Chapter

Chapter 2: The Benefits of Trade Integration in the Southern African Customs Union

Author(s):
Joannes Mongardini, Tamon Asonuma, Olivier Basdevant, Alfredo Cuevas, Xavier Debrun, Lars Engstrom, Imelda Flores Vazquez, Vitaliy Kramarenko, Lamin Leigh, Paul Masson, and Genevieve Verdier
Published Date:
April 2013
Share
  • ShareShare
Show Summary Details
Author(s)
Lars Engstrom and Geneviève Verdier

Despite being the oldest customs union in the world, the Southern African Customs Union (SACU) has often been criticized as a trade agreement benefiting its largest and most powerful member, South Africa, and being detrimental to its smaller members by hindering industrialization and inducing trade diversion. This chapter provides empirical evidence supporting the view that SACU has benefited all of its members and has outperformed other trade arrangements in Africa.

The chapter seeks to assess the extent to which SACU has expanded trade among member countries. It also assesses the impact of other trade agreements in Africa that can serve as benchmarks for SACU performance. The analysis estimates a gravity model of trade, which allows trade expansion, diversion, and creation to be separated. In models of this kind, bilateral trade between a pair of countries is explained by country-specific variables (e.g., size), pair-specific variables (e.g., shared border or language), and the presence of a trade agreement or a regional agreement to promote economic cooperation.

The quantitative literature on the impact of preferential trade agreements is extensive and can be divided into two groups.1 Ex ante studies focus on estimating the effect of a preferential trade agreement before it is put into place by using trade patterns and estimated elasticities in computable general equilibrium models.2 Ex post studies focus on analyzing bilateral trade flows after trade agreements have been put in place. The tool of choice in these papers—and in this chapter as well—is a gravity model of bilateral trade, which has become the workhorse model for analyzing bilateral trade.

Ex post studies draw on aggregate data as well as sector-level or commodity-level data. Aggregate bilateral trade data have been used to study the impact of preferential trade agreements (Frankel, 1997; Endoh, 1999; Krueger, 1999; Magee, 2008), currency unions (Rose, 2000; Rose and Van Wincoop, 2001; Frankel and Rose, 2002), and border effects (McCallum, 1995), as well as other historical or cultural influences on trade. Studies using more disaggregated data include Clausing (2001), who estimates the effect of the Canada-U.S. Free Trade Agreement, and Romalis (2005), who focuses on the North American Free Trade Agreement (NAFTA).

A number of studies analyze African trade. Foroutan and Pritchett (1993), with a focus on intra–sub-Saharan African trade, and Coe and Hoffmaister (1999), with greater emphasis on African trade with industrial countries, both find that the gravity model accurately predicts African bilateral trade patterns. Other papers sometimes reach different conclusions. Although they both estimate the impact of the Common Market for Eastern and Southern Africa (COMESA) on bilateral trade, Cernat (2001) finds evidence of trade creation, while Subramanian and Tamirisa (2001) suggest trade diversion. A more recent example is the work of Mayda and Steinberg (2007), who focus on Uganda’s commodity-level trade within COMESA and find that the trade agreement has not increased Uganda’s trade with member countries.

The analysis in this chapter is closest to the work of Carrère (2004). Using aggregate bilateral trade data, Carrère (2004) finds that trade agreements in Africa generated a significant increase in trade between 1962 and 1996 although initially often through trade diversion. In contrast to that study, this chapter offers an estimate of trade creation for SACU—which was not included in her study—and updates the estimation to 2008.

The gravity model estimated in this chapter provides evidence supporting the view that SACU has promoted trade:

  • SACU outperforms other African regional trade agreements (RTAs). Although most African trade agreements are trade-creating, the magnitude of trade creation is much larger for SACU.
  • Based on bilateral exports data, intra-SACU trade is 57 times higher than expected compared with a world reference, and 145 times higher than expected compared with an Africa-only sample. There is no evidence of export diversion but some evidence of import diversion, that is, part of the increase in intra-SACU trade has replaced imports from the rest of the world (ROW).
  • Estimates based on bilateral imports data confirm the trade-creating effects of SACU membership, but the positive effects are smaller than in the data set of bilateral exports and, interestingly, these estimates do not confirm import diversion. That is, SACU has increased intra-SACU trade—and in particular imports to Botswana, Lesotho, Namibia, and Swaziland (BLNS) from South Africa—but there is little evidence of imports from the ROW being crowded out.

To sum up, this analysis finds that the positive effects from trade expansion (increasing intra-SACU trade) are significant, whereas the negative effects from trade diversion are limited (or possibly zero). SACU has increased total intra-SACU trade and BLNS’s imports of goods from South Africa, but whether this has reduced BLNS’s imports from the ROW is questionable. This result provides important information to policymakers. Any future reform of the current SACU arrangements should ensure that the benefits of trade creation are increased.

The chapter is organized as follows: The next section reviews stylized facts about trade patterns within SACU and is followed by sections describing the chapter’s methodology and presenting the results. The final section offers concluding remarks.

Trade Patterns within SACU

BLNS, except Swaziland, mainly export primary commodities to markets outside SACU, but South Africa has a significant market share in BLNS. About 70 percent of imports to BLNS originate from South Africa. From consumer durables to machinery and equipment, BLNS rely heavily on imports from South Africa, with little imports from outside SACU (see Figures 2.1 and 2.2).

Figure 2.1Exports and Imports of BLNS

Source: IMF staff estimates.

Note: BLNS = Botswana, Lesotho, Namibia, and Swaziland.

Figure 2.2Geographic Distribution of Exports of Goods

(Percentage of the total value of exports)

Sources: Botswana Central Statistics Office; Central Bank of Lesotho; Lesotho Central Bureau of Statistics; UN Comtrade database; and IMF staff estimates.

Natural resource endowments (diamonds) and trade agreements that provide tariff preferences explain much of the export pattern of BLNS. About 90 percent of Botswana’s exports, dominated by diamonds, are destined to markets outside SACU. Namibia also exports diamonds, in addition to ores, printing materials, and fish, with about 70 percent of exports destined for markets outside SACU in 2005–08. A key reason for Lesotho’s export pattern is the African Growth and Opportunity Act that gives Lesotho preferential access to the U.S. market; its exports are dominated by garments exported to the United States. In total, 60 percent of Lesotho’s exports were headed to markets outside SACU during 2005–10. Swaziland’s export pattern is more closely linked with South Africa. Swaziland’s exports include processed and semiprocessed food products—such as sugar, soft drink concentrates, and canned fruits—and South Africa received about 50 percent of Swaziland’s exports in 2005–10.

Previous studies on SACU have been focused on cost- and consumer price–raising effects of tariffs. Flatters and Stern (2006) argue that a common external tariff escalates consumer prices. The gross cost-raising effect is the difference between prices faced in the presence of the tariff and those that would prevail in its absence. They estimate the associated probable welfare loss to be in the range of 2–10 percent of GDP for BLNS (Table 2.1). This estimate, however, ignores the impact of possible transportation costs, the fact that BLNS could have had their own tariffs in the absence of SACU, and the structural effects of SACU (e.g., trade diversion and trade creation).

Table 2.1BLNS: Gross Cost-Raising Effects of SACU Tariffs in 2006
Import of goods in 2006, f.o.b.Estimated cost-raising impact for 2006
(Millions of U.S. dollars)Percentage of value of imports1Millions of U.S. dollars2Percentage of GDP
ProbablePossibleProbablePossibleProbablePossible
Botswana3,0539.214.828145224
Lesotho1,4699.616.61412441017
Namibia2,7997.415.120742335
Swaziland1,9157.914.1151270610
Source: Authors’ compilation.Note: f.o.b. = freight on board.

Estimates provided in Flatters and Stern (2006).

Estimates in U.S. dollars deviate from numbers presented in Flatters and Stern (2006) due to differences in trade data used for the calculations.

Source: Authors’ compilation.Note: f.o.b. = freight on board.

Estimates provided in Flatters and Stern (2006).

Estimates in U.S. dollars deviate from numbers presented in Flatters and Stern (2006) due to differences in trade data used for the calculations.

Methodology and Estimation Issues

Gravity models state that economic interactions between two countries are proportional to the size of these entities and inversely related to the distance between them. In addition to distance, gravity models also relate trade flows to country-specific characteristics, such as linguistic and cultural factors. Gravity models also estimate the effect of bilateral and RTAs on trade flows.

This chapter estimates variants of the following equation:

in which xijt is bilateral trade (measured by exports or imports) between countries i and j at time t, βt are year effects, Yi is real GDP in country i, Pit is population in country i, distij is the distance between countries i and j, Areai is country i’s area in square kilometers, RTAijt is a dummy variable that is equal to 1 if countries i and j share a regional trade agreement at time t, RTA_ROWij is a dummy variable that is equal to 1 if the exporting country is a member of the RTA and 0 otherwise, ROW_RTAij is a dummy variable that is equal to 1 if the importing country is a member of the RTA and 0 otherwise, and Dij are time-invariant dummies that capture characteristics of the country pair (i.e., whether they share a common language or a border).3

For this equation to be estimated successfully, a number of econometric issues must be addressed, including the treatment of zero-trade observations, nonlinearities, omitted variables, and selection bias.4 These challenges and others are discussed in Appendix 2A. Of particular importance to this work are the choice of control variables, possible endogeneity, and the measurement of trade creation.

  • Controlling for country-pair characteristics. Trade is more likely between countries with strong cultural and historical ties or trade arrangements. Following the literature, the analysis controls for whether countries share a common border, have had a colonial link, share a common language, and are a part of the same trade arrangement.
  • Controlling for monetary arrangements. A large literature exists on the positive effects of currency unions on bilateral trade.5 In empirical studies, currency unions are usually controlled for by adding a 0–1 dummy variable. In many cases in Africa, however, membership in a currency union is indistinguishable from membership in an RTA (for example, SACU). The alternative here is to follow Carrère (2004) and use the volatility of the bilateral nominal exchange rate to control for the exchange rate regime. Volatility is defined as the standard deviation of the log change in the monthly bilateral nominal exchange rate.
  • Dealing with endogeneity. As noted in the literature (e.g., see Krugman, 1991), countries may be “natural trading partners,” that is, countries with high bilateral trade flows will tend to form regional agreements. To deal with this endogeneity problem, a bilateral random effect (∈ij) is introduced, which controls for unobserved factors that may explain bilateral trade (see Appendix 2A for more details).
  • Measuring trade expansion, diversion, and creation. The gravity equation can be used to estimate the extent and magnitude of trade expansion, diversion, and creation. Trade expansion occurs when bilateral trade is higher for members of a trade arrangement (βRTA > 0). Trade diversion occurs when intra-bloc trade increases but trade with the ROW [imports (βROW_RTA < 0) or exports (βRTA_ROW < 0)] falls. There is trade creation if intra-bloc trade increases without reducing trade with the ROW or if trade expansion exceeds trade diversion. Following Magee (2008), counterfactuals from estimated coefficients of the gravity equations can be used to determine not only whether trade creation occurred but also its magnitude (see Appendix 2A for further details).

Data

The quality of trade data for BLNS is low, with unreliable and often inconsistent numbers. To deal with this problem, the gravity models are estimated using two separate data sets for bilateral exports and bilateral imports, respectively:

  • Bilateral exports data are taken from the IMF’s Direction of Trade Statistics (DOTS) database, which provides an unbalanced panel of exports in U.S. dollars for 150 countries and a total of 405,171 observations for the period 1990–2008. In the DOTS database, BLNS are treated as one group because of the lack of credible data for trade between each BLNS country and the ROW. However, the data for bilateral trade between the BLNS group and South Africa (intra-SACU trade) appear unreliable. This measurement problem is corrected by replacing intra-SACU trade with bilateral exports data from the UN Comtrade database, which provides trade data in U.S. dollars for 150 countries for the period 1962–2008.6
  • Bilateral imports data are taken from the UN Comtrade database for the period 1990–2008. Import data tend to be more reliably reported because of the collection of import tariffs. SACU countries are also available as individual entities in the UN Comtrade database, which means that trade creation and diversion can be estimated for individual countries in SACU. In this context, it is important to note that imports data could overstate the BLNS group’s reliance on imports from South Africa either because the country of origin is not properly recorded or South African distributors are selling re-exported goods previously imported to South Africa.

The control variables of interest are the dummy variables for RTAs from the World Trade Organization’s database and include the East African Community (EAC), the Economic Community of West African States (ECOWAS), the Central African Economic and Monetary Community (CEMAC), COMESA, SACU, the Southern African Development Community (SADC), and the West African Economic and Monetary Union (WAEMU) for Africa, as well as the European Union (EU), NAFTA, the Southern Common Market (MERCOSUR), the Andean Community (a customs union comprising Bolivia, Colombia, Ecuador, and Peru; ANDEAN), and the ASEAN Free Trade Area (AFTA). Because African trade agreements are often overlapping, dummy variables for African trade agreements are redefined to exclude overlapping effects. For example, because all WAEMU countries are also members of ECOWAS, including unmodified dummy variables for both would make it difficult to identify the separate effects of these trade agreements. The ECOWAS dummy variable is therefore redefined to exclude intra-WAEMU trade as well as trade between WAEMU and non-ECOWAS countries.7

Control variables are from (1) the World Bank’s World Development Indicators for real GDP and population; (2) the IMF’s International Financial Statistics for exchange rates; and (3) the Centre d’Etudes Prospectives et d’Informations Internationales database for gravity variables (distance, area, country-pair characteristics).

Because the DOTS database only contains data for BLNS as a group, adjustments must be made to the control variables for BLNS. Each BLNS observation is computed as the weighted average of individual country observations, with population shares used as weights. For example, the distance between each BLNS country and South Africa is the population-weighted average of the distances between the capitals of South Africa and Botswana, Lesotho, Namibia, and Swaziland.

Results

Table 2.2 presents the results for bilateral exports from the DOTS database for three samples: the full sample, non-Organization for Economic Cooperation and Development (non-OECD) member countries, and African countries only. The three samples provide three different reference points for evaluating the performance of RTAs in Africa. In the full sample, African RTAs are evaluated against a world norm.8 The non-OECD sample excludes intra-OECD trade, that is, trade between advanced economies. Because trade between OECD countries is dominated by intra-industry trade, the non-OECD sample may be more appropriate for assessing SACU and African RTAs in general. Finally, the analysis evaluates African RTAs against an African norm in the Africa-only sample, which only includes intra-African trade and trade between Africa and the ROW. For each sample, two types of estimations are shown: the first uses standard random effects and the second uses the Hausman-Taylor instrumental variable method to deal with the potential correlation between random effects and some of the regressors. A Hausman test reveals a correlation between the bilateral random effects and ln(YiYj) and ln (YiYj/PiPj) justifying the use of the Hausman-Taylor method.

Table 2.2Bilateral Exports Using the Direction of Trade Statistics Database
Full sampleNon-OECDAfrica only
(1)(2)(3)(4)(5)(6)
Dependent variable: ln(1 + real exports)
Constant−18.35***−18.27***−15.53***−15.72***−7.132***−7.553***
(−25.94)(−136.01)(−22.28)(−110.32)(−7.08)(−26.44)
ln(distij)−0.712***−0.741***−0.654***−0.673***−0.670***−0.790***
(−15.80)(−94.70)(−13.40)(−78.82)(−5.66)(−41.35)
ln(YiYj)0.564***0.582***0.502***0.518***0.328***0.392***
(27.72)(168.04)(23.46)(133.10)(13.68)(43.13)
ln(YiYjPiPj)0.03390.00637−0.00764−0.0194***−0.0474−0.0939***
(1.37)(1.18)(−0.30)(−3.37)(−1.40)(−8.67)
Landlocked−0.422***−0.470***−0.445***−0.473***−0.296***−0.329***
(−7.21)(−38.35)(−7.29)(−34.87)(−4.00)(−13.91)
ln(AreaiAreaj)−0.0974***−0.107***−0.0879***−0.0979***−0.0748***−0.107***
(−6.39)(−39.96)(−5.39)(−33.36)(−4.01)(−18.97)
Common language0.352***0.331***0.315***0.296***0.09140.0481
(3.78)(20.50)(3.43)(17.71)(0.86)(1.71)
Contiguous0.4540.304***0.4960.354***0.5010.450***
(1.91)(5.91)(1.62)(5.89)(1.47)(4.78)
Colony1.690***1.683***2.123***2.176***2.691*2.580***
(6.15)(32.29)(5.85)(31.22)(2.56)(23.81)
Total year0.0244***0.0268***0.0246***0.0255***0.0362**0.0368***
(4.54)(12.88)(5.00)(12.28)(2.98)(7.26)
Entire year−0.128**−0.157***−0.167***−0.174***−0.0863−0.134***
(−2.95)(−11.61)(−1.09)(−12.57)(−1.09)(−3.92)
Difference_year0.01390.0151−0.0167−0.0169−0.0571***−0.0220
(1.05)(1.24)(−1.34)(−1.35)(−5.59)(−1.21)
Volatility of exchange rate−0.0257**−0.0265**−0.0123−0.01360.000631−0.0113
(−3.03)(−2.81)(−1.04)(−1.18)(0.03)(−0.70)
SACU3.871***4.014**4.268***4.420***5.079***4.956***
(16.51)(2.92)(14.46)(3.32)(15.76)(4.03)
SACU_ROW0.545*0.592***0.504*0.503***1.214***1.257***
(2.31)(12.48)(2.46)(10.47)(7.84)(20.65)
ROW_SACU−0.0917−0.0430−0.231−0.230***−0.463**−0.420***
(−0.40)(−0.97)(−1.18)(−5.01)(−2.67)(−7.54)
SADC0.2650.246***0.2940.293***0.09510.145**
(1.60)(4.08)(1.76)(5.23)(1.05)(2.86)
SADC_ROW0.05480.0498***0.04650.0436**0.0916**0.134***
(1.05)(3.57)(0.99)(3.14)(2.65)(8.81)
ROW_SADC0.182***0.177***0.182***0.181***0.05800.104***
(3.47)(12.67)(3.74)(12.31)(1.28)(5.42)
COMESA0.244**0.253***0.247***0.263***0.216***0.248***
(3.01)(7.49)(3.34)(8.37)(3.49)(9.09)
COMESA_ROW−0.120**−0.116***−0.0508−0.0454***0.156***0.190***
(−3.09)(−9.15)(−1.33)(−3.64)(4.67)(14.72)
ROW_COMESA0.219***0.223***0.207***0.214***0.137**0.172***
(5.79)(17.43)(5.39)(15.98)(3.10)(9.25)
ECOWAS0.314*0.273***0.1020.0789−0.246−0.253***
(2.16)(4.19)(1.08)(1.23)(−1.54)(−4.38)
ECOWAS_ROW−0.130−0.155***−0.121−0.129***−0.0534−0.0357
(−1.87)(−9.75)(−1.85)(−8.01)(−0.91)(−1.82)
ROW_ECOWAS0.188**0.164***0.125*0.118***−0.0392−0.0123
(2.89)(11.21)(1.98)(7.71)(−0.58)(−0.60)
(3.13)(6.59)(3.48)(7.90)(4.15)(11.30)
Dependent variable: ln(1 + real exports)
EAC0.799**0.820***0.879***0.911***1.005***1.116***
(3.13)(6.59)(3.48)(7.90)(4.15)(11.30)
EAC_ROW−0.156*−0.149***−0.124−0.112***0.06190.119***
(−2.38)(−7.84)(−1.91)(−5.99)(1.02)(5.74)
ROW_EAC−0.0119−0.00416−0.009890.00293−0.144*−0.0883***
(−0.18)(−0.22)(−0.14)(0.15)(−1.97)(−3.55)
CEMAC−0.0455−0.0215−0.0291−0.01440.03660.188*
(−0.17)(−0.19)(−0.12)(−0.14)(0.17)(2.10)
CEMAC_ROW−0.0656−0.0590**−0.112−0.114***−0.04190.0247
(−0.90)(−3.00)(−1.78)(−5.77)(−0.87)(1.16)
ROW_CEMAC−0.0148−0.00773−0.0202−0.0213−0.03650.0352
(−0.20)(−0.39)(−0.32)(−1.04)(−0.63)(1.41)
WAEMU0.832*0.766***0.6470.617***0.3510.400***
(2.16)(7.67)(1.73)(6.49)(0.87)(4.77)
WAEMU_ROW−0.0299−0.0538***−0.00673−0.01330.04760.0816***
(−0.44)(−3.51)(−0.10)(−0.86)(0.66)(4.60)
ROW_WAEMU0.169*0.144***0.09430.0866***0.04510.0949**
(1.97)(6.25)(1.23)(3.46)(0.51)(2.93)
Observations375,054375,054284,243284,243109,824109,824
Year effectsyesyesyesyesyesyes
Cluster country pairyesnoyesnoyesno
Random effectsyesyesyesyesyesyes
Hausman-Taylornoyesnoyesnoyes
Source: IMF staff.Note: CEMAC = Central African Economic and Monetary Community; COMESA = Common Market for Eastern and Southern Africa; EAC = East African Community; ECOWAS = Economic Community of West African States; ROW = rest of the world; SACU = Southern African Customs Union; SADC = Southern African Development Community; WAEMU = West African Economic and Monetary Union. These regressions also include country-specific and country-pair characteristics as well as dummies for trade agreements in non-African countries. In some regressions (“cluster country pair”), the standard errors are computed assuming zero correlation across groups but allow a non-zero within-cluster or within-country-pair correlation, t-statistics are in parentheses. *, **, and ***, denote significance at the 5 percent, 1 percent, and 0.1 percent levels of confidence.
Source: IMF staff.Note: CEMAC = Central African Economic and Monetary Community; COMESA = Common Market for Eastern and Southern Africa; EAC = East African Community; ECOWAS = Economic Community of West African States; ROW = rest of the world; SACU = Southern African Customs Union; SADC = Southern African Development Community; WAEMU = West African Economic and Monetary Union. These regressions also include country-specific and country-pair characteristics as well as dummies for trade agreements in non-African countries. In some regressions (“cluster country pair”), the standard errors are computed assuming zero correlation across groups but allow a non-zero within-cluster or within-country-pair correlation, t-statistics are in parentheses. *, **, and ***, denote significance at the 5 percent, 1 percent, and 0.1 percent levels of confidence.

Let us consider the rationale for the gravity variables first. Larger countries (as measured by GDP) trade more, whereas greater distance tends to reduce bilateral trade. Country-pair characteristics matter: countries that share a common language, are contiguous, or share a common colonial past trade more. Because membership in a currency union is often indistinguishable from membership in a trade agreement, it is also important to separate their different effects on trade. As indicated above, the analysis controls for membership in a currency union by adding exchange rate volatility as an explanatory variable. As expected, exchange rate volatility reduces bilateral trade (which is consistent with the literature on estimating the impact of currency unions on trade) but only in the full sample; it does not appear to matter in the smaller samples.

How has SACU performed since 1990? SACU has a large positive effect on intra-SACU trade. The results also provide evidence of trade expansion in SADC, COMESA, EAC, ECOWAS, and WAEMU, but the largest expansion takes place within SACU. These estimates are converted into a measure of the additional trade by taking the exponential of individual estimated coefficients. Intra-SACU trade is more than 55 times higher [= exp(4.014), Table 2.2, column (2) for the full sample, and Table 2.4, column (1) for the full sample] than the expected level compared with a world reference, and 142 times higher [= exp(4.956), Table 2.2, column (6) and Table 2.4, column (3)] in the Africa-only sample. By contrast, intra-COMESA trade is only 1.3 times higher, intra-EAC trade is about 2–3 times higher, and intra-WAEMU trade is about 1.5–2 times higher than the norm [Table 2.4, columns (1)–(3)]. Surprisingly, there is only weak evidence of trade expansion in CEMAC and ECOWAS.

Table 2.3Bilateral Imports Using UN Comtrade Database
Full sampleNon-OECDAfrica only
(1)(2)(3)(4)(5)(6)
Dependent variable: ln(1 + real imports)
Constant−25.61***−25.60***−24.75***−24.75***−29.49***−29.47***
(−67.94)(−67.93)(−54.49)(−54.46)(−26.07)(−26.01)
ln(distij)−1.251***−1.251***−1.219***−1.219***−0.612***−0.613***
(−60.04)(−60.05)(−47.67)(−47.67)(−9.71)(−9.71)
ln(YiYj)1.064***1.064***0.992***0.992***1.182***1.182***
(103.41)(103.42)(73.25)(73.25)(31.71)(31.71)
ln(YiYjPiPj)0.109***0.109***0.167***0.167***−0.217***−0.218***
(6.64)(6.64)(8.19)(8.18)(−4.60)(−4.61)
Landlocked−0.496***−0.495***−0.301***−0.300***−0.125−0.123
(−15.79)(−15.77)(−7.83)(−7.81)(−1.77)(−1.73)
ln(AreaiAreaj)−0.184***−0.184***−0.136***−0.136***−0.273***−0.274***
(−23.61)(−23.61)(−13.49)(−13.50)(−11.97)(−11.98)
Common language0.816***0.816***0.761***0.762***0.417***0.417***
(19.19)(19.19)(15.50)(15.50)(4.67)(4.66)
Contiguous0.2050.1970.445**0.429**1.784***1.764***
(1.72)(1.65)(2.89)(2.77)(6.40)(6.27)
Colony1.254***1.252***1.016***1.009***2.287***2.276***
(11.01)(10.99)(6.13)(6.09)(7.29)(7.24)
Total year0.0526***0.0526***0.0783***0.0784***0.0972***0.0977***
(24.48)(24.50)(29.25)(29.28)(14.83)(14.89)
Entire year0.0704***0.0702***0.306***0.306***0.239**0.237**
(4.08)(4.07)(11.94)(11.93)(3.17)(3.15)
Difference_year0.454***0.454***0.408***0.408***0.333***0.333***
(34.22)(34.22)(27.57)(27.57)(14.11)(14.10)
Dependent variable: ln(1 + real imports)
Volatility of exchange rate0.007890.00787−0.0188−0.0188−0.0149−0.0150
(0.35)(0.35)(−0.56)(−0.56)(−0.24)(−0.25)
Infrastructure(i)0.04620.04500.02830.0263−0.151**−0.157**
(1.70)(1.66)(0.84)(0.78)(−2.62)(−2.73)
Infrastructure(j)−0.176***−0.175***−0.159***−0.158***−0.0358−0.0303
(−7.83)(−7.79)(−5.48)(−5.42)(−0.71)(−0.60)
SACU1.4841.0992.213*
(1.86)(1.38)(2.40)
ZAF_BLNS2.2691.9672.357
(1.77)(1.54)(1.60)
BLNS_ZAF3.912**3.658**4.159**
(3.04)(2.85)(2.82)
BLNS0.381−0.07721.494
(0.37)(−0.08)(1.26)
SACU_ROW−0.0965−0.0965−0.127−0.1280.1410.142
(−1.18)(−1.18)(−1.39)(−1.39)(1.08)(1.09)
ROW_SACU0.1540.1540.1900.1901.554***1.556***
(1.88)(1.88)(1.91)(1.91)(9.90)(9.90)
SADC0.864***0.864***0.669***0.670***0.500*0.499*
(4.94)(4.94)(3.76)(3.77)(2.54)(2.54)
SADC_ROW0.399***0.399***0.225***0.225***−0.0876−0.0882
(8.87)(8.87)(4.48)(4.47)(−1.34)(−1.35)
ROW_SADC0.272***0.272***0.142**0.142**0.208*0.207*
(6.25)(6.25)(2.64)(2.64)(2.48)(2.46)
COMESA0.690***0.690***0.624***0.625***0.560***0.560***
(6.43)(6.44)(5.73)(5.74)(5.01)(5.02)
COMESA_ROW0.498***0.498***0.372***0.372***0.101*0.101*
(12.66)(12.66)(8.69)(8.69)(1.98)(1.98)
ROW_COMESA−0.355***−0.355***−0.403***−0.403***−0.00914−0.00909
(−9.16)(−9.16)(−8.49)(−8.49)(−0.11)(−0.11)
ECOWAS0.963***0.964***0.631***0.631***0.784***0.783***
(5.34)(5.35)(3.34)(3.35)(3.69)(3.68)
ECOWAS_ROW0.611***0.611***0.743***0.743***0.289***0.289***
(12.09)(12.09)(12.88)(12.88)(3.73)(3.72)
ROW_ECOWAS−0.423***−0.423***−0.190***−0.191***−0.119−0.121
(−9.53)(−9.53)(−3.53)(−3.54)(−1.40)(−1.42)
EAC0.6770.6780.7470.7480.7330.733
(1.71)(1.71)(1.86)(1.86)(1.83)(1.83)
EAC_ROW0.02860.02840.1040.103−0.294***−0.297***
(0.49)(0.48)(1.61)(1.60)(−3.49)(−3.52)
ROW_EAC−0.289***−0.289***−0.0800−0.08000.499***0.499***
(−4.91)(−4.91)(−1.13)(−1.13)(4.80)(4.80)
CEMAC0.6120.6130.4500.4520.7150.715
(1.66)(1.66)(1.20)(1.21)(1.91)(1.92)
CEMAC_ROW−0.0743−0.0742−0.0956−0.09570.02390.0235
(−1.21)(−1.21)(−1.40)(−1.41)(0.30)(0.29)
ROW_CEMAC−0.284***−0.284***−0.414***−0.414***−0.0583−0.0592
(−4.83)(−4.83)(−5.85)(−5.86)(−0.56)(−0.57)
WAEMU1.341***1.342***1.023***1.023***0.761*0.758*
(4.56)(4.56)(3.39)(3.39)(2.39)(2.38)
WAEMU_ROW_lyear_20080.313***0.313***0.295***0.294***−0.133−0.135
Observations186,062186,062123,711123,71140,16640,166
Year effectsyesyesyesyesyesyes
Cluster country pairnononononono
Random effectsyesyesyesyesyesyes
Hausman-Tayloryesyesyesyesyesyes
Source: IMF staff.Note: BLNS = Botswana, Lesotho, Namibia, and Swaziland; CEMAC = Central African Economic and Monetary Community; COMESA = Common Market for Eastern and Southern Africa; EAC = East African Community; ECOWAS = Economic Community of West African States; SACU = Southern African Customs Union; SADC = Southern African Development Community; ROW = rest of the world; WAEMU = West African Economic and Monetary Union; ZAF = imports from South Africa. These regressions also include country-specific and country-pair characteristics as well as dummies for trade agreements in non-African countries. In some regressions (“cluster country pair”), the standard errors are computed assuming zero correlation across groups but allow a non-zero within-cluster or within-country-pair correlation, t-statistics are in parentheses. *, **, and ***, denote significance at the 5 percent, 1 percent, and 0.1 percent levels of confidence.
Source: IMF staff.Note: BLNS = Botswana, Lesotho, Namibia, and Swaziland; CEMAC = Central African Economic and Monetary Community; COMESA = Common Market for Eastern and Southern Africa; EAC = East African Community; ECOWAS = Economic Community of West African States; SACU = Southern African Customs Union; SADC = Southern African Development Community; ROW = rest of the world; WAEMU = West African Economic and Monetary Union; ZAF = imports from South Africa. These regressions also include country-specific and country-pair characteristics as well as dummies for trade agreements in non-African countries. In some regressions (“cluster country pair”), the standard errors are computed assuming zero correlation across groups but allow a non-zero within-cluster or within-country-pair correlation, t-statistics are in parentheses. *, **, and ***, denote significance at the 5 percent, 1 percent, and 0.1 percent levels of confidence.
Table 2.4Trade Effects of Regional Trade Agreements in Africa
Bilateral exports, DOTS databaseBilateral imports, UN Comtrade database
Full sampleNon-OECDAfrica onlyFull sampleNon-OECDAfrica only
(1)(2)(3)(4)(5)(6)
SACUTrade expansion55.483.1142.00.00.09.1
Import diversion0.0−0.2−0.30.00.00.0
Export diversion1.81.73.50.00.04.7
Trade creation57.284.5145.20.00.013.9
SADCTrade expansion1.31.31.22.42.01.6
Import diversion1.21.20.01.51.30.0
Export diversion1.11.01.11.31.21.2
Trade creation3.53.62.35.24.42.9
COMESATrade expansion1.31.31.32.01.91.8
Import diversion1.21.21.21.61.51.1
Export diversion−0.10.01.2−0.3−0.30.0
Trade creation2.42.53.73.43.02.9
ECOWASTrade expansion1.30.0−0.22.61.92.2
Import diversion1.21.10.01.82.11.3
Export diversion−0.1−0.10.0−0.3−0.20.0
Trade creation2.31.0−0.24.13.83.5
EACTrade expansion2.32.53.10.00.00.0
Import diversion0.00.0−0.10.00.0−0.3
Export diversion−0.1−0.11.1−0.30.0−0.4
Trade creation2.12.44.1−0.30.0−0.6
CEMACTrade expansion0.00.01.20.00.00.0
Import diversion0.00.00.00.00.00.0
Export diversion−0.1−0.10.0−0.2−0.30.0
Trade creation−0.1−0.11.2−0.2−0.30.0
WAEMUTrade expansion2.21.91.53.82.82.1
Import diversion1.21.11.11.41.30.0
Export diversion−0.10.01.1−0.5−0.2−0.5
Trade creation3.32.93.74.73.91.7
Source: IMF staff.Note: CEMAC = Central African Economic and Monetary Community; COMESA = Common Market for Eastern and Southern Africa; DOTS = Direction of Trade Statistics; EAC = East African Community; ECOWAS = Economic Community of West African States; OECD = Organization for Economic Cooperation and Development; SACU = Southern African Customs Union; SADC = Southern African Development Community; WAEMU = West African Economic and Monetary Union. Trade expansion and diversion are computed using significant coefficients. When the coefficient is insignificant, the expansion/diversion magnitude is zero.
Source: IMF staff.Note: CEMAC = Central African Economic and Monetary Community; COMESA = Common Market for Eastern and Southern Africa; DOTS = Direction of Trade Statistics; EAC = East African Community; ECOWAS = Economic Community of West African States; OECD = Organization for Economic Cooperation and Development; SACU = Southern African Customs Union; SADC = Southern African Development Community; WAEMU = West African Economic and Monetary Union. Trade expansion and diversion are computed using significant coefficients. When the coefficient is insignificant, the expansion/diversion magnitude is zero.

Trade expansion, however, is not direct evidence of trade creation if it is accompanied by substantial export and import diversion. Overall, these estimates do not suggest large levels of diversion (βRTA_ROW is small). Export diversion ranges from 0 to 10 percent in COMESA, ECOWAS, EAC, CEMAC, and WAEMU [Table 2.4, columns (1)–(3), based on estimates in Table 2.2]. For SACU, there is no evidence of export diversion, but the data do support some import diversion (βROW_RTA < 0) in the restricted samples; for example, imports from the ROW are 30 percent lower than the norm in the Africa-only sample. There is also evidence of import diversion for EAC countries, 10 percent below the norm in the Africa-only sample.

Table 2.3 presents the results using bilateral imports from the UN Comtrade database. Arguably, these data are more reliable because country authorities must collect import tariffs and might therefore better track their imports. As noted previously, gravity variables have the expected signs. The analysis also controls for the volatility of the exchange rate and infrastructure. In contrast with the previous data set, exchange rate volatility does not reduce bilateral trade. The negative effects from volatility could possibly be picked up by random effects.

Estimates based on the second data set reconfirm the previous finding that most RTAs in Africa have expanded trade. However, the results are not fully robust to the change in the data source (e.g., the benefits of EAC membership are no longer visible).9 The large estimated effect of SACU membership in the first data set is now significantly reduced and only present in the Africa-only sample.

Let us consider the SACU results a bit more closely. The public debate in SACU has often focused on the claim that there is a negative polarization impact on BLNS from being in a customs union with a much larger and more developed economy. However, in the Africa-only sample, the analysis estimates that SACU has a large positive effect on intra-SACU trade (βSACU > 0) without reducing trade with the ROW. This suggests that BLNS would have had close trade links with South Africa even in the absence of SACU, given the size of the South African economy, its proximity, the commonality of language, and other independent variables. In other words, trade diversion has been minimal, whereas trade expansion has been large, particularly relative to the African norm. Consequently, any possible negative externalities could be related to the fact that BLNS are in the periphery of a strong core (South Africa) rather than to the impact of SACU itself (see, for example, Ramcharan, 2009).

A closer inspection indicates that the gains from SACU come mainly from trade between South Africa and BLNS—and not from trade among BLNS. The effect of separate trade flows is estimated by replacing the SACU dummy variable with three variables: South African imports from the BLNS (ZAF_BLNS), BLNS imports from South Africa (BLNS_ZAF), and intra-BLNS trade (BLNS). As shown in Table 2.3, columns (2), (4), and (6), BLNS countries do trade more with South Africa, but no evidence indicates that this has reduced trade from the ROW.

Table 2.4 summarizes all results and the implied magnitude of trade creation,10 providing reliable evidence of trade creation for SACU, SADC, COMESA, ECOWAS, and WAEMU. The evidence for EAC, and particularly for CEMAC, is weaker, and not robust to the change in database. SACU appears to outperform other RTAs in Africa, although this result is muted in the UN Comtrade samples.

The magnitude of the SACU effect can be presented by graphing trade creation across time, as suggested by Magee (2008).11Figure 2.3 presents four panels representing (1) trade creation in nominal levels; (2) trade creation as a percentage of own counterfactual exports without SACU (that is, the sum of predicted intra-SACU exports in the absence of SACU, and exports to the ROW in the absence of SACU); (3) trade creation as a percentage of total exports; and (4) trade creation as a percentage of predicted intra-SACU exports in the absence of SACU. The panels show that, although South Africa gains more in nominal levels, BLNS appear to gain more using alternative measures.

Figure 2.3Southern African Customs Union: Trade Creation, 2000–07

Source: IMF staff estimates.

Note: BLNS = Botswana, Lesotho, Namibia, and Swaziland; SACU = Southern African Customs Union.

A number of caveats should be noted. First, it is possible that the analysis has not fully eliminated the possible endogeneity between bilateral trade and regional agreements and, if so, part of the estimated coefficients would reflect a correlation rather than a causal link. Second, the positive effect of bilateral trade agreements is most likely dynamic and growing over time as institutions are set up to deal with the demands of increased trade. This may explain, in part, the overperformance of the SACU arrangement: SACU is the oldest customs union in Africa and has well-established institutions that support trade. Third, compared with other customs unions, SACU member countries may have more incentives to report bilateral trade. Revenue sharing among SACU members is, in part, based on bilateral trade data, and members receive a higher share of collected revenues the higher their share of bilateral trade. Informal, unrecorded bilateral trade is therefore likely to be less of a problem than in other parts of Africa.12 Nevertheless, robust results for SACU across databases support the view that SACU membership seems to have promoted trade creation.

Conclusion

The analysis presented in this chapter finds that the SACU members have benefited from trade integration induced by SACU. The positive effects from trade expansion (increasing intra-SACU trade) are significant, whereas the negative effects from trade diversion are limited (or possibly zero).

Using two separate databases and measures of trade, the chapter finds solid evidence of trade creation, net of any potential trade diversion. Compared with a world reference based on bilateral exports data, intra-SACU trade is 57 times higher than expected. Compared with an Africa-only reference based on bilateral imports data, intra-SACU trade is 9 times higher than expected. Furthermore, South Africa has gained more from trade creation than BLNS if measured in nominal levels, but BLNS have gained more from trade creation measured in percentage of counterfactual exports without SACU.

The results in this chapter have implications for the ongoing discussion about how to reform SACU and its revenue-sharing formula. Trade creation has benefited all SACU members and future reforms should, therefore, support greater integration and not introduce new barriers to trade. Even if the evidence of trade diversion is unclear, a case for compensatory transfers can still be made if the structure of the common external tariff decidedly benefits one member of a customs union at the expense of other members. For example, a significant proportion of SACU import tariffs are collected on automobiles, an industry almost entirely based in South Africa. BLNS could, therefore, make legitimate claims that they should be compensated for extra-normal profits earned by South African automobile producers as a result of the common external tariff.

The chapter’s results cast doubt on the proposal to allocate customs revenues according to the final destination principle. Such a mechanism for revenue allocation would require establishing capacity for tracking and collecting revenues by country of final destination. It may, therefore, result in the erection of stricter fiscal frontiers between members and increase the cost of doing business and discourage intra-SACU trade. Stricter fiscal frontiers would be detrimental to the welfare of all SACU members.

Appendix 2A. Methodology and Estimation Issues

This chapter estimates a variant of the following equation:

in which xijt is bilateral trade (measured by exports or imports) between countries i and j at time t, Yi is real GDP in country i, Pit is population in country i, distij is the distance between countries i and j, RTAijt is a dummy variable that is equal to 1 if countries i and j share a trade agreement at time t, and Dij are time-invariant dummies that capture characteristics of the country pair (i.e., whether they share a common language or a border).13

A number of econometric issues must be addressed for this equation to be estimated successfully (see Baldwin and Taglioni, 2006, for a review of estimation issues relating to gravity equations):

  • Inappropriate deflation. Bilateral trade is in current U.S. dollars and must be deflated by a U.S. aggregate price index to obtain volumes. Such deflation, however, may create spurious correlations because there are global trends in inflation rates. The analysis follows Rose (2000) and others by adding time dummies to adjust for global trends in inflation rates.
  • Zero-trade observations. There are many zero observations for bilateral trade. The literature considered two options for dealing with them. One alternative is to assume that these observations are random measurement errors and drop them from the sample. Another option is to treat these observations as true values and transform the data so that they can be used in the estimation. The transformation log(1 + exports) is used here. Note that these observations may not be random, which would preferably require the use of censored data methods.
  • Nonlinearities. There may be nonlinearities in the relationship between bilateral trade and macroeconomic variables. These nonlinearities are captured by using a parsimonious quadratic specification:
  • Omitted variables. Countries with strong ties are likely to trade more and are more likely to form trade unions. Thus, the error term is correlated with the trade agreement dummy. This can be controlled for to some extent by introducing dummies that capture cultural and historical ties (common language and the like), but it is difficult to control for all those effects. Country-pair fixed effects can be introduced, which will capture the impact on trade of factors that do not change over time (distance, land area, cultural ties, and so on). Introducing country-pair fixed effects would normally require the removal of variables, such as distance and common border, that are time invariant. The analysis, therefore, models these country-pair effects ∈ij as random variables:

Note that some time-varying macroeconomic variables may be correlated with random effects. One way to deal with this issue is to use the Hausman and Taylor (1981) instrumental variable technique, which uses exogenous time-varying regressors from periods other than the current as instruments.14

  • Selection bias. Because the panel is unbalanced, a selection bias may be associated with the presence of a country pair in the sample. To correct for this potential, the analysis follows Nijman and Verbeek (1992) and adds three variables: the number of years a country pair has been present in the sample (total year), one dummy that is equal to 1 if the country pair is in the sample for the entire period (entire year), and another dummy that is equal to 1 if the country pair was present at t − 1 (difference_year).
  • Predicting bilateral trade. As discussed below, the estimation results will be used to compute predicted trade and various counterfactuals. Wooldridge (2006) discusses two ways of obtaining consistent predictions for the left-hand-side variable when it is expressed in logs. That method is used here, giving the most accurate prediction for trade. The predicted level of trade is x^=α0elog(x)^ where α0 is the coefficient in a linear regression (without an intercept) of x on log(x)^ and log(x)^ is the fitted value from the regression of ln(xijt) on all the right-hand-side variables.
  • Controlling for currency unions. A large literature exists on the positive effects of currency unions on bilateral trade (a well-known example is Rose, 2000). In empirical papers, currency unions are usually controlled for by adding a 0–1 dummy variable. In many cases in Africa, however, membership in a currency union is indistinguishable from membership in an RTA (e.g., SACU). The alternative is to follow Carrère (2004) and use the volatility of the bilateral nominal exchange rate to control for the exchange rate regime. Volatility is defined as the standard deviation of the log change in the monthly bilateral nominal exchange rate.
  • Controlling for infrastructure. Although the analysis uses the traditional gravity model assumption that transport costs are captured by distance, there is evidence that this variable captures only a small fraction of overall transport costs (Limao and Venables, 2001). Infrastructure, for example, the communication and transportation network, should be an important determinant of transport costs, and ultimately bilateral trade. The analysis follows Limao and Venables (2001) and Carrère (2004) and measures infrastructure as the average of the density of paved roads, railway, and the number of telephone lines per capita.
  • Measuring trade expansion, diversion, and creation. The gravity equation can be used to estimate the extent and magnitude of trade diversion. Following Carrère (2004), the analysis estimates the following equation:

in which

  • RTA_ROW is equal to 1 if the exporting country is a member of an RTA and the importing country is from the ROW, 0 otherwise.
  • ROW_RTA is equal to 1 if the exporting country is from the ROW and the importing country is a member of an RTA.

Trade expansion, diversion, and creation can be defined as follows when the dependent variable is exports:

  • Trade expansion: βRTA > 0.
  • Trade creation: βRTA > 0 and βRTA–ROW = 0, that is, intra-bloc trade increases without reducing exports to the ROW.
  • Import diversion: βRTA > 0 and βROW–RTA < 0, that is, intra-bloc trade increases but imports from the ROW fall.
  • Export diversion: βRTA > 0 and βRTA–ROW < 0, that is, intra-bloc trade increases but exports to the ROW fall.

The magnitude of trade creation can also be estimated (see Magee, 2008). Trade expansion is associated with increases in intra-bloc trade above the trade level that would have occurred within the trading bloc in the absence of an agreement. Setting RTAij, RTA_ROWij and RTA_RTAij to zero gives an estimate of the level of bilateral exports in the absence of the RTA. Then trade expansion can be defined as:

in which

is country j ‘s total exports from other countries within the RTA, and

is the predicted level of exports from other countries in the RTA in the absence of the RTA.

Trade diversion is implied if the rise in intra-bloc trade is accompanied by a decline in extra-bloc trade:

in which

is country j’s total exports from other countries outside of the RTA, and

is the predicted level of exports from other countries outside of the RTA if it had not been signed.

Trade creation is occurring if part of the increase in intra-bloc trade is not offset by a decline in extra-bloc trade:

Appendix 2B. Sample Data and Sources
Table 2B.1Sample of Countries
Afghanistan
Albania
Algeria
Angola
Argentina
Armenia
Australia
Austria
Bahrain
Bangladesh
Belarus
Belgium
Belize
Benin
Bolivia
Brazil
Bulgaria
Burkina Faso
Burundi
Cambodia
Cameroon
Canada
Cape Verde
Central African Rep.
Chad
Chile
China
Comoros
Congo, Dem. Rep. of
Congo, Republic of
Costa Rica
Côte d’Ivoire
Croatia
Cyprus
Czech Republic
Denmark
Djibouti
Dominica
Egypt
El Salvador
Equatorial Guinea
Estonia
Ethiopia
Fiji
Finland
France
Gabon
Gambia, The
Georgia
Germany
Ghana
Greece
Grenada
Guatemala
Guinea
Guinea-Bissau
Guyana
Haiti
Honduras
Hungary
Iceland
India
Indonesia
Ireland
Israel
Italy
Jamaica
Japan
Jordan
Kazakhstan
Kenya
Korea
Kyrgyz Republic
Lao P.D.R.
Latvia
Lebanon
Liberia
Libya
Lithuania
Luxembourg
Macedonia (FYR Macedonia)
Madagascar
Malawi
Malaysia
Maldives
Mali
Mauritius
Mexico
Moldova
Mongolia
Morocco
Mozambique
Nepal
Netherlands
New Zealand
Niger
Nigeria
Norway
Pakistan
Panama
Papua New Guinea
Paraguay
Peru
Philippines
Poland
Portugal
Romania
Russian Federation
Rwanda
St. Kitts and Nevis
St. Lucia
St. Vincent and the Grenadines
Samoa
São Tomé and Príncipe
Saudi Arabia
Senegal
Seychelles
Sierra Leone
Singapore
Slovak Republic
Solomon Islands
South Africa
Spain
Sri Lanka
Sudan
Suriname
Sweden
Switzerland
Tajikistan
Tanzania
Thailand
Togo
Tonga
Trinidad and Tobago
Tunisia
Turkey
Uganda
Ukraine
United Kingdom
Uruguay
Vanuatu
Venezuela
Vietnam
Zambia
Table 2B.2Data
VariableDefinitionSources
Nominal exportsBilateral nominal exports (millions of U.S. dollars)DOTS
Real exportsBilateral nominal exports deflated by U.S. CPIAuthors’ calculation
Nominal importsBilateral nominal imports (millions of U.S. dollars)UN Comtrade database
Real importsBilateral nominal imports deflated by U.S. CPIAuthors’ calculation
U.S. CPIU.S. consumer price indexIFS
Real GDPGross domestic product, constant international dollarsWDI
PopulationPopulationWDI
VolatilityStandard deviation of the log change in the monthly bilateral nominal exchange rateIFS and authors’ calculation
DistanceGeodesic distance between largest citiesCEPII
AreaArea in kilometers squaredCEPII
Landlocked=1 if the country is landlocked, 0 otherwiseCEPII
Locked=landlockedi + landlockedjAuthors’ calculation
Contiguous=1 if the two countries are contiguous, 0 otherwiseCEPII
Colony=1 if the two countries ever had a colonial link, 0 otherwiseCEPII
Common language=1 if the two countries share a common official language, 0 otherwiseCEPII
SACUij=1 if both countries i and j are members of the Southern African Customs Union, 0 otherwiseWTO and authors’ calculation
ZAF_BLNS=1 if exporting country is South Africa and importing country is BLNS, 0 otherwise in DOTS data; =1 if importing country is South Africa and exporting country is BLNS, 0 otherwise in UN Comtrade dataAuthors’ calculation
BLNS_ZAF=1 if exporting country is BLNS and importing country is South Africa, 0 otherwise in DOTS data; =1 if importing country is BLNS and exporting country is South Africa, 0 otherwise in UN Comtrade dataAuthors’ calculation
SADCij=1 if countries i and j are members of the Southern African Development Community and neither country i nor j is a member of SACU, 0 otherwise.WTO and authors’ calculation
SADC_ROWij=1 if exporting country (importing) is a member of SADC in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
ROW_SADCij=1 if exporting (importing) country is a member of SADC in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
COMESAij=1 if countries i and j are members of the Common Market for Eastern and Southern Africa, neither country i nor j is a member of SADC, and neither country i nor j is a member of EAC; 0 otherwiseWTO and authors’ calculation
COMESA_ROWij=1 if exporting country (importing) is a member of COMESA, 0 otherwise in DOTS database (UN Comtrade database)WTO and authors’ calculation
ROW_COMESAij=1 if exporting (importing) country is a member of COMESA in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
ECOWASij=1 if both countries i and j are members of the Economic Community of West African States and neither country i nor j is a member of WAEMU, 0 otherwiseWTO and authors’ calculation
ECOWAS_ROWij=1 if exporting country (importing) is a member of ECOWAS in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
ROW_ECOWASij=1 if exporting (importing) country is a member of ECOWAS in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
EACij=1 if both countries i and j are members of the East African Community, 0 otherwiseWTO and authors’ calculation
EAC_ROWij=1 if exporting (importing) country is a member of EAC in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
ROW_EACij=1 if exporting (importing) country is a member of EAC in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
CEMACij=1 if both countries i and j are members of the Central African Economic and Monetary Community, 0 otherwiseWTO and authors’ calculation
CEMAC_ROWij=1 if exporting (importing) country is a member of CEMAC in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
ROW_CEMACij=1 if exporting (importing) country is a member of CEMAC in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
WAEMUij=1 if exporting country is a member of the West African Economic and Monetary Union, 0 otherwiseWTO and authors’ calculation
WAEMU_ROWij=1 if exporting (importing) country is a member of WAEMU in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
ROW_WAEMUij=1 if exporting (importing) country is a member of WAEMU in DOTS database (UN Comtrade database), 0 otherwiseWTO and authors’ calculation
ECij=1 if both countries i and j are members of the European Community, 0 otherwiseWTO and authors’ calculation
NAFTAij=1 if both countries i and j are members of the North American Free Trade Agreement, 0 otherwiseWTO and authors’ calculation
ANDEANij=1 if both countries i and j are members of the Andean Community, 0 otherwiseWTO and authors’ calculation
MERCOSURij=1 if both countries i and j are members of the Southern Common Market, 0 otherwiseWTO and authors’ calculation
AFTAij=1 if both countries i and j are members of the ASEAN Free Trade Area, 0 otherwiseWTO and authors’ calculation
Note: BLNS = Botswana, Lesotho, Namibia, and Swaziland. CEPII = Centre d’Etudes Prospectives et d’Informations Internationales; CPI = consumer price index; DOTS = IMF, Direction of Trade Statistics; IFS = IMF, International Financial Statistics; WDI = World Bank, World Development Indicators; WTO = World Trade Organization.
Note: BLNS = Botswana, Lesotho, Namibia, and Swaziland. CEPII = Centre d’Etudes Prospectives et d’Informations Internationales; CPI = consumer price index; DOTS = IMF, Direction of Trade Statistics; IFS = IMF, International Financial Statistics; WDI = World Bank, World Development Indicators; WTO = World Trade Organization.
Table 2B.3Data for BLNS in the Direction of Trade Statistics Sample
VariableDefinitionSources
Real exportsBilateral nominal exports deflated by U.S. CPIAuthors’ calculation
Nominal exportsSum of bilateral exports for BLNSUN Comtrade for trade between BLNS and South Africa, DOTS for trade between BLNS and the rest of the world, and authors’ calculation
U.S. CPIU.S. consumer price indexIFS
Real GDPSum of GDP for BLNSWDI and authors’ calculation
PopulationSum of population for BLNSWDI and authors’ calculation
VolatilityWeighted average of standard deviation of the log change in the monthly bilateral nominal exchange rate (weights are population shares)IFS and authors’ calculation
DistanceWeighted average of geodesic distance between largest cities (weights are population shares)CEPII and authors’ calculation
AreaSum of area in kilometers squared for BLNSCEPII and authors’ calculation
Landlocked=1CEPII and authors’ calculation
Locked=landlocked + landlockedAuthors’ calculation
Contiguous=1 if the individual dummy is 1 for Botswana, Lesotho, Namibia, or Swaziland, 0 otherwiseCEPII and authors’ calculation
Colony=1 if the individual dummy is 1 for Botswana, Lesotho, Namibia, or Swaziland, 0 otherwiseCEPII and authors’ calculation
Common language=1 if the individual dummy is 1 for Botswana, Lesotho, Namibia, or Swaziland, 0 otherwiseCEPII and authors’ calculation
SACUSouthern African Customs Union, =1WTO and authors’ calculation
SADCSouthern African Development Community, =1WTO and authors’ calculation
COMESACommon Market for Eastern and Southern Africa, =1WTO and authors’ calculation
ECOWASEconomic Community of West African States, =0WTO and authors’ calculation
EACEast African Community, =0WTO and authors’ calculation
CEMACCentral African Economic and Monetary Community, =0WTO and authors’ calculation
WAEMUWest African Economic and Monetary Union, =0WTO and authors’ calculation
ECEuropean Community, =0WTO and authors’ calculation
NAFTANorth American Free Trade Agreement, =0WTO and authors’ calculation
ANDEANAndean Community, =0WTO and authors’ calculation
MERCOSURSouthern Common Market, =0WTO and authors’ calculation
AFTAASEAN Free Trade Area, =0WTO and authors’ calculation
ZAF_BLNS=1 if exporting country is South Africa and importing country is BLNS, 0 otherwiseAuthors’ calculation
BLNS_ZAF=1 if exporting country is BLNS and importing country is South Africa, 0 otherwiseAuthors’ calculation
Note: BLNS = Botswana, Lesotho, Namibia, and Swaziland; CEPII = Centre d’Etudes Prospectives et d’Informations Internationales; CPI = consumer price index; DOTS = IMF, Direction of Trade Statistics; IFS = IMF, International Financial Statistics; WDI = World Bank, World Development Indicators; WTO = World Trade Organization.
Note: BLNS = Botswana, Lesotho, Namibia, and Swaziland; CEPII = Centre d’Etudes Prospectives et d’Informations Internationales; CPI = consumer price index; DOTS = IMF, Direction of Trade Statistics; IFS = IMF, International Financial Statistics; WDI = World Bank, World Development Indicators; WTO = World Trade Organization.
The authors would like to thank Vitaliy Kramarenko, participants in the IMF African Department seminar series, and other colleagues in the IMF and the World Bank for helpful comments.
1Theoretical underpinnings with clear microfoundations first emerged with Anderson (1979) followed by Bergstrand (1985). More recent contributions include Anderson and Van Wincoop (2001).
2See Baldwin and Venables (1995) for a review of this literature and Caliendo and Parro (2011) for a recent application to the North American Free Trade Agreement (NAFTA).
3As noted in the introduction, this work is similar to Carrère (2004). For a more detailed discussion of estimation issues, see Appendix 2A, Methodology and Estimation Issues.
4See Baldwin and Taglioni (2006) for a review of estimation issues relating to gravity equations.
5A well-known example is Rose (2000).
6This estimation only covers 1990–2008.
7For more details, see Appendix 2B, Sample Data and Sources.
8Although these results are not shown, the full-sample regressions also include controls for non-African RTAs: ASEAN, EU, MERCOSUR, and NAFTA.
9This could be the result of dynamic effects. The EAC customs union was officially launched in 2005 but allowed for a five-year transition during which member states could maintain internal tariffs on certain sensitive products.
10The numbers in Table 2.4 are computed using columns (2), (4), and (6) in Tables 2.2 and 2.3. Trade effects are computed as eβ^ if β^>0(1eβ^) if β^<0, where β^ is the estimated coefficient. Note that in the DOTS database, the import diversion coefficient is the coefficient on ROW_RTA and export diversion corresponds to RTA_ROW. The reverse is true for the UN Comtrade database.
11We estimate a regression similar to column 1 in Table 2.2. Setting RTAij, RTA_ROWij, and ROW_RTAij to zero gives an estimate of the level of bilateral exports in the absence of the RTA. Details about the measure of trade expansion using the Magee methodology are given in Appendix 2A.
12SACU members may even have an incentive to over-report their bilateral trade.
13This work is similar to Carrère (2004). That analysis, however, excludes SACU and covers 1962–96. The present work focuses on the more recent period of 1990–2008.
14Let X = [X1, X2] be a matrix of time-varying dependent variables, where X1 is exogenous and X2 is endogenous, and Z denotes a matrix of time-invariant dependent variables. The appropriate matrix of instruments is [QX1, QX2, PX1, Z] where Q is a matrix that obtains deviations from individual means and P averages observations across time for each country.

    Other Resources Citing This Publication