirical estimations. Using bilateral trade data on CFA countries and their European partners from 1980 through 1999, we empirically test the relationships between 2

A monetary union (or currency union) is defined by a single currency, a single central bank and a single monetary policy. 3 Transaction cost savings following the adoption of the euro were estimated between 0.4 per cent (Emerson et al. 1992) and 1 per cent of the GDP of the 12 members of the European Community (Gros and Thygesen 1998). 4 Wei (1996) introduces several variables for exchange rate volatility to test whether this uncertainty causes a home bias, but he finds that the estimated home bias coefficient is not much affected by the inclusion of volatility. However, in another contribution, Parsley and Wei (2001) conclude that exchange rate variability together with distance and unitshipping costs explain a substantial part of the border effect between the United States and Japan.

424

Review of World Economics 2005, Vol. 141 (3)

currency unions and border effects.5 The CFA Franc Zone is a relevant zone for our purpose. First, there is very little literature estimating the border effect for less developed countries and particularly for this area.6 Second, until the launch of the European Economic and Monetary Union in 1999, this zone was the only long-lasting reference in terms of currency union.7 Third, this zone comprises two separate monetary unions: the West African Economic and Monetary Union (WAEMU) and the Central African Economic and Monetary Community (CAEMC).8 This particular feature allows us to compare intra-currency unions and extra-currency unions trade relationships. We find that CFA countries display large border effects and currency barriers explain between 17 and 28 per cent of the overall border effect. The remainder of the paper is organized as follows. In Section 2, we describe the theoretical setting and model the trade cost factor to take into account both the currency union effect and the border effect. In Section 3, we deal with the estimation strategy and focus on the measurement of intranational distances and relative prices. Then, we present the data in Section 4, the results in Section 5 and some sensitivity tests in Section 6. In the final section, we conclude. 2 Model The first attempts to derive theoretical foundations for the gravity equation date back to Anderson (1979) and Bergstrand (1985). Subsequently, various models stemming from different general frameworks have been developed, such as monopolistic competition or product differentiation models. We 5

Trade data for African countries are available for a larger time period (see e.g. Nitsch 2005). However, we also need output data to compute border effects, which is more difficult to find before the eighties (see Appendix B for details). 6 One exception is Helliwell (1998). 7 Taking into account this feature, Nitsch (2005) explores the trade effect of the gradual enlargement of the CFA franc zone. 8 These two monetary unions have two different central banks and two different currencies with the same acronym, CFA Franc. The meaning is different however: Franc of the Communaute´ Financie`re d’Afrique (African Financial Community) for WAEMU countries, and Franc of the Coope´ration Financie`re en Afrique Centrale (Financial Cooperation in Central Africa) for CAEMC countries. The former is issued by the Banque Centrale d’Afrique de l’Ouest (BCEAO) located in Dakar (Senegal) and the latter, corresponding to CAEMC countries, is issued by the Banque des Etats de l’Afrique Centrale (BEAC) located in Yaound´e (Cameroon).

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

425

follow Deardorff (1998) and Anderson and van Wincoop (2003) and assume that each country is specialized in a single good. The model starts from the utility of a representative consumer in country i (Ui ), which is assumed to be:
  (1−σ)/σ (σ−1)/σ σ/(σ−1) Ui = βj cij , (1) j

where cij is the consumption by consumers of country i for goods from country j, σ > 1 the elasticity of substitution between goods and βj a positive distribution parameter. The consumers in country i maximize their utility subject to the following budget constraint:  (2) Yi = pij cij , j

with Yi the income of country i and pij the price of good j for consumers in i.9 Assuming that trade costs are borne by sellers and take the “iceberg” form,10 we redefine pij as: pij = tij pj ,

(3)

where pj is the price received by sellers and tij the trade cost factor (see below). The result of the consumers’ utility maximization problem is:
 βj pij 1−σ 1 , (4) cij = Yi pij Pi with


Pi =

 (βj pij )1−σ

1/(1−σ) .

(5)

j

Therefore, the value of imports of i from j (Mij ) is given by:
 βj pj tij 1−σ . Mij = Yi Pi 9

(6)

Recall that i and j can be countries as well as goods since we assume that each country is specialized in a single good. 10 Assuming an iceberg form for trade costs amounts to suppose that for each good transported from country j to country i a proportion “melts” in transit.

426

Review of World Economics 2005, Vol. 141 (3)

Finally, using a market clearing condition, we obtain (see Appendix A for details):
 Yi Yj tij 1−σ Mij = , (7) Yw Pi Pj with Yw the world income, θj country j’s share of world income and
1−σ tij  1−σ . Pi = θj Pj j

(8)

As in traditional gravity equations, the value of trade depends positively on the economic size of each country and negatively on a trade barrier factor (tij ). But trade is also affected by the price indexes of both countries. Recalling that σ > 1, Anderson and van Wincoop (2003) interpret them as “multilateral resistance” variables. This interpretation is quite intuitive: “the more resistant to trade with all others a region is, the more it is pushed to trade with a given bilateral partner” (Anderson and van Wincoop 2003: 3). The next step is to model trade costs. We assume that the trade cost factor consists of three terms corresponding to two different types of costs, non-border costs (d) and border costs (b and cb): ρ

tij = dij cbij bij .

(9)

The first term (dijρ ) relates to non-border costs. It is usually proxied by the distance between two countries. The following terms relate to border costs: • (cbij ) is the currency barrier resulting from the use of different currencies in trade relationships. In a more formal way, cbij = cbncuij with ncuij = 1 if i and j do not use the same currency and 0 otherwise. We expect a negative coefficient for this variable, capturing the effect of currency barriers on trade. • (bij ) is the net border effect which measures all impediments to trade related to the national border, except currency barriers, i.e. the use of different national currencies. Formally, bij = bbeij with beij = 1 for international trade (i
= j) and 0 for intra-national trade (i = j). We expect again a negative coefficient for this variable, reflecting the impact of border barriers on trade. This specification allows us to measure the impact of national borders as well as the effect of currency barriers on trade.

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

427

Transforming (7) in log terms and replacing the trade cost factor with (9) yields:
 Mij (10) ln = a0 + a1 ln(dij ) + a2 (ncuij ) + a3 (beij ) Yi Yj − (1 − σ) ln(Pi ) − (1 − σ) ln(Pj ), where the coefficients to be estimated are: a1 = (1 − σ)ρ, a2 = (1 − σ) ln(cb), and a3 = (1 − σ) ln(b). Once we have estimated a3 , we can compute the ad valorem tariff equivalent of the border barrier (τb ). Since b = 1 + τb , we obtain τb = b − 1 = exp(a3 /(1 − σ)) − 1.

3 Estimation Strategy The estimation of (10) raises some methodological problems related to the measurement of “intra-national” observations (in trade and distance) and relative prices (Pi and Pj ). 3.1 How to Measure Intra-National Observations? The border effect approach has been first applied to measure the effect of borders in North American trade (McCallum 1995). Its extension to other countries generates two major problems. First, while we do have intra-national trade flows for the United States and Canada (resp. interstate and inter-provincial trade flows), equivalent data are unavailable for other countries. An intuitive solution to measure the “intra-national trade” of a country consists in subtracting its total exports from its total national output (Wei 1996; see Appendix B for details). The second problem lies in the measurement of intra-national distances. In the core of the paper, we follow the seminal work of Wei (1996) and measure the intra-national distance as one-quarter of the distance to the nearest neighbour (see Appendix B). However, other approaches have been suggested and used (see robustness checks in Section 6). 3.2 How to Measure Relative Prices? Equation (10) includes two relative price terms (Pi and Pj ). In order to take account of these unobserved price indexes, Anderson and van Wincoop

428

Review of World Economics 2005, Vol. 141 (3)

(2003) suggest to estimate the gravity equation with non-linear least squares or alternatively to replace relative price terms with country-specific dummies.11 Both methods lead to consistent estimates, and while the former is more efficient, its “benefit seems to be relatively small compared to the computational simplicity of the fixed effect approach” (Feenstra 2004). Accordingly, to take account of the price indexes, we introduce in (10) fixed effects for both destination and source countries. Finally, we estimate: 
Mij (11) = a0 + a1 ln(dij ) + a2 (ncuij ) + a3 (beij ) ln Yi Yj + a4 φi + a5 φj + νt It + εij , where i denotes the destination country and j the source country. The variables are defined as follows: • • • • • •

• •

11

Mij is the value of bilateral imports between i and j; Y is the GDP; dij denotes the distance between i and j; ncuij is a dummy variable equal to one if i and j do not use the same currency and zero otherwise; beij is a dummy variable equal to one for international trade (i
= j) and zero for intra-national trade (i = j); φi and φj are country fixed effects respectively for i and j. The coefficients a4 and a5 are supposed to estimate the multilateral resistance terms: a4 = (σ − 1) ln(Pi ) and a5 = (σ − 1) ln(Pj ); It is a vector of year dummies that takes a value of one in year t for t = 1, ..., T; εij is an error term, reflecting measurement error in trade.

Another attempt to approximate these price variables consists in using GDP deflators (Bergstrand 1985). However, making use of available data on price indexes may introduce a bias, as all trade costs are not pecuniary costs. An alternative approach amounts to proxy these price terms with remoteness variables representing bilateral distances relative to distance with all trading partners (see e.g. Wei 1996). Nevertheless, the inclusion of remoteness indexes in gravity equations is “atheoretic” and contributes to a disconnection between the theory and the functional form of the gravity model (Anderson and van Wincoop 2003).

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

429

4 Data The database includes 11 CFA countries12 (7 belonging to the WAEMU: Benin, Burkina Faso, Ivory coast, Mali, Niger, Senegal, Togo; 4 to the CAEMC: Cameroon, Chad, Congo and Gabon) and 14 European countries.13 Our trade variable (Mij ) is the value of imports of each CFA country from all other countries j on the period 1980–1999, plus its imports from itself i.e. its intra-national trade. This represents a potential of 11 × 25 × 20 = 5,500 observations.14 5 Results All estimations are carried out using ordinary least squares (OLS) and heteroscedasticity is corrected with White’s method. Following our model, we constrain the coefficients on GDP terms at unity.15 In this way, we control for a potential problem of simultaneity between income and trade. Indeed, there are good reasons to suspect that income is endogenous, i.e. influenced by the level of trade. We report only the results using Wei’s measure of distance with a regional breakdown (see below and Appendix B). Other methods for measuring bilateral and intra-national distances do not alter our main results (see robustness checks in Section 6). Table 1 provides the results of two different specifications. Column (1) reports the estimation of the impact on trade of the average currency barrier and the average border effect. As expected, the coefficient on the “average currency barrier” variable is negative and highly significant (p < 0.01). It highlights the influence of currency barriers on CFA trade.16 The coefficient on the “average border effect” variable is also negative and 12

The CFA zone consists currently of 14 countries, but we do not retain Guinea-Bissau, Equatorial Guinea and Central African Republic because of too many missing observations. 13 Austria, Belgium-Luxembourg, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Portugal, Spain, Sweden, the United Kingdom. 14 However, some observations are missing. These missing values range from 6 to 29 per cent of total observations depending on the year considered. See Appendix C. 15 Our main results are robust to the estimation of non-unitary income elasticities. Full results are available upon request. 16 Our “currency barrier effect” is different from the “Rose effect” (see Rose 2000), not only because we estimate the negative impact of using different currencies, but also because our estimation is computed with intra-national trade observations. If we drop these

430

Review of World Economics 2005, Vol. 141 (3)

Table 1: Average Border Effect and Currency Barrier for the CFA Franc Zone

Distance Average currency barrier Average border effect WAEMU currency barrier WAEMU border effect CAEMC currency barrier CAEMC border effect Country fixed effects Year dummies R-sq Number of observations

(1)

(2)

−0.52∗ (0.10) −1.70∗ (0.13) −6.35∗ (0.29)

−0.47∗ (0.10)

Yes Yes 0.71 4659

−1.27∗ −6.29∗ −2.56∗ −6.53∗

(0.14) (0.32) (0.28) (0.40)

Yes Yes 0.71 4659

∗ significant

at the 1 per cent level. Note: Dependent variable: Log of imports value divided by the GDP of importers and exporters. Robust standard errors in parentheses. Constant, fixed effects and year dummies are not reported. In column (1), we estimate (11). In column (2), we estimate: ln (Mij /Yi Yj ) = a0 + a1 ln(dij ) + a2.1 (ncu(waemu)ij ) + a2.2 (ncu(caemc)ij ) + a3.1 (be(waemu)ij ) + a3.2 (be(caemc)ij ) + a4 φi + a5 φj + νt It + εij ,

where ncu(waemu)ij is a dummy variable equal to one if i and j do not use the WAEMU’s currency; be(waemu)ij is a dummy variable equal to one for international trade and zero for intra-national trade of WAEMU countries; ncu(caemc)ij and be(caemc)ij are similar dummies for CAEMC countries.

significant. It is interpreted as the estimate of the net average impact of national border barriers (excluding currency barriers).17 For the sake of illustration, we calculate the ad valorem tariff equivalent of the estimated observations from the sample, we can estimate a variant of (11):   ln Mij /Yi Yj = a0 + a1 ln(dij ) + a2 (cuij ) + a3 φi + a4 φj + νt It + εij , where (cuij ) is a dummy variable equal to one if i and j use the same currency. We find that the coefficient of the currency union dummy is 1.36 (p < 0.01) (overall results are available upon request). See also Rose (2004) for a stimulating meta-analysis of the impact of a common currency on trade. 17 As highlighted by Helliwell (1998: 46): “interpretation of national border effects becomes more complicated with the addition of other variables affecting the international flows of goods among subsets of countries [...] After qualitative variable relating to some bilateral trade flows have been added, the border effects for a given country depend on which trading partners are being considered”. Thus, in column (1), the “net average border effect” coefficient measures the border effect of a CFA country involved in a trade relation-

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

431

border effect. As discussed above, the border effect coefficient is the product of the substitution elasticity and the log of the tariff equivalent. Suppose an intermediate degree of substitutability between goods (σ = 10), then the estimated coefficient of the average border effect (−6.35) implies a tariff equivalent of about 102 per cent [= exp(−6.35/(1 − 10)) − 1]. The overall border barrier amounts to −8.05 [= −6.35 − 1.70]. Thus, here, currency barriers explains 21 per cent [= (−1.70)/(−8.05)] of the overall border effect. In column (2), we compute the border effect and the currency barrier effect for the two currency unions separately. We estimate the same equation as below (11) but we differentiate each dummy depending on whether it concerns WAEMU or CAEMC countries. CAEMC countries exhibit a larger border effect than WAEMU countries, but the main difference lies in the currency barrier coefficient. For CAEMC countries, the existence of different currencies seems to largely impede trade. Currency barriers explain 28 per cent [= (−2.56)/(−6.53 − 2.56)] of the overall barrier to trade, while it explains only 17 per cent [= (−1.27)/ (−6.29 − 1.27)] of the border effect in the case of WAEMU countries. To sum up, Table 1 highlights three results. First, a border effect exists. Other things being equal, a CFA country trades more with itself than with another country. Second, currency barriers matter. Other things being equal, the border effect of two countries using different currencies is higher than the border effect of two countries sharing the same currency. However, the third result is that currency barriers do not solve the border effect puzzle. Currency barriers explain between 17 and 28 per cent of the overall border effect, depending on the currency union. Consequently, a puzzling issue remains: why is the border effect so large? Our estimations are in line with existing estimates on other developing countries (Helliwell 1998: 55–56). The consensus is that the size of the border effect is larger for less developed countries than for OECD countries. For example, Zimbabwe is found to trade around two hundred times more with itself than with other countries (Helliwell 1998: 55–56). This empirical evidence can be explained by several arguments. First of all, large border effects can reflect the specialization of the supply structure. For instance, the specialization of CFA countries in primary goods (which are mostly non-traded intra-regionally) will probably be translated into ship with a country sharing the same currency, i.e. the net border effect, besides currency issues.

432

Review of World Economics 2005, Vol. 141 (3)

high border effects. Moreover, the existence of unrecorded trade which is considered to be large for CFA countries can increase border effects in lowering official international trade. We try to control for this factor in robustness checks (see Section 6). Finally, a large border effect can be explained by the existence of non-monetary barriers: remaining tariff and non-tariff barriers coming from an inability and/or unwillingness to carry out trade liberalization measures, a very poor transportation infrastructure, low economic potentials, ethnic, cultural or linguistic diversity, and very high political instability (IMF 2002).

6 Robustness Checks 6.1 Evolution of Border Effects Are our results robust across sub-periods? We try to answer to this question by dividing our sample into two sub-periods: before 1990 and after 1990. The results by sub-periods are presented in Table 2. These estimates are in line with those of Table 1. It is noteworthy that the average border effect of a CFA country has increased during the overall period (columns 1 and 2). It means that a CFA country trades more and more with itself than with other countries. On the other hand, the coefficient on the currency barrier variable has decreased, suggesting that the impact of different currencies on trade has lowered. Currency barriers explain 25 per cent [(−1.94)/(−5.81 − 1.94)] of the overall barrier to trade for the period 1980–1989, and 17 per cent for 1990–1999. These conclusions hold for both WAEMU and CAEMC countries (columns 3 and 4), the border effect increase being particularly remarkable in the case of the CAEMC. 6.2 Informal Trade We perform some sensitivity tests to account for informal trade, which is commonly defined as transactions outside official channels. These unrecorded trade flows concern both legal and illegal goods. There seems to be a consensus on the importance of informal trade flows in CFA countries, while they are by definition difficult to evaluate. For instance, Benin is known for being a platform of re-export. It legally imports merchandises from Europe or Asia and exports them fraudulently to neighbouring countries, especially Nigeria.

433

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

Table 2: Evolution of Average Border Effects and Currency Barriers

Distance Average currency barrier Average border effect WAEMU currency barrier WAEMU border effect CAEMC currency barrier CAEMC border effect Country fixed effects Year dummies R-sq Number of observations

1980–1989 (1)

1990–1999 (2)

1980–1989 (3)

1990–1999 (4)

−0.61∗ (0.14) −1.94∗ (0.20) −5.81∗ (0.39)

−0.45∗ (0.14) −1.44∗ (0.18) −6.79∗ (0.42)

−0.55∗ (0.14)

−0.44∗ (0.14)

−1.59∗ −5.92∗ −2.82∗ −5.75∗

−1.08∗ −6.54∗ −1.99∗ −7.23∗

Yes Yes 0.71 2228

Yes Yes 0.73 2431

(0.22) (0.43) (0.42) (0.52)

Yes Yes 0.71 2228

(0.18) (0.47) (0.37) (0.56)

Yes Yes 0.73 2431

∗ significant

at the 1 per cent level. Note: Dependent variable: Log of imports value divided by the GDP of importers and exporters. Robust standard errors in parentheses. Constant, fixed effects and year dummies are not reported. In columns (1) and (2), we estimate (11). In columns (3) and (4), we estimate: ln (Mij /Yi Yj ) = a0 + a1 ln(dij ) + a2.1 (ncu(waemu)ij ) + a2.2 (ncu(caemc)ij ) + a3.1 (be(waemu)ij ) + a3.2 (be(caemc)ij ) + a4 φi + a5 φj + νt It + εij , where ncu(waemu)ij is a dummy variable equal to one if i and j do not use the WAEMU’s currency; be(waemu)ij is a dummy variable equal to one for international trade and zero for intra-national trade of WAEMU countries; ncu(caemc)ij and be(caemc)ij are similar dummies for CAEMC countries.

Unrecorded trade flows are expected to modify the size of the border effect, as it can generate an underestimation of international trade flows and lead to an overestimation of intra-national trade.18 Both mismeasurements are likely to decrease the size of the border effect. Accordingly, in Table 3, we perform two kinds of tests. Firstly, we assume that CFA total exports are underestimated. This assumption is likely to affect the size of the border effect via the measure of intra-national trade. Since CAEMC exports consist mainly in oil products,19 we suppose that they are less underestimated than 18

Recall that intra-national trade is calculated taking total production minus total exports. Hence, if total exports are underestimated and production is well-measured, intranational trade will be overestimated. 19 The composition of CAEMC exports is highly biased toward oil products. For example, in Gabon, oil products represent 75 per cent of total exports in 1999.

434

Review of World Economics 2005, Vol. 141 (3)

Table 3: Sensitivity Tests Related to Informal Trade (1)

(2)

(3)

Distance Average currency barrier Average border effect

−0.51∗ (0.10) −1.74∗ (0.13) −6.15∗ (0.30)

−0.52∗ (0.10) −1.70∗ (0.13) −5.65∗ (0.29)

−0.52∗ (0.10) −1.70∗ (0.13) −4.96∗ (0.29)

Country fixed effects Year dummies R-sq Number of observations

Yes Yes 0.71 4659

Yes Yes 0.73 4659

Yes Yes 0.75 4659

∗ significant

at the 1 per cent level. Note: Dependent variable: Log of imports value divided by the GDP of importers and exporters. Standard errors in parentheses. Regressions (1) to (3) report the results of estimating (11). Constant, fixed effects and year dummies are not reported. We deal with the problem of informal trade by multiplying total exports of WAEMU countries by 1.5 in column (1); multiplying bilateral intra-CFA zone imports by 2 in column (2); multiplying bilateral intra-CFA zone imports by 4 in column (3).

WAEMU exports, and we simply increase exports of WAEMU countries.20 Results for this first test are given in column (1). Secondly, various reports point out that effective trade flows in West and Central African countries are at least twice the official flows. In order to deal with this issue, we multiply bilateral imports (the dependent variable) by 2 and then by 4. Columns (2) and (3) respectively show the results of this second assumption.21 Our main findings do not appear to be sensitive to unrecorded trade. In the first regression (column 1), the average border effect is almost unchanged compared to column 1 of Table 1. When modifying bilateral trade flows (columns 2 and 3), the average border effect is as expected significantly lower [resp. −5.65 and −4.96 vs −6.35 (column 1 of Table 1)]. Thus, the border effect is smaller when accounting for informal trade flows and currency barriers still explain between 22 (column 1) and 25 per cent (column 3) of the overall border effect. 6.3 Distance Measurements Distance mismeasurements can inflate border effect estimates (Head and Mayer 2002). Consequently, we undertake some sensitivity tests related 20

We multiply total exports of these countries by 1.5. We only report the results for the average border effect and currency barrier, but separate results for WAEMU and CAEMC countries are available upon request.

21

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

435

Table 4: Sensitivity Tests Related to Distance Measurements Wei’s method (1) (2) Distance Average currency barrier Average border effect Country fixed effects Year dummies R-sq Number of observations

Wolf ’s method (3) (4)

Leamer’s method (5) (6)

−0.52∗ (0.10) −1.70∗ (0.13)

−0.62∗ (0.10) −1.60∗ (0.14)

−0.44∗ (0.09) −1.74∗ (0.13)

−0.51∗ (0.10) −1.66∗ (0.14)

−0.67∗ (0.11) −1.61∗ (0.14)

−0.78∗ (0.11) −1.49∗ (0.14)

−6.35∗ (0.29)

−6.16∗ (0.29)

−6.80∗ (0.24)

−6.71∗ (0.24)

−6.79∗ (0.20)

−6.71∗ (0.20)

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

0.71 4659

0.71 4659

0.71 4659

0.71 4659

0.71 4659

0.71 4659

∗ significant

at the 1 per cent level. Note: Dependent variable: Log of imports value divided by the GDP of importers and exporters. Standard errors in parentheses. Regressions (1) to (6) report the results of estimating (11). Constant, fixed effects and year dummies are not reported. We use different measurements of the internal and external distances: Wei’s method with a regional breakdown method in column (1); original Wei’s method in column (2); Wolf ’s method with a regional breakdown in column (3); original Wolf ’s method in column (4); Leamer’s method with a regional breakdown in column (5) and original Leamer’s method in column (6).

to the use of different measurements of distance. Results are reported in Table 4. We compute six different measures of distance, following different treatments of external and internal distances. Regarding external distances, in columns (1), (3) and (5), we use a regional disaggregation to compute bilateral great circle distances between the main cities of each region (see Appendix B for more details). In columns (2), (4) and (6), we simply calculate external distances as great circle distances between the main economic center of each country. Internal distances are computed in three different ways.22 The first approach is based on Wei’s method (columns 1 and 2). Wei (1996) proposes 22

Calculating internal distances is an “intrinsically difficult problem” (Head and Mayer 2002) and these methods are not totally satisfying.

436

Review of World Economics 2005, Vol. 141 (3)

to measure intra-national distance as one-quarter of the distance to the nearest neighbour (see Appendix B for details). The second possible approach is to compute internal distances using the great circle methodology (Wolf 1997). Therefore, we also calculate the internal distance as the distance between the two main economic centers in each country (columns 3 and 4). Finally, we follow a third approach based on area measures. As in Leamer (1997), we approximate the economic geography of a country with a disk, assuming that producers concentrate in the center and consumers are randomly distributed throughout the rest of the area. √ The internal distance then corresponds to the radius of the disk: area/π (columns 5 and 6). Whatever the methodology used, all average border effects computed with a regional breakdown method are slightly larger than otherwise. Besides, Wolf ’s and Leamer’s methods yield slightly larger estimates for the average border effect than Wei’s approach, without altering our main conclusions. Currency barriers explain between 18 (column 6) and 21 per cent (column 1) of border impediments to trade.

7 Concluding Remarks In this paper, we use a gravity model to examine the extent to which currency barriers explain the border effect puzzle. We focus on the CFA Franc Zone in West and Central Africa, as it exhibits particularly interesting features. First, it is the most long-lasting experience of monetary unions, existing for over 50 years. Second, there is very little literature estimating the border effect for less developed African countries. Finally, the presence of two currency unions (the WAEMU and the CAEMC) allows comparisons between intraand extra-currency unions trade flows. Our results indicate that West and Central African currency unions do not explain much of the border effect puzzle. Currency barriers account for 17 to 28 per cent of the overall border effect, depending on the currency union. This result seems quite robust to the influence of informal trade and distance measurements. West and Central African countries exhibit a large home bias, despite the sharing of a common currency. This finding allows us to draw some policy implications for emerging and developing countries contemplating to adopt a single currency (e.g. the Economic Community of West African States (ECOWAS) members among others). A common currency reduces the border effect, but it is not a decisive factor. Hence, our

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

437

results provide indirect evidence for promoting institutional framework and economic integration before the adoption of a single currency. Appendix A Technical Appendix: The Market Clearing Condition The market clearing condition implies that income in j is equal to spending in all imports23 : Yj =

 i

Mij =

 i


Yi

βj pj tij Pi

1−σ .

(12)

From this equation we obtain the equilibrium scaled prices: (βj pj )1−σ =

Yj .   tij 1−σ Yi Pi i

(13)

Substituting this expression into (6) gives:
 tij 1−σ Yi Yj Mij = , Yw Pi Πj

(14)

where Πj =


  tij 1−σ 1/(1−σ) θi , Pi i

(15)

and θi = Yi /Yw country i’s share of world income Yw . Then using (13) and (5) we have: 
  tij 1−σ 1/(1−σ) Pi = θj . Πj j

(16)

If we assume symmetric trade costs, then comparing (15) with (16) yields Πj = Pj . Thus, we obtain (7) and (8).

23

This condition is equivalent to the balanced trade assumption.

438

Review of World Economics 2005, Vol. 141 (3)

B Data International Trade Our database includes 11 CFA countries (7 belonging to the WAEMU: Benin, Burkina Faso, Ivory coast, Mali, Niger, Senegal,Togo; and 4 to the CAEMC: Cameroon, Chad, Congo and Gabon) and 14 European countries (Austria, BelgiumLuxembourg, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Portugal, Spain, Sweden, the United Kingdom). We consider imports of the CFA countries from all the other countries on the period 1980–1999. The import data, in current U.S. dollar, come from the International Monetary Fund’s (IMF) Direction of Trade. Intra-National Trade We define countries’ imports from themselves as in Wei (1996). Thus, intra-national trade is computed, for each CFA country, as the difference between total goods production and total exports to the rest of the world. We use a three-step procedure. 1. 2. 3.

We compute output/value added ratios for each country, using output data coming from United Nations’ National Accounts Statistics. We compute the average ratio when some output data are missing. We calculate total goods production by multiplying output/value added ratio by total value added in agriculture and industry, taken from the World Development Indicators (World Bank). We compute the difference between total goods production and total exports to the rest of the world, taken from the IMF’s Direction of Trade.

GDP GDP, in current U.S. dollar, come from the World Development Indicators (World Bank). Distance The border effect methodology requires the calculation of both intra-national and international measures of distance. Various methods have been applied to test the sensitivity of our results (see robustness checks in Section 6.3). In the core of the paper, we use mainly the following method: (a) International distances are calculated using a regional disaggregation.24 We divide large countries in different regions.25 Then, we compute great circle dis24

This approach is similar to Head and Mayer (2000), who evaluate the European Union’s border effect. 25 The regional disaggregation of EU countries follows the NUTS Classification (level 1). For Africa, we disaggregate each of the largest countries (Cameroon, Congo, Niger and Chad) into two areas.

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

439

tances between the main cities of the regions of each country and weight those distances by the population of the regions. Finally, we calculate the distance between two countries by summing the inter-regional-weighted distances.26 (b) Intra-national distances are computed using Wei’s seminal method. Wei (1996) proposes to measure intra-national distances as one-quarter of the distance to the nearest neighbour’s capital. Following Wei, we measure it as one-eighth for the smallest countries (Togo). C Missing Values Table A1: Missing Trade Observations Missing data ratio 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

0.29 0.20 0.16 0.19 0.16 0.18 0.20 0.20 0.17 0.14 0.14 0.16 0.16 0.10 0.07 0.08 0.06 0.13 0.12 0.13

Importer country WAEMU: Benin Burkina Faso Ivory Coast Mali Niger Senegal Togo CAEMC: Cameroon Chad Congo Gabon

Missing data ratio

0.14 0.17 0.09 0.16 0.15 0.05 0.13 0.12 0.34 0.15 0.19

Note: Missing data ratio is calculated for each year or for each importer country as the number of missing observations divided by the total number of observations.

26

As an example, to calculate the distance between Cameroon and France, we divided Cameroon into two areas (one centered on the capital: Yaound´e, the other around Douala) and France in eight regions corresponding to the NUTS1 level (Bordeaux, Lille, Lyon, Marseille, Nantes, Paris, Rouen, Strasbourg). Bilateral distance between Cameroon and France is the average weighted distance between Yaound´e-Bordeaux, Yaound´e-Lille, ... Yaound´eStrasbourg, Douala-Bordeaux, ... Douala-Strasbourg.

440

Review of World Economics 2005, Vol. 141 (3)

References Anderson, J. (1979). A Theoretical Foundation for the Gravity Equation. American Economic Review 69 (1): 106–116. Anderson, J., and E. van Wincoop (2003). Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review 93 (1): 170–192. Bergstrand, J. (1985). The Gravity Equation in International Trade: Some Microeconomic Foundations and Empirical Evidence. Review of Economics and Statistics 67 (3): 474–481. Deardorff, A. (1998). Determinants of Bilateral Trade: Does Gravity Work in a Neoclassical World? In J. A. Frankel (ed.), The Regionalization of the World Economy. Chicago: University of Chicago Press. Emerson, M., D. Gros, A. Italianer, J. Pisany-Ferry, and H. Reichenbach (1992). One Market, One Money. Oxford: Oxford University Press. Feenstra, R. (2004). Advanced International Trade: Theory and Evidence. Princeton: Princeton University Press. Gros, D., and N. Thygesen (1998). European Monetary Integration: From the European Monetary System to European Monetary Integration. London: Longman. Head, K., and T. Mayer (2000). Non-Europe: The Magnitude and Causes of Market Fragmentation in Europe. Review of World Economics/Weltwirtschaftliches Archiv 136 (2): 284–314. Head, K., and T. Mayer (2002). Illusory Border Effects: Distance Measurement Inflates Estimates of Home Bias in Trade. CEPII Working Paper 2002-01. Centre d’Etudes Prospectives et d’Informations Internationales, Paris. Helliwell, J. (1996). Do National Borders Matter for Quebec’s Trade? Canadian Journal of Economics 29 (3): 507–522. Helliwell, J. (1998). How Much Do National Borders Matter? Washington, D.C.: Brooking Institution. Hillberry, R. (1998). Regional Trade and the Medicine Line: The National Border Effect in U.S. Commodity Flow Data. Journal of Borderlands Studies 8 (2): 1–17. IMF Country Report (2002). Central African Economic and Monetary Community—Paper on Recent Developments and Regional Policy Issues. Technical Report 203, International Monetary Fund, Washington, D.C. Leamer, E. (1997). Access to Western Markets, and Eastern Effort Levels. In S. Zecchini (ed.), Lessons from the Economic Transition: Central and Eastern Europe in the 1990’s. Dordrecht: Kluwer Academic Publishers. McCallum, J. (1995). National Borders Matter: Canada-US Regional Trade Patterns. American Economic Review 85 (3): 615–623. Nitsch, V. (2005). Currency Union Entries and Trade. Forthcoming in J. Hölscher and H. Tomann, Globalization of Capital Markets and Monetary Policy. Palgrave Macmillan.

de Sousa/Lochard: Do Currency Barriers Solve the Border Effect Puzzle?

441

Obstfeld, M., and K. Rogoff (2000). The Six Major Puzzles in International Macroeconomics: Is There a Common Cause? NBER Working Paper 7777. National Bureau of Economic Research, Cambridge, Mass. Parsley, D., and S.-J. Wei (2001). Explaining the Border Effect: The Role of Exchange Rate Variability, Shipping Costs and Geography. Journal of International Economics 55 (1): 87–105. Rose, A. (2000). One Money, One Market. Estimating the Effect of Common Currencies on Trade. Economic Policy 30: 7–45. Rose, A. (2004). A Meta-Analysis of the Effect of Common Currencies on International Trade. NBER Working Paper 10373. National Bureau of Economic Research, Cambridge, Mass. Wei, S.-J. (1996). Intra-National versus Inter-National Trade: How Stubborn Are Nations in Global Integration? NBER Working Paper 75531. National Bureau of Economic Research, Cambridge, Mass. Wolf, H. (1997). Patterns of Intra- and Inter-State Trade. NBER Working Paper 5939. National Bureau of Economic Research, Cambridge, Mass.