Journal Issue
Share
Article

Are Africa’s Currency Unions Good for Trade?

Author(s):
International Monetary Fund. Research Dept.
Published Date:
November 2009
Share
  • ShareShare
Show Summary Details

Newton’s theory of universal gravitation relates the force of attraction between two objects to their respective masses and the distance between them. Tinbergen (1962) and Pöyhönen (1963) proposed that the same principle may be applied to explain bilateral trade flows where trade is estimated as an increasing function of the trading partners’ incomes, and a decreasing function of the distance between them. Although simply specified, gravity models have performed well empirically, and are widely applied in empirical trade analysis.1

In his seminal work, Rose (2000) uses gravity models to empirically investigate the impact of currency arrangements on bilateral trade. He concludes that the effect of common currency on international trade is “large”: two countries sharing a currency tend to trade roughly three times as much as they would otherwise. Rose and van Wincoop (2001), Frankel and Rose (2002), and Glick and Rose (2002) confirm this result, and show that it is robust to various specifications and estimation techniques. Frankel and Rose (2002) further find that the growth-enhancing benefits of currency unions are achieved through increased trade only and not through other channels (such as reduced inflation).

These results have generated immense interest and controversy in the academic community, and numerous studies have followed. Most of these studies point out methodological or data limitations in the earlier analyses and find a smaller impact of currency unions on trade (see, for example, Tenreyro, 2001; Nitsch, 2002; and Baxter and Kouparitsas, 2006). In general, these studies challenge the magnitude of the estimated effect of currency unions in the research conducted by Rose, but do not question the validity of its existence or the direction of the effect.2

From a theoretical standpoint, currency unions promote trade through various channels. For example, they reduce transaction costs between trading partners and facilitate exchange; create a larger market in which there are potential gains from economies of scale and production efficiency (assuming factor mobility and flexible wages); bring macroeconomic stability by signaling the central bank’s commitment to reduce inflation; enhance the credibility of the monetary authority; and reduce uncertainty. Less uncertainty about prices can help allocate resources more efficiently in a region, while the absence of exchange rate risk in countries sharing a common currency encourages investment, facilitates capital mobility, and helps to generate trade.

Nonetheless, the potential gains from forming a currency union must be compared with the potential losses from membership. The potential losses depend on the extent to which countries in the union face asymmetric shocks, and whether countries in the region are sufficiently flexible to absorb or mitigate such shocks. Flexibility implies that factors are mobile and an appropriate system of transfers between countries is in place to act as a shock absorber and reduce costs of asymmetric shocks even if the shocks are large.

This paper extends the existing literature on how currency unions affect trade in three important directions. First, the main focus of the paper is Africa. We concentrate on Africa because it has a rich history of currency unions, although its integration in world trade has remained limited.3 With ongoing debates over existing and proposed monetary unions in Africa, the benefits of forming a currency union are at the forefront of policy agendas and debates across the continent. We empirically investigate certain stylized facts of currency unions in Africa—including trade diversion and the impact of currency unions on price and output comovements, and on trade stability—and compare them with the world estimates to examine whether Africa’s experience is different. Second, we investigate the time dimensional aspects of currency unions by estimating the effects of dissolution of a currency union on trade and the effect of the duration of currency union membership on bilateral trade flows. As the effect of joining a currency union on trade may evolve over time, it is important to quantify the changes (if any) in the trade-generating effect of currency unions. Third, we conduct a comprehensive set of robustness checks to address the methodological and econometric concerns often highlighted in estimating gravity models.

Our empirical findings suggest that African countries stand to benefit at least as much from currency union membership as other countries in the world; therefore, currency union benefits are not region-specific. Specifically, our results for both the world and Africa show that (1) countries belonging to a currency union trade, on average, about one-and-a-half times more with each other than with other comparable countries that do not share a currency; (2) currency unions are associated with trade creation and increased comovement of prices, but not with the comovement of output among members; (3) the effect of currency unions on trade stability is ambiguous; and (4) the duration of sharing a common currency matters significantly for bilateral trade: the longer a country participates in a currency union, the greater the benefit is, albeit with some diminishing returns. Although the trade-stimulating effect of currency unions holds significantly for Africa, given the continent’s small trade base and the protracted period necessary following the establishment of a currency union to increase trade, our analysis suggests that currency union membership may not be a primary tool to achieve high levels of trade in the region, especially in the short run.

I. Literature Review

Baldwin’s (2005) survey of the evidence on how currency union has an impact on trade correctly places Andrew Rose’s research at the center of this literature. Rose and his coauthors find that membership in a currency union promotes trade substantially. In his seminal work, Rose (2000) shows that trade between countries belonging to the same currency union is about three times larger than trade between comparable countries that do not share a currency. His study also shows that the effect of lower exchange rate volatility is positive, although the coefficient is much smaller than that associated with currency unions. In fact, the increase in trade associated with currency unions is much larger in order of magnitude than that associated with the complete elimination of exchange rate volatility.

Rose presents several arguments for his results, such as a common currency can induce financial integration that has consequences for trade. He also argues that by entering a currency union, a government signals its commitment to long-term integration, thereby promoting trade.4Rose and van Wincoop (2001) investigate further the trade-generating effect of currency union membership by using Anderson and van Wincoop’s (2003) structural model to address country-specific idiosyncrasies. This approach, which was applied only to countries with complete bilateral data, somewhat reduces the effect of currency unions on trade to about two-and-a-half times. Glick and Rose (2002) explore the time-series dimension of the data, and introduce country-pair specific intercepts in the gravity model. They find a relatively smaller, but still large, impact of currency unions on trade: joining a currency union almost doubles bilateral trade. Although Rose’s subsequent research and other studies on the subject have established different magnitudes of the estimated currency union effect, there seems to be general agreement in the literature on the existence of a common-currency effect, which, as suggested by Frankel (2005), “is probably substantially smaller than a tripling.”5

A growing body of evidence describes the potential channels through which currency unions may affect trade. For example, Alesina and Barro (2002) show that besides country size and the volume of trade exchange with a potential anchor, the comovement of prices, outputs, inflation, and the volatility of inflation matter for currency unions. Thus, countries that stand to gain the most from joining currency unions are those having the largest comovements of outputs and prices with a potential anchor, and those with a history of high and volatile inflation. Tenreyro (2001) tests the predictions of Alesina and Barro (2002) using a probit model, and finds that countries with a higher comovement of prices have a higher propensity to form a currency union, but that the comovement of output has no effect on a country’s decision. Alesina and Tenreyro (2002) and Tenreyro and Barro (2003) analyze the impact of currency unions on the comovement of output and prices between trading partners and investigate how comovements of outputs and prices would respond to the formation of a currency union. They find that sharing a currency enhances trade and increases price comovements, but decreases the comovement of shocks to real gross domestic product (GDP) (that is, increases specialization).

Keeping in view the potential and estimated impact of currency unions on trade, Africa presents an interesting case to assess the relative impact of currency unions and free trade agreements (FTAs) on intraregional and international trade. This is because Africa has a rich history of currency unions and preferential regional trade agreements, but its participation in world trade remains limited. Various explanations have been proposed for Africa’s marginalization in global trade activity, including slow economic growth, unfavorable geographical and exogenous factors, poor infrastructure, ill-planned trade policies, weak governance and institutions, barriers to intraregional trade, and constraints on factor mobility. Further, the substantial savings on transaction costs that accrue from a monetary union and imply an increase in trade benefits may be limited in Africa because of lower diversification and a heavy dependence on primary commodities.6 The loss of nominal exchange rate flexibility makes real adjustments to asymmetric shocks more difficult, especially in view of the poor systems of fiscal transfers and the limited development of the banking and financial sectors in Africa.

In this context, Debrun, Masson, and Pattillo (2005) show that gains from adopting a common currency depend, among other factors, on the correlation of terms-of-trade shocks. This, in turn, is connected to the countries’ dependence on primary commodities and their prices. They also show that the existence of interest groups affects incentives to join a currency union or accept a new member in a multilateral union. This effect is noteworthy as it implies that differences in government spending propensities may be more important than asymmetric shocks for the benefits/losses arising from joining a currency union.

Coe and Hoffmaister (1999) analyze North-South trade and show that, on average, Africa trades more with the rest of the world (ROW) than other developing countries. However, Subramanian and Tamirisa (2003) stress the importance of distinguishing between Anglophone and Francophone countries while assessing the integration of African countries in global markets. Using Glick and Rose’s (2002) specification, Masson and Pattillo (2004) examine the impact of currency unions on trade in Africa. Their estimated effect of currency unions on African bilateral trade with the ROW is almost the same as for the world: currency unions increase trade threefold in both Africa and the world.7

This paper contributes to the existing literature by quantifying a series of “stylized facts” relating to the trade and currency union nexus in Africa, and investigating whether Africa’s experience is different from that of the rest of the world. In particular, we examine the trade-generating impact of currency union membership, the impact on trade creation, and the effect of currency unions on trade stability and on the comovement of prices and outputs. Importantly, we also investigate whether the duration of currency union membership has any significant effect on trade. To explore these issues, we apply an augmented gravity model of trade that includes variables for FTAs, years of currency union membership, and trade diversion, and use an extended data set that covers more countries and years than previous empirical work on the subject.

II. Methodology

Analytical Framework

Traditionally, gravity models represent trade between two economies as a function of their respective economic masses and obstacles to trade such as the distance between them. However, this basic model has been extended in recent years to incorporate a variety of other factors that may hinder or promote trade, for example, common language, historical ties, common border, geographical location, and so forth.

In line with recent literature, we begin by investigating the effect of currency unions on trade by defining the following augmented gravity model:

where i and j denote the exporting and importing countries, respectively; Xij denotes the value of bilateral trade between i and j; CUij is a binary variable that is unity if i and j share the same currency;8 γ is the estimate of currency union’s trade-generating effect; and Zk is a vector consisting of other variables that includes (log of) product of real GDP, real GDP per capita, land areas of the trading partners, and the distance between them, and dummy variables that are equal to 1 if the countries share an FTA, historical ties, language, or a border; are part of the same nation; or were colonies of the same colonizer after the year 1945, and 0 otherwise.

Next, we investigate the possibility that the stimulus to trade among members of a currency union comes at the expense of trade diversion with nonmembers. To do this, we follow Frankel and Rose (2002) and define a dummy variable that is unity if the trading partners are not in the same currency union but (at least) one is in a currency union with another country. A negative (and significant) coefficient of this variable would indicate the existence of potentially harmful trade diversion, and could be interpreted as implying that increased trade among members of a union comes at the expense of reduced trade with nonmembers.

Then, to investigate the impact of currency unions on comovements of output and prices, we construct the variables measuring comovement of prices and output, as in Alesina and Tenreyro (2002), and use them as dependent variables in equation (1). Specifically, the price comovement between countries is determined by estimating a second-order autoregressive equation of annual price data for every pair of countries with more than 20 observations:

Estimated residuals from equation (2) are then used to obtain a measure of comovement of prices with higher VPij representing greater synchronization of prices between countries i and j:

In a similar fashion, we construct a measure for the comovement of output, with û denoting estimated residuals from the following autoregressive process:

and

Next, to assess the currency union impact on trade stability, we follow Rose (2005) and estimate an equation similar to the gravity equation, with the coefficient of variation of log of real trade as the dependent variable of equation (1). We calculate values for the dependent variable for the periods 1950–76 and 1976–2003, so we have two observations per pair. In addition, as a robustness check of the dependent variable, we use (1) the maximum absolute value (during the 27-year sample period) of the difference between the log of real trade and the sample average of trade of every country, scaled by the sample average; (2) the mean absolute value of the difference between exports and their sample average of every country (scaled by average exports); and (3) the standard deviation of the residual from a conventional gravity equation of exports in levels. All the explanatory variables are averaged over the corresponding time periods.

Finally, to investigate the effects of membership duration, we construct another variable of interest—the number of years that a given trading partner has shared a common currency. We modify equation (1) to include this variable as follows:9

Estimation Issues

While estimating trade flows using the gravity model, several relevant methodological issues need to be discussed. These issues are derived from various critiques of the estimation of the gravity equation and include the three “classic gravity model mistakes” pointed out by Baldwin (2005), as well as the critique on the correct functional form of the gravity equation pointed out by Santos Silva and Tenreyro (2006).10 To the extent that these critiques relate to the analysis presented in this paper, we discuss our attempts to address them through robustness checks of the estimated results.

First, we begin with the issue of the omitted variables bias stemming from the correlation of any protrade omitted variables with the currency union dummy. This has been labeled as the “gold medal mistake” in Baldwin’s (2005) critique. Research following Rose (2000) attempts to control for this bias by introducing country-specific idiosyncrasies in the model, both in the context of cross-section and in panel estimations. In cross-section analysis, country fixed effects (using country-specific dummy variables) can be introduced to account for Anderson and van Wincoop’s (2003) “multilateral resistance” factor, according to which trade between two countries does not only depend on the characteristics of the countries but also on the barriers between them and the ROW.11 However, given that there is a time-series element to the potential bias that is not eliminated with this procedure, we employ a panel data fixed effects procedure (the fixed effects “within” estimator) that adds country-pair specific effects to the equation, and thus exploits the time-series dimension of the data around country-pair averages.

Second, as forming a currency union (or continuing to stay in a currency union) may also be an endogenous choice, some of the large trade-creating effects of currency union may actually be a reflection of reverse causality. The use of instrumental variables could be a solution to the potential endogeneity problem. However, an appropriate instrument for a currency union is hard to find, which is further complicated by the fact that currency union membership is proxied by a dummy variable. Nevertheless, attempts by Alesina and Tenreyro (2002) to address the endogeneity problem using an instrumental variable based on client-anchor relationship have shown that the effect of currency union on trade remains high even after accounting for this potential endogeneity.12 In addition, Rose and van Wincoop (2001) argue that “reverse causality also does not explain away the findings; there is little evidence in the political science literature that countries join currency unions to increase trade, and instrumental variables only increase the impact of currency unions on trade.” This political dimension is particularly important for the case of Africa. Masson and Pattillo (2004) underscore the political aspect of the decision to form or participate in a currency union and argue that the experience of Africa shows that political objectives are important to the formation of monetary unions.13 Hence, in our analysis we choose to treat currency unions as an exogenous variable with respect to trade.

Finally, there are several issues relating to model misspecification. These include (1) the aggregation of exports and imports as the dependent variable (the “silver medal mistake” of Baldwin); (2) inappropriate deflation of nominal trade values by the U.S. aggregate price index (the “bronze medal mistake” of Baldwin); (3) possible nonlinear effects entering the gravity equation; and (4) the treatment of zero-trade observations in the sample.14 We attempt to address all these issues in our robustness checks.

On the aggregation issue, some critics argue that although theory supports the use of bilateral exports as the dependent variable, the use of bilateral trade as the dependent variable without properly aggregating imports and exports can seriously bias the results. We address this critique by taking the dependent variable as the sum of the logarithms of exports and imports in addition to the logarithm of the sums, and re-estimating equation (1). To account for the potential bias arising from inappropriate deflation by the aggregate U.S. price index, we add time dummies. This procedure corrects for global trends in inflation rates, as every bilateral trade flow is divided by the same price index adjusted for time effects. To address the possibility of nonlinear effects operating in gravity equation estimations (for example, due to sample nonhomogeneity), we add quadratic terms for both output and output per capita as in Glick and Rose (2002).

The issue of zero-trade observations arises because many observations in bilateral trade data sets appear as zeros either because some pairs of countries did not trade, or because of rounding errors and missing observations. Using the log-linear form of the gravity equation as in equation (1) implies including only those observations for which the dependent variable is positive. Given that the value of trade flows between some pairs of countries—typically pairs of small countries—tends to be zero, this may lead to a sample selection problem. The truncation at zero may result in inconsistent estimators when ordinary least squares (OLS) are used.15 We check the sensitivity of our results to the inclusion of zero-trade observations mainly in two ways. First, we apply the Tobit estimation method to account for the censored nature of the dependent variable of the model. Second, we apply the pseudomaximum likelihood (PML) approach applied by Santos Silva and Tenreyro (2006), which takes the real value of trade as the dependent variable, and includes zero observations. An additional advantage of using the PML approach is that it may have a superior functional form than the log-linear gravity model. This is because, as noted by Santos Silva and Tenreyro (2006), Jensen’s inequality can have important implications for log-linear models in the presence of heteroscedasticity: if the error term is heteroscedastic with the variance depending on the regressors, then the parameters estimated by OLS can be severely biased.16 Also, as commonly done in empirical gravity model literature, we avoid the truncation of observations by adding a positive constant to all trade observations and taking the log, that is, we use log(constant + Xij) as the dependent variable.

Data

The data set used in this paper is an extended version of Glick and Rose’s data set. It includes 217 countries and political units over the time period 1948–2003, which constitute the world sample in our analysis.17 The Africa sample is a subset of the world sample, consisting of the bilateral trading patterns of 49 African countries. The intra-Africa and African trade with the ROW (Africa-ROW) samples include those pairs of countries where both trading partners are in Africa, and where only one partner is in Africa, respectively.

The data for the paper have been compiled from various sources. The annual bilateral trade observations are obtained from the IMF’s Direction of Trade Statistics and are expressed in real U.S. dollars using the U.S. consumer price index for all urban consumers; GDP and prices (purchasing power parity of GDP) data have been taken from University of Pennsylvania’s World Tables 6.1 and the World Bank’s World Development Report 2005; the terms-of-trade data are from the IMF’s World Economic Outlook 2005; information on colonial past, distance, and language has been compiled from the Central Intelligence Agency (CIA) World Factbook 2004; data on currency unions for the period 1998–2003 are from the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions; and data on FTAs and currency unions are from Glick and Rose (2002). Table A1 provides details on data construction and sources. Tables A2A4 present the countries, currency unions, and FTAs used in the study as well as the summary statistics for the variables of interest.

III. Results

The Trade-Generating Effect

We begin by investigating the general aspects of bilateral trade in the world and in Africa by using the extended data set and applying the gravity model to establish some benchmark results. In addition to the FTA dummy used by Glick and Rose (2002), we construct another FTA dummy variable that addresses an important limitation of Rose and Glick’s data set because it takes into account the FTAs with Africa, and uses it as an alternate variable. In the next step, we estimate the characteristics of bilateral trade in Africa and trade creation, and address robustness issues.

Benchmark results

Table 1 shows the benchmark results of estimating equation (1) using the extended sample. Following the literature, we use simple pooled OLS estimation (columns 1–3) and then employ the panel fixed effects estimation technique to address the potential omitted-variable bias by introducing the country-pair specific idiosyncratic effects (columns 4–6).18 The specification in column 1 uses the world sample and replicates the benchmark pooled results of Glick and Rose (2002) very closely: countries sharing a common currency trade about two-and-a-half times more than countries not involved in a currency union. The coefficients on the standard determinants of the gravity models, such as income, population, and distance, have the correct sign, are statistically significant, and yield plausible elasticity estimates broadly in line with those obtained in earlier literature. In column 2, we allow the free trade area dummy to include additional agreements, especially those operating in Africa. The coefficient of currency union decreases slightly to 2.3 compared with the original Glick and Rose specification.

Table 1.Benchmark Results
Sample:WorldWorldAfricaWorldWorldAfricaWorldWorldAfrica
Estimation:OLSOLSOLSFixed effectsFixed effectsFixed effectsPMLPMLPML
Specification:(1)(2)(3)(4)(5)(6)(7)(8)(9)
Currency union1.00***0.84***0.99***0.55***0.60***0.54**0.15**0.15**0.65***
(8.55)(7.75)(6.57)(14.83)(16.15)(7.91)(2.49)(2.51)(5.05)
Log distance−1.08***−1.05***−1.06***
(52.68)(52.08)(19.77)
Log product real GDP0.92***0.93***1.01***0.43***0.40***0.14**1.01***1.01***1.14***
(104.20)(105.38)(61.62)(27.34)(25.14)(3.78)(16.52)(16.52)(12.62)
Log product real GDP/capita0.42***0.44***0.34***0.36***0.40***0.45**0.000.00−0.19**
(32.28)(33.80)(14.77)(23.66)(26.73)(11.83)(0.03)(0.03)(2.55)
Common language0.36***0.35***0.29***
(9.25)(9.31)(4.66)
Common land border0.52***0.40***1.22***
(4.71)(3.85)(7.30)
Free trade agreement (FTA)1.14***0.75***0.47***
(10.40)(33.41)(3.38)0.47***0.61***
FTA (with Africa)1.19***0.92***0.33***0.15**(3.40)(3.96)
−14.32−7.46(13.36)(3.19)
Number landlocked in the pair−0.23***−0.25***−0.35***
(8.17)(8.82)(8.86)
Number islands in the pair0.030.03−0.31***
(0.92)(0.73)(5.00)
Log product of areas−0.08***−0.08***−0.18***
(10.75)(11.26)(14.70)
Common colonizer0.55***0.51***0.30***
(8.78)(8.27)(3.48)
Current colony0.95***0.99***−0.390.34***0.32***−0.100.94**0.94**0.10
(3.91)(4.02)(0.86)(7.98)(7.50)(1.25)(2.34)(2.33)(0.55)
Ever colony1.31***1.32***2.14***
(10.75)(10.97)(14.42)
Same nation−0.20−0.231.63***
(0.20)(0.22)(3.30)
Observations265,262265,262100,597265,262265,262100,597380,512380,512176,712
R2 (within)0.140.140.03
R2 (between)0.60.590.30
R2 (overall)0.680.680.520.550.540.24
Ramsey F-test (p-value)0.000.000.00
Fixed effects F-test (p-value)0.000.000.000.000.000.00
Hausman test (p-value)0.000.000.000.000.000.00
Wald χ2 (p-value)0.000.000.00
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the gravity model of bilateral trade using the world and Africa (at least one country in the pair is in Africa) samples. The dependent variable is log of real trade between trading partners (log(Xij)) in the OLS and fixed effects regressions, and real trade between trading partners (Xij) in the PML regressions. Real trade is defined as the average of exports and imports between the trading partners deflated by the U.S. consumer price index. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement (Glick and Rose’s (2002) dummy variable equal to 1 if the two countries share a free trade agreement, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). Year dummies included in all regressions. The table reports robust t-statistics in parentheses of columns 1–6, and robust z-statistics in columns 7–9. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the gravity model of bilateral trade using the world and Africa (at least one country in the pair is in Africa) samples. The dependent variable is log of real trade between trading partners (log(Xij)) in the OLS and fixed effects regressions, and real trade between trading partners (Xij) in the PML regressions. Real trade is defined as the average of exports and imports between the trading partners deflated by the U.S. consumer price index. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement (Glick and Rose’s (2002) dummy variable equal to 1 if the two countries share a free trade agreement, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). Year dummies included in all regressions. The table reports robust t-statistics in parentheses of columns 1–6, and robust z-statistics in columns 7–9. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.

Column 3 presents the estimation results for Africa. Interestingly, the currency union trade-generating effect is larger than for the world sample, and the marginal impacts of other determinants change.19 The effect of sharing a border is now much larger, reflecting the poor transportation links between many African countries and the tendency to trade more with neighboring countries. The impact of having a common language is however less important in Africa, as the currency unions in the sample are the Francophone countries of West and Central Africa and the South Africa Common Monetary Area. The variable reflecting having been a colony (ever colony) has a greater impact that reflects the colonial ties of African countries. Surprisingly, there is a reversal in the sign of the coefficient of number of islands, which is hard to interpret. However, this may be a result of the omitted-variable bias caused by ignoring country-specific fixed effects.

The Ramsey regression specification error test for omitted variables confirms the existence of omitted variables in the three pooled regressions. Thus, in columns 4–6, we re-estimate the earlier specifications as a panel and add a set of country-pair specific intercepts to the equation.20 The F-test for the joint significance of the fixed effects shows that the fixed effects are significant in each case. The magnitude of the trade-generating coefficient changes, which suggests that not accounting for country-specific idiosyncratic effects may lead to biased estimates. For both the world and Africa, the size of the currency union coefficient falls, and the trade-generating effect becomes 1.8 and 1.7 for the world and Africa, respectively.

Turning to the issue of zero-trade observations, columns 7–9 in Table 1 present the results when the PML approach is used. Not eliminating the zero-trade observations increases the world and Africa samples by about 45 and 31 percent, respectively. The results show that the effect of currency unions is still positive and significant in all cases but much smaller in magnitude. Currency unions now increase trade by a factor of about 0.2 and 0.7 in the world and in Africa, respectively. The inclusion of zero observations reduces the trade-generating elasticity for the world sample by more than the elasticity for Africa; hence, the effect of currency unions on trade is now larger for Africa than for the world.21

Characteristics of trade in Africa

Next, we investigate the characteristics of Africa’s trade in more detail. Table 2 presents the results of the relative trade performance of African countries by introducing dummies for intra-Africa trade, and trade between Africa and the ROW in equation (1). We first estimate the augmented equation for the world sample for comparison purposes and observe that the effect of currency unions on trade changes only marginally: the trade-generating effect of a currency union is now 2.2. Both the Africa-ROW and intra-Africa trade dummies are positive and highly significant, indicating that if at least one of the bilateral trading partners is in Africa, trade increases by no less than 1.3 times.

Table 2.Africa Trade Details
Sample:WorldAfrica-ROWIntra-AfricaWorldWorldAfrica-ROWIntra-AfricaWorldAfrica-ROWIntra-Africa
Estimation:OLSOLSOLSOLSFixed effectsFixed effectsFixed effectsPMLPMLPML
Specification:(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)
Currency union0.79***0.530.55***0.76***0.60***0.55***0.43***0.15**0.60***0.62***
(7.03)(1.18)(2.69)(6.23)(16.15)(7.77)(4.94)(2.51)(4.30)(2.70)
Africa-ROW0.27***
(7.90)
Intra-Africa0.42***
(4.58)
Log distance−1.06***−0.96***−1.29***−1.06***
(52.00)(15.64)(9.92)(-52.27)
Log product real GDP0.94***1.06***0.62***0.93***0.40***0.11***0.88***1.01***1.18***0.60***
(104.26)(64.47)(9.83)(104.11)(25.14)(2.78)(4.59)(16.52)(12.34)(5.05)
Log product real GDP/capita0.48***0.33***0.50***0.47***0.40***0.51***−0.46**0.00−0.13−0.10
(35.94)(13.95)(5.80)(35.04)(26.73)(11.79)(2.44)(0.03)(1.54)(0.79)
Common language0.33***0.27***0.40**0.34***
(8.76)(4.26)(2.01)(8.76)
Common land border0.43***1.10***0.98***0.42***
(4.17)(3.21)(4.27)(4.04)
Free trade agreement (with Africa)1.15***0.78***1.18***0.33***0.31***0.47***0.30**
(13.08)(5.24)(13.81)(13.36)(5.56)(3.40)(2.09)
Number landlocked in the pair−0.28***−0.37***−0.33***−0.28***
(9.97)(8.82)(3.07)(-9.82)
Number islands in the pair0.04−0.39***0.170.03
(1.09)(6.07)(0.83)(0.88)
Log product of areas−0.09***−0.20***−0.01−0.09***
(12.35)(16.14)(0.15)(-11.88)
Common colonizer0.49***0.32***0.250.51***
(7.91)(3.46)(1.09)(8.02)
Current colony0.98***−0.031.02***0.32***−0.070.94**0.11
(3.92)(0.06)(3.99)(7.50)(0.82)(2.33)(0.59)
Ever colony1.30***2.12***1.31***
(11.25)(14.08)(11.16)
Same nation−0.31

(0.31)
1.47***

(2.62)
−0.30

(-0.30)
Intra-Francophone0.36**

(2.13)
Intra-Anglophone0.02

(0.10)
Francophone-ROW0.19***

(4.81)
Anglophone-ROW0.22***

(5.19)
Observations265,26285,75914,838265,262265,26285,75914,838380,512146,46830,244
R2 (within)0.140.040.02
R2 (between)0.590.270.160.000.000.00
R2 (overall)0.680.540.400.680.540.210.110.000.000.00
Hausman test (p-value)0.000.000.000.000.000.00
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the gravity model of bilateral trade using the world, Africa (at least one country in the pair is in Africa), intra-Africa (both countries in the pair are in Africa), and Africa-rest of the world (ROW) (only one country in the pair is in Africa) samples. The dependent variable is log of real trade between trading partners (log(Xij)) in the OLS and fixed effects regressions, and real trade between trading partners (Xij) in the PML regressions. Real trade is defined as the average of exports and imports between the trading partners deflated by the U.S. consumer price index. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise); intra-Francophone (dummy variable equal to 1 if the both partners are African Francophone countries, and 0 otherwise); intra-Anglophone (dummy variable equal to 1 if both partners are African Anglophone countries, and 0 otherwise); Francophone-ROW (dummy variable equal to 1 if only one country in the pair is an African Francophone country, and 0 otherwise); and Anglophone-ROW (dummy variable equal to 1 if only one country in the pair is an African Anglophone country, and 0 otherwise). Year dummies included in all regressions. The table reports robust t-statistics in parentheses of columns 1–7, and robust z-statistics in columns 8–10. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the gravity model of bilateral trade using the world, Africa (at least one country in the pair is in Africa), intra-Africa (both countries in the pair are in Africa), and Africa-rest of the world (ROW) (only one country in the pair is in Africa) samples. The dependent variable is log of real trade between trading partners (log(Xij)) in the OLS and fixed effects regressions, and real trade between trading partners (Xij) in the PML regressions. Real trade is defined as the average of exports and imports between the trading partners deflated by the U.S. consumer price index. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise); intra-Francophone (dummy variable equal to 1 if the both partners are African Francophone countries, and 0 otherwise); intra-Anglophone (dummy variable equal to 1 if both partners are African Anglophone countries, and 0 otherwise); Francophone-ROW (dummy variable equal to 1 if only one country in the pair is an African Francophone country, and 0 otherwise); and Anglophone-ROW (dummy variable equal to 1 if only one country in the pair is an African Anglophone country, and 0 otherwise). Year dummies included in all regressions. The table reports robust t-statistics in parentheses of columns 1–7, and robust z-statistics in columns 8–10. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.

Columns 2 and 3 of Table 2 present the results when the sample is confined to those trading pairs where one partner country is in Africa, and where both partner countries are in Africa, respectively. In both cases, membership in a currency union increases trade by about 1.7 times. In columns 5–7 panel fixed effects “within” estimates are shown for the world, Africa-ROW, and intra-Africa samples, respectively. The results confirm that currency union participation in Africa has beneficial intraregional trade-generating effects as well as Africa-ROW trade-generating effects: the trade-generating effect is 1.7 for intra-African trade and 1.9 for Africa-ROW trade. Columns 8–10 employ the PML estimation technique. Once again, the magnitude of the coefficient for currency union is smaller than for pooled OLS and fixed-effects estimators.

Finally, following Subramanian and Tamirisa (2003), we also investigate whether Anglophone and Francophone African countries differ in their trading characteristics (column 4 of Table 2). The results show that the trade-generating effect remains about 2, but there is some evidence of differences in the trade patterns of Francophone and Anglophone Africa. The intra-Francophone trade coefficient is significant, suggesting that trade among the Francophone Africa countries is about 43 percent more than the average. This suggests that even after controlling for the currency union effect (essentially the effects of the CFA franc zone) trade among Francophone Africa has increased. We also find evidence of increasing integration of both Francophone and Anglophone Africa with the ROW, confirming the recent pattern of increase in intra-African trade as well as Africa-Europe trade. The coefficients of trade with the ROW are positive and significant, suggesting that in Francophone and Anglophone Africa, overall trade is about 21 and 23 percent more than the average, respectively.

Trade creation or diversion?

In Table 3, we investigate the possibility that the stimulus to trade among members of a currency union comes at the expense of trade with nonmembers. To do so, we add the trade diversion dummy to the specifications of Tables 1 and 2. All the OLS and fixed-effects specifications show that the coefficient associated with the trade diversion dummy is positive. This suggests that there is a significant trade creation (rather than trade diversion) effect, but this effect is smaller for Africa than for the world. Moreover, although there is a significant trade creation effect in both intra-Africa and Africa-ROW specifications, the intra-Africa trade creation effect is about five times larger, which reflects the observed trade patterns in Africa’s trade arrangements. The PML estimation for the world sample, however, gives a negative estimated coefficient for the trade diversion dummy, and insignificant coefficients for the Africa and Africa-ROW samples, suggesting that the earlier results may not be robust.22 The evidence from all estimation methods suggests that currency unions are associated with trade creation within Africa, which is significant and almost identical in magnitude for the PML and fixed-effects estimates.

Table 3.Trade Creation or Diversion?
Sample:WorldAfricaAfrica-

ROW
Intra-

Africa
WorldAfricaAfrica-

ROW
Intra-

Africa
WorldAfricaAfrica-

ROW
Intra-

Africa
Estimation:OLSOLSOLSOLSFixed

effects
Fixed

effects
Fixed

effects
Fixed

effects
PMLPMLPMLPML
Specification:(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)
Currency union0.97***

(8.77)
1.05***

(6.81)
0.61

(1.38)
0.59***

(2.58)
0.77***

(20.42)
0.75***

(10.77)
0.74***

(10.00)
0.91***

(9.15)
0.12**

(2.16)
0.72***

(4.10)
0.64***

(3.39)
0.91***

(3.92)
Trade diversion0.23***

(8.66)
0.10**

(2.35)
0.09

(1.92)
0.09

(0.55)
0.23***

(20.81)
0.22***

(11.66)
0.18***

(9.22)
0.52***

(8.52)
−0.20***

(3.63)
0.10

(0.80)
0.03

(0.46)
0.51***

(3.53)
Log distance−1.05***

(52.02)
−1.05***

(19.62)
−0.96***

(15.61)
−1.29***

(9.86)
Log product real GDP0.93***

(105.28)
1.01***

(61.82)
1.06***

(64.45)
0.63***

(9.89)
0.38***

(24.23)
0.13***

(3.49)
0.11***

(2.65)
0.80***

(4.20)
1.01***

(16.59)
1.14***

(12.68)
1.17***

(12.38)
0.64***

(4.90)
Log product real GDP/capita0.44***

(33.56)
0.33***

(14.46)
0.32***

(13.64)
0.50***

(5.61)
0.41***

(27.56)
0.45***

(11.93)
0.51***

(11.76)
−0.40**

(2.15)
−0.01

(0.17)
−0.18***

(2.58)
−0.13

(1.56)
−1.40

(1.24)
Common language0.33***

(8.80)
0.28***

(4.57)
0.27***

(4.20)
0.40**

(2.00)
Common land border0.42***

(4.04)
1.22***

(7.29)
1.11***

(3.19)
0.98***

(4.26)
Free trade agreement (with Africa)1.17***

(14.03)
0.93***

(7.55)
0.80***

(5.13)
0.34***

(13.68)
0.18***

(3.80)
0.40***

(6.98)
0.47***

(3.39)
0.62***

(4.07)
0.21

(1.49)
Number landlocked in the pair−0.24***

(8.38)
−0.35***

(8.71)
−0.36***

(8.70)
−0.33***

(3.03)
Number islands in the pair0.01

(0.24)
−0.31***

(5.01)
−0.39***

(6.07)
0.17

(0.86)
Log product of areas−0.08***

(11.48)
−0.18***

(14.82)
−0.20***

(16.19)
−0.01

(0.20)
Common colonizer0.52***

(8.42)
0.31***

(3.60)
0.33***

(3.51)
0.26

(1.15)
Current colony1.03***

(4.17)
−0.35

(0.77)
−0.02

(0.03)
0.38***

(8.72)
−0.08

(0.95)
−0.06

(0.69)
0.86**

(2.32)
0.10

(0.54)
0.10

(0.58)
Ever colony1.25***

(10.57)
2.10***

(13.94)
2.09***

(13.65)
Same nation−0.25

(0.24)
1.60***

(3.22)
1.46***

(2.59)
Observations265,262100,59785,75914,838265,262100,59785,75914,838380,512176,712146,46830,244
R2 (within)0.140.030.040.03
R2 (between)0.590.270.260.17
R2 (overall)0.680.520.540.400.530.220.200.12
Hausman test (p-value)0.000.000.000.000.000.000.000.00
Wald χ20.000.000.000.000.000.000.000.00
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the trade diversion effect of currency unions using the world, Africa (at least one country in the pair is in Africa), intra-Africa (both countries in the pair are in Africa), and Africa-rest of the world (ROW) (only one country in the pair is in Africa) samples. The dependent variable is log of real trade between trading partners (log(Xij)) in the OLS and fixed effects regressions, and real trade between trading partners (Xij) in the PML regressions. Real trade is defined as the average of exports and imports between the trading partners deflated by the U.S. consumer price index. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); trade diversion (dummy variable equal to 1 if the trading partners are not in the same currency union but at least one is in a currency union with another country); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). Year dummies included in all regressions. The table reports robust t-statistics in parentheses of columns 1–8, and robust z-statistics in columns 9–12. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the trade diversion effect of currency unions using the world, Africa (at least one country in the pair is in Africa), intra-Africa (both countries in the pair are in Africa), and Africa-rest of the world (ROW) (only one country in the pair is in Africa) samples. The dependent variable is log of real trade between trading partners (log(Xij)) in the OLS and fixed effects regressions, and real trade between trading partners (Xij) in the PML regressions. Real trade is defined as the average of exports and imports between the trading partners deflated by the U.S. consumer price index. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); trade diversion (dummy variable equal to 1 if the trading partners are not in the same currency union but at least one is in a currency union with another country); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). Year dummies included in all regressions. The table reports robust t-statistics in parentheses of columns 1–8, and robust z-statistics in columns 9–12. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.

Sensitivity analysis

We check the robustness of our results presented in Tables 1 and 2, by changing our methodology in a number of ways and reporting estimates of the coefficients of interest. In particular, we conduct sensitivity checks of the specification by estimating the dependent variable as the average of the logarithm of exports and imports (rather than the logarithm of the average); adding quadratic terms for output and output per capita to control for possible sample nonlinearities; modifying the dependent variable by adding a constant to trade observations before taking the logarithm to avoid truncation of the data;23 and using Tobit to account for the censored nature of the dependent variable.

The results in Table 4 show the estimates of the currency union trade-generating effect γ for the various robustness checks. First, for the benchmark panel fixed effects estimates, adding nonlinear terms and time effects, and changing the specification of the dependent variable to the average of the logs has a marginal effect on the size of γ. The trade-generating effect is about 1.8 for the world and 1.7 for Africa (which is smaller than that estimated using pooled OLS). Next, including zero-trade observations makes a difference to the results. This is because excluding zero values tends to drive the overall elasticity upward, whereas their inclusion drives it downward.24 The Tobit estimates that overcome the bias that may result from the censored nature of the data are sufficiently close to the cross-section results obtained in Table 1. However, they may suffer from the omitted-variable bias resulting from not controlling for country-pair specific fixed effects. The PML estimates that take into account all zero-trade observations as well as country-pair fixed effects find a positive and significant but smaller impact of currency unions on trade.

Table 4.Sensitivity Analysis of Currency Union Effect(Various Estimation Methods)
Sample:WorldAfrica
γEffectγEffect
Ordinary least squares
With time effects (from Table 1)0.84***

(7.75)
2.30.99***

(6.57)
2.7
Without time effects0.91***

(7.69)
2.50.97***

(5.70)
2.6
With time and country fixed effects (Anderson van-Wincoop)0.85**

(7.86)
2.31.14

(7.82)
3.1
Dependent variable log(100 + Xij) (with time effects)0.57***

(4.85)
1.80.58***

(4.05)
1.8
Tobit estimation (with time effects)0.79***

(20.30)
2.20.85***

(16.09)
2.4
Fixed effects
With time effects (from Table 1)0.60***

(16.15)
1.80.54***

(7.91)
1.7
Without time effects0.58***

(15.54)
1.80.55***

(7.94)
1.7
Dependent variable calculated as average of logs (with time effects)0.65***

(17.17)
1.90.60***

(8.86)
1.8
Nonlinearities added (with time effects)0.45***

(11.83)
1.60.50***

(7.23)
1.7
PML panel estimates with fixed effects including zero trade observations
With time effects (from Table 1)0.15**

(2.51)
0.20.65***

(5.05)
0.7
Without time effects0.09

(1.34)
0.10.65***

(7.20)
0.7
Source: Authors’ calculations.Note: This table presents the results of the currency union trade-generating effect for various robustness checks using the world and Africa samples. The dependent variable is log of real trade (log(Xij)) in the ordinary least squares (OLS) and fixed effects regressions, and real trade (Xij) in the Poisson pseudomaximum likelihood (PML) regressions. The table reports robust t-statistics in parentheses of OLS and fixed effects estimation results and robust z-statistics in parentheses of PML results.*, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.
Source: Authors’ calculations.Note: This table presents the results of the currency union trade-generating effect for various robustness checks using the world and Africa samples. The dependent variable is log of real trade (log(Xij)) in the ordinary least squares (OLS) and fixed effects regressions, and real trade (Xij) in the Poisson pseudomaximum likelihood (PML) regressions. The table reports robust t-statistics in parentheses of OLS and fixed effects estimation results and robust z-statistics in parentheses of PML results.*, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.

Despite the differences in the magnitudes among estimation methods, the results always show a statistically significant trade-generating effect for the Africa sample, which is comparable to the world sample trade-generating effect. Overall, the currency union trade-generating effect is strong for Africa, and may be larger than that for the world. However, the magnitude of the currency union trade-generating effect seems to be sensitive to the inclusion of zero-trade observations as well as the estimation methodology employed, particularly when zero observations are included and the PML approach is used. Taken together, all panel estimates (fixed effects “within” and PML) suggest that the average trade-generating effect is 1.2 for the world and 1.4 for Africa.

Years of Membership, Comovements, and Trade Stability

We now turn our attention to three different but related issues. In particular, we investigate the impact of the duration of currency union membership on bilateral trade, the currency union impact on output and price comovements, and the effect of currency union membership on trade stability.

Membership duration

To investigate the “time effect” of currency union membership, we examine how the effects of leaving a currency union evolve over time. To do so, we follow Glick and Rose (2002) and define a dummy variable that is equal to unity for the observations associated with the year a union is dissolved between a given pair of countries, and equal to zero otherwise.25 Importantly, we also include lags of this dummy variable in the equation, that is, we add a dummy variable for one year after dissolution, for two years after dissolution, and so on.

Figures 1 and 2 plot the coefficients of subsequent lags to trace out the response of bilateral trade (the “response coefficients”) to the dissolution of a currency union. This clearly illustrates the impact of currency union dissolution on trade over time.26 As the time-dimension of our data set is longer than that used by Glick and Rose, we show more convincingly that exit from a currency union is associated with a decline in bilateral trade.

Figure 1.Estimated Impact of Dissolution of Currency Union (CU) on Trade: World Sample

Notes: The horizontal lines in this figure correspond to the estimate of the coefficient of the currency union dummy (the γ coefficient for the gravity equation (1) extended to include the vector of lagged variables described in the text) for the world and Africa samples, respectively. The corresponding lag of the dummy variable associated with CU dissolution is statistically significant if the corresponding error bands exclude zero.

Figure 2.Estimated Impact of Dissolution of Currency Union (CU) on Trade: Africa Sample

Notes: The horizontal lines in this figure correspond to the estimate of the coefficient of the currency union dummy (the γ coefficient for the gravity equation (1) extended to include the vector of lagged variables described in the text) for the world and Africa samples, respectively. The corresponding lag of the dummy variable associated with CU dissolution is statistically significant if the corresponding error bands exclude zero.

For both samples, trade is mostly lower after the dissolution of a currency union compared with trade during the currency union. This is particularly true for Africa, where—with the exception of a small “blip” that occurs in the 18th year—the “response coefficients” are estimated to be substantially lower than before the dissolution. For the world sample, the “response coefficients” are around the level of trade during the currency union until about 27 years after the exit. Further, the adverse effect of exiting a currency union is smaller for Africa in the first few years but increases later. After 10 years of currency union exit, bilateral trade declines cumulatively by about 30 percent for Africa and 45 percent for the world, but 25 years after the exit, the cumulative decline in bilateral trade is greater in Africa than in the world sample.

Statistically significant effects of currency union dissolution are visible (in both samples) after about 13 years. However, after about 22 and 28 years for Africa and the world samples, respectively, these effects become gradually negligible (and insignificantly different from zero), suggesting that the effect of the currency union dissolution disappears. Overall, the results suggest that leaving a currency union has a significant impact on trade (perhaps more for Africa than the world), but this relationship is far from linear. This observation motivates the inclusion of a quadratic term in the currency union duration effect analysis.

In addition to the dissolution of currency union membership, we examine whether the duration of sharing a common currency matters for trade. We use a novel approach that introduces (sequentially) to equation (1) a variable that measures the number of years that a given trading partner has shared a common currency, and the variable’s quadratic term to capture potential nonlinear effects of currency union duration. Results presented in Table 5 show that the duration of currency union membership is important in all cases. According to the fixed-effects estimates, one additional year of membership increases trade by about 1.4 and 2.5 percent for the world and Africa, respectively. Holding all else constant, after 10 years of currency union membership, trade increases by about 15 percent for the world and 28 percent for Africa. This implies that currency union participation would double the level of trade in about 67 years for the world and in 36 years for Africa. Interestingly, the estimated coefficient for the quadratic term is negative and statistically significant (albeit small) in all specifications. This result points at diminishing returns of currency union membership.

Table 5.Duration of Currency Union Membership
Sample:WorldWorldAfricaAfricaWorldWorldAfricaAfricaWorldWorldAfricaAfrica
Estimation:OLSOLSOLSOLSFixed

effects
Fixed

effects
Fixed

effects
Fixed

effects
PMLPMLPMLPML
Specification:(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)
Years in currency union0.03***

(7.47)
(7.49)0.03***

(5.99)
0.06***

(5.58)
0.01***

(8.05)
0.04***

(13.36)
0.02***

(8.54)
0.04***

(7.84)
0.01*

(1.73)
0.05***

(3.80)
0.03***

(4.71)
0.06***

(4.94)
(Years in currency union)−0.001***

(4.86)
−0.001***

(3.26)
−0.001***

(10.89)
−0.001***

(4.81)
−0.002***

(3.42)
−0.001***

(2.94)
Log distance−1.05***

(52.02)
−1.05***

(52.00)
−1.06***

(19.88)
−1.06***

(19.82)
Log product real GDP(105.60)0.93***

(105.64)
(61.60)(61.70)0.38***

(24.39)
0.39***

(24.81)
0.12***

(3.04)
0.13***

(3.38)
1.01***

(16.40)
1.01***

(16.59)
1.16***

(12.54)
1.15***

(12.54)
Log product real GDP/capita0.44***

(33.88)
0.44***

(33.87)
0.34***

(14.77)
0.34***

(14.72)
0.42***

(27.60)
0.41***

(26.88)
0.48***

(12.57)
0.46***

(11.95)
0.01

(0.17)
−0.01

(0.08)
−0.16**

(2.11)
−0.18**

(2.35)
Common language0.35***

(9.26)
0.35***

(9.22)
0.29***

(4.74)
0.29***

(4.67)
Common land border0.41***

(3.94)
0.40***

(3.89)
1.23***

(7.29)
1.23***

(7.29)
Free trade agreement

(with Africa)
1.19***

(14.35)
1.19***

(14.41)
0.91***

(7.36)
0.92***

(7.42)
0.33***

(13.38)
0.35***

(13.94)
0.13***

(2.69)
0.14***

(2.85)
0.47***

(3.40)
0.47***

(3.40)
0.62***

(3.87)
0.62***

(3.94)
Number landlocked in the pair−0.25***

(8.82)
−0.25***

(8.82)
−0.36***

(8.90)
−0.36***

(8.88)
Number islands in the pair0.02

(0.67)
0.02

(0.68)
−0.31***

(5.00)
−0.31***



(5.02)
Log product of areas−0.08***

(11.30)
−0.08***

(11.38)
−0.18***

(14.67)
−0.18***

(14.74)
Common colonizer0.51***

(8.24)
0.50***

(8.06)
0.31***

(3.53)
0.30***

(3.41)
Current colony1.17***

(4.80)
1.06***

(4.49)
−0.04

(0.09)
−0.21

(0.47)
0.51***

(12.19)
0.41***

(9.58)
0.09

(1.14)
−0.03

(0.36)
0.97**

(2.39)
0.86**

(2.31)
0.29

(1.29)
0.22

(1.03)
Ever colony1.34***

(11.10)
1.33***

(11.04)
2.15***

(14.69)
2.14***

(14.60)
Same nation−0.24

(0.21)
−0.22

(0.19)
1.62***

(3.23)
1.73***

(3.60)
Observations265,262265,262100,597100,597265,262265,262100,597100,597380,512380,512176,712176,712
R2 (within)0.140.140.030.03
R2 (between)0.590.590.250.28
R2 (overall)0.680.680.520.520.530.540.210.22
Hausman test (p-value)0.000.000.000.000.000.000.000.00
Wald χ2 (p-value)0.000.000.000.00
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the trade creation effect of currency union membership duration using the world and Africa (at least one country in the pair is in Africa) samples. The dependent variable is log of real trade between trading partners (log (Xij)) in the OLS and fixed effects regressions, and real trade between trading partners (Xij) in the PML regressions. Real trade is defined as the average of exports and imports between the trading partners deflated by the U.S. consumer price index. The independent variables are years in currency union (number of years since joining the currency union) and its square; log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). Year dummies included in all regressions. The table reports robust t-statistics in parentheses of columns 1–8, and robust z-statistics in columns 9–12. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the trade creation effect of currency union membership duration using the world and Africa (at least one country in the pair is in Africa) samples. The dependent variable is log of real trade between trading partners (log (Xij)) in the OLS and fixed effects regressions, and real trade between trading partners (Xij) in the PML regressions. Real trade is defined as the average of exports and imports between the trading partners deflated by the U.S. consumer price index. The independent variables are years in currency union (number of years since joining the currency union) and its square; log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). Year dummies included in all regressions. The table reports robust t-statistics in parentheses of columns 1–8, and robust z-statistics in columns 9–12. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.

Comovements of prices and output

In order to investigate the effects of currency unions on the comovements of prices and output we follow the approach of Alesina and Tenreyro (2002), and Tenreyro and Barro (2003). Specifically, we test whether their results hold in a bilateral approach to currency unions, especially in the context of Africa. Hence, we estimate equation (1), sequentially substituting the dependent variable with VPij and VYij as calculated in Equations (4) and (6).27 The results of the estimated impact on comovements of output and prices are presented in Table 6. We find that currency unions tend to increase the comovement of prices in both the world and Africa, but that they are not systematically related to the comovement of outputs. These findings support the results of Alesina and Tenreyro (2002). However, we also find that the marginal effect of currency unions on the price comovements is higher in Africa. This is an important observation because price comovements tend to be lower in Africa (see summary statistics in Tables A2A4). The insignificant effect of currency unions on the comovement of outputs is not surprising considering that the theoretical link between the two is ambiguous, and largely depends on the extent to which trade is intra- or interindustry.28 The positive estimated effect of currency union on price comovement is relatively clear because countries that are members of currency unions avoid inflation and nominal exchange rate volatility that characterizes other regimes.

Table 6.Currency Union Impact on Comovements of Outputs and Prices(Dependent Variables: Comovement of Outputs (Vyij) for (1)–(4); and Comovement of Prices (Vpij) for (5)–(8))
Sample:WorldAfricaWorldAfricaWorldAfricaWorldAfrica
Specification:(1)(2)(3)(4)(5)(6)(7)(8)
Currency union−0.002

(0.61)
0.002

(0.55)
0.000

(0.12)
0.000

(0.25)
0.06***

(9.07)
0.07***

(9.35)
0.03***

(11.19)
0.04***

(11.19)
Log distance0.002***

(3.56)
0.008***

(7.58)
−0.001***

(3.56)
−0.001

(0.96)
−0.003*

(1.79)
0.001

(0.26)
−0.004***

(5.33)
−0.003**

(2.23)
Log product real GDP0.005***

(22.10)
0.005***

(14.19)
0.003***

(5.21)
0.000

(0.30)
0.002***

(4.18)
−0.003***

(3.96)
0.007***

(3.34)
−0.008***

(3.57)
Log product real GDP/capita0.005***

(16.98)
0.001

(1.63)
−0.004***

(4.39)
0.001

(0.56)
0.012***

(14.76)
0.016***

(16.61)
−0.004

(1.17)
0.011***

(3.47)
Common language0.001

(0.65)
−0.002

(1.54)
0.001**

(2.05)
0.000

(0.13)
0.001

(0.42)
0.000

(0.09)
0.002

(1.43)
0.001

(1.15)
Common land border0.006**

(2.52)
0.005

(1.44)
0.003***

(3.65)
0.001

(0.82)
−0.005

(0.66)
−0.007

(0.77)
0.004

(1.46)
0.004

(1.24)
Free trade agreement

(with Africa)
0.012***

(4.59)
0.010***

(2.65)
0.003**

(2.35)
−0.001

(0.60)
0.018***

(2.84)
0.015*

(1.89)
0.007**

(2.50)
−0.007**

(2.21)
Number landlocked in the pair0.005***

(8.19)
0.007***

(8.65)
0.007***

(3.88)
0.014***

(7.30)
Number islands in the pair0.005***

(7.14)
0.010***

(8.08)
0.012***

(6.77)
0.012***

(5.45)
Log product of areas0.000***

(3.07)
0.000

(0.66)
−0.004***

(9.62)
−0.001*

(1.94)
Common colonizer0.004***

(3.14)
0.009***

(5.45)
0.002***

(3.49)
0.001

(0.98)
0.010***

(3.35)
0.002

(0.50)
0.005***

(4.06)
0.006***

(4.48)
Current colony−0.018*

(1.70)
−0.044

(1.47)
−0.007**

(2.03)
−0.002

(0.23)
0.049*

(1.90)
−0.040

(0.68)
0.025**

(2.15)
−0.020

(0.70)
Ever colony0.004*

(1.70)
0.018**

(2.01)
0.000

(0.03)
0.001

(0.55)
0.007

(0.72)
0.034*

(1.77)
−0.006

(1.40)
0.005

(0.70)
Same nation
Country fixed effectsNoNoYesYesNoNoYesYes
Observations6,9923,7936,9923,7937,0003,7967,0003,796
R20.250.150.920.920.120.130.870.91
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS) and fixed effects estimation results for the effects of currency unions on the comovements of prices and output using the world and Africa (at least one country in the pair is in Africa) samples. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). The table reports robust t-statistics in parentheses. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS) and fixed effects estimation results for the effects of currency unions on the comovements of prices and output using the world and Africa (at least one country in the pair is in Africa) samples. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). The table reports robust t-statistics in parentheses. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.
Table 7.Currency Union Impact on Trade Stability

(Dependent Variable is Trade Variability, the Coefficient of Variation of log(Xij))

Sample:WorldAfricaWorldAfricaWorldAfrica
Estimation:OLSOLSFixed effectsFixed effectsPMLPML
Specification:(1)(2)(3)(4)(5)(6)
Currency union−0.04**

(3.41)
−0.06***

(4.15)
−0.02

(0.84)
−0.01

(0.34)
0.08

(0.85)
−0.01

(0.07)
Log distance0.05***

(22.17)
0.07***

(11.88)
Log product real GDP−0.03***

(30.50)
−0.05***

(19.75)
−0.02

(1.29)
0.02*

(1.71)
−0.12***

(7.94)
−0.08***

(3.21)
Log product real GDP/capita−0.02***

(9.93)
−0.02***

(3.94)
−0.04***

(3.61)
−0.07***

(4.74)
−0.09***

(8.11)
−0.12***

(7.80)
Common language−0.03***

(6.25)
−0.03***

(4.25)
Common land border0.02**

(2.28)
−0.03**

(2.50)
Free trade agreement (with Africa)−0.06**

(5.95)
−0.08***

(5.50)
−0.02

(1.34)
−0.03*

(1.80)
−0.05

(0.99)
−0.03

(0.56)
Number landlocked in the pair0.01***

(3.19)
0.02***

(3.18)
Number islands in the pair0.02***

(3.56)
0.06***

(6.13)
Log product of areas0.00***

(2.71)
0.01***

(5.49)
Common colonizer−0.02***

(2.67)
0.00

(0.11)
Current colony−0.01

(0.66)
0.12***

(3.59)
0.04***

(2.71)
0.06***

(2.69)
0.70***

(4.21)
0.30*

(1.72)
Ever colony−0.02**

(2.19)
−0.04***

(2.75)
Same nation−0.02

(0.73)
−0.13***

(3.88)
Observations18,8198,06918,8198,06913,2126,176
R2 (within)0.060.11
R2 (between)0.110.01
R2 (overall)0.170.160.110.01
Hausman test (p-value)0.010.000.000.00
Wald χ2 (p-value)0.000.00
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the effect of currency unions on trade stability using the world and Africa samples. The dependent variable is the coefficient of variation of log of real trade in columns 1–4 and of real trade in columns 5 and 6. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). The table reports robust t-statistics in parentheses of columns 1–4, and robust z-statistics in columns 5 and 6. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.
Source: Authors’ calculations.Note: This table reports pooled ordinary least squares (OLS), fixed effects, and Poisson pseudomaximum likelihood (PML) estimation results for the effect of currency unions on trade stability using the world and Africa samples. The dependent variable is the coefficient of variation of log of real trade in columns 1–4 and of real trade in columns 5 and 6. The independent variables are currency union (dummy variable equal to 1 if money is interchangeable between the two countries at a 1:1 par for an extended period of time, and 0 otherwise); log of geographical distance between the trading partners; log of the product of real GDP of the trading partners; log of the product of real GDP per capita of the trading partners; common language (dummy variable equal to 1 if the two countries share a language, and 0 otherwise); common land border (dummy variable equal to 1 if the two countries share a border, and 0 otherwise); free trade agreement with Africa (a more comprehensive version of the variable free trade agreement that also takes into account free trade agreements with African countries); number of landlocked countries in the pair (for example, 0, 1, or 2); number of islands in the pair (for example, 0, 1, or 2); log of product of land areas of the two countries; common colonizer (dummy variable equal to 1 if the trading partners share the colonizer, and 0 otherwise); current colony (dummy variable equal to 1 if one country in the pair is colonized by the other country, and 0 otherwise); ever colony (dummy variable equal to 1 if one trading partner has ever been a colony of the other, and 0 otherwise); and same nation (dummy variable equal to 1 if both partners are part of the same nation, and 0 otherwise). The table reports robust t-statistics in parentheses of columns 1–4, and robust z-statistics in columns 5 and 6. *, **, *** denote significance at the 10, 5, and 1 percent levels, respectively.

Last, we examine whether currency unions make trade more stable by reducing exchange rate volatility. Table 7 shows estimations of the gravity model with the coefficient of variation of log of real trade as dependent variable in columns 1–4, and of real trade in columns 5 and 6. The OLS results show that currency unions increase trade stability, with the marginal impact being slightly higher for Africa than for the world. However, this effect becomes statistically insignificant when country-pair specific effects are introduced in the model (columns 3–6). This indicates that the evidence of currency unions on trade stability is tentative and not robust.

Sensitivity analysis

Similar to the previous section, we check the robustness of the results presented in Tables 57, by conducting sensitivity checks of the specification. In particular, we estimate the dependent variable as the average of the logarithm of exports and imports, add quadratic terms for output and output per capita to control for possible sample nonlinearities, and add time effects. Estimates in Table 5 are robust to all the above sensitivity checks: after a country in the world (Africa) has been in a currency union for 10 years, trade increases by 14–22 percent (20–29 percent), and the diminishing returns of currency union membership are always significant. The conclusions of Tables 6 and 7 are also robust to changes in the specification.29

IV. Conclusions

This paper has provided some insights into several aspects of the performance of currency unions using an augmented version of the gravity model and focusing on two samples, the world and Africa. Our analysis confirms earlier results that the impact of currency unions is positive and significant on trade. We show that the trade-generating effect of currency unions is strong for Africa—and may be larger than that for the world—suggesting that Africa stands to benefit at least as much from currency union participation as the ROW. However, the magnitude of the currency union trade-generating effect seems to be sensitive to the inclusion of zero-trade observations as well as the estimation methodology employed. On average, currency unions increase trade by factors of 1.2 and 1.4 in the world and Africa, respectively. Also, our results show that the duration of currency union membership matters: the longer the duration, the greater the benefits in terms-of-trade creation, albeit with some diminishing returns. These two effects combined suggest that although currency unions enhance trade by a factor of 1.2–1.4, the effect is not immediate: according to estimates, currency union participation doubles the level of trade in about 67 years for the world and 36 years for Africa. In addition, we find evidence that currency union participation increases price comovements among member countries, but has no significant effect on output comovements among members. Finally, the currency union effect on trade stability appears to be ambiguous.

Although our results indicate that several aspects of currency unions operate more or less the same in Africa as elsewhere, we would like to emphasize that the marginal effects and mechanisms of transmission may vary across the two samples. The methodology herein does not constitute an explicit investigation into how trade and its underlying determinants are connected, or the extent to which currency unions can promote growth and reduce poverty. Identifying the similarities and some of the differences across the samples is only the first step in investigating the dynamics of currency unions in Africa and may raise more questions than it answers.

Future research on the issue may explore the omitted factors that induce countries to join a currency union and to trade more, as suggested by Rose and van Wincoop (2001). Joining a currency union and opening up to the world are very often political decisions, so that such omitted factors may lie in a set of political and institutional variables. This is not a new realization: in recent papers on “deeper” determinants of economic growth, both institutions and trade openness determine a country’s performance in the long run (see, for example, Acemoglu, Johnson, and Robinson, 2001; and Rodrik, Subramanian, and Trebbi, 2002). In the complicated web of relationships describing income, its determinants, and the linkages between the determinants, the interdependence of trade, and institutions is a recurring theme that is difficult to handle, not least because of the issues of causality and construction of proper instruments.

Appendix I

See Tables A1A4 and Figure A1.

Table A1.Sample Data: Variable Definitions and Sources
VariableDescriptionSource
Dependent variable
XijtThe average value of real bilateral trade between i and j at time tIMF’s Direction of Trade Statistics (DoT); Average of available values for export from a to b, export from b to a, import into a from b, import into b from a. Deflated by U.S. consumer price index
Explanatory variables
YiReal GDPWorld Bank’s World Development Indicators
PopiPopulationWorld Bank’s World Development Indicators, Penn World Tables 6.1;Glick and Rose (2002) data used for 1948–50
DijThe distance between i and jCIA’s World Factbook; Great Circle Method used to calculate distance
LangijA binary variable that is unity if i and j have a common languageCIA’s World Factbook
ContijA binary variable that is unity if i and j share a land borderCIA’s World Factbook
LandlThe number of landlocked countries in the country pair (0, 1, or 2)CIA’s World Factbook
IslandThe number of island nations in the country pair (0, 1, or 2)CIA’s World Factbook
AreaLand area of country n, n = i,jCIA’s World Factbook
ComColyijBinary variable that is unity if i and j were colonies after 1945 with the same colonizerCIA’s World Factbook
CurColijBinary variable that is unity if i and j are colonies at time tCIA’s World Factbook
ColonyijBinary variable that is unity if i colonized j or vice versaCIA’s World Factbook
ComNatijBinary variable that is unity if i and j remained part of the same nation during the sampleCIA’s World Factbook
FTAijBinary variable that is unity if i and j belong to the same regional trade agreementWTO publication [www.wto.org/english/tratop_e/region_e/region_e.htm]
CUijBinary variable that is unity if i and j use the same currency at time tGlick and Rose (2002). For 1998–2002: IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions
PiPPP of GDPPenn World Tables 6.1; price level of GDP (P) is the purchasing power parity over GDP divided by the exchange rate and multiplied by 100
TotiTerms of tradeIMF’s World Economic Outlook 2004
Table A2.Summary Statistics
World Sample

(265,262 observations)
Africa Sample

(100,597 observations)
VariableMeanSDMeanSD
Log of real trade10.0203.2468.8132.951
Currency union0.0140.1190.0270.161
Log distance8.1670.8098.1750.661
Log product real GDP47.9742.66247.0752.193
Log product real GDP/capita16.1061.47015.1801.259
Common language0.2160.4110.2660.442
Common land border0.0300.1700.0290.169
Free trade agreement(with Africa)0.0340.1800.0450.207
Number landlocked in the pair0.2620.4790.4020.553
Number islands in the pair0.3380.5370.2500.470
Log product of areas24.1393.29924.6702.904
Common colonizer0.0990.2980.1550.362
Current colony0.0020.0410.0020.043
Ever colony0.0200.1390.0170.129
Same nation0.0000.0160.0000.017
Years in currency union0.3593.4810.7054.840
Comovements of outputs1−0.0750.030−0.0870.031
Comovements of prices1−0.1550.080−0.1630.070
Trade volatility20.1810.2580.2240.294

One value for country pair (6,992/7,000 observations for the world for comovements of outputs/prices and 3,793/3,796 for Africa samples, respectively).

Two values for country pair (18,156 observations for the world and 7,845 for the Africa sample).

One value for country pair (6,992/7,000 observations for the world for comovements of outputs/prices and 3,793/3,796 for Africa samples, respectively).

Two values for country pair (18,156 observations for the world and 7,845 for the Africa sample).

Table A3.Free Trade Agreements in the Sample
Regional AgreementsMembers (date of entry)
Africa:
ECOWAS: Economic Community of West African StatesBenin (1975), Ghana (1975), Niger (1975), Burkina Faso (1975), Guinea (1975), Nigeria (1975), Cape Verde (1975), Liberia (1975), Senegal (1975), Côte d’Ivoire (1975), Mali (1975), Sierra Leone (1975), Gambia, The (1975), Mauritania (1975), Togo (1975)
COMESA: Common Market for Eastern and Southern AfricaAngola (1981), Malawi (1981), Tanzania (1981), Burundi (1981), Mauritius (1981), Uganda (1981), Comoros (1981), Mozambique (1981), Zambia (1981), Djibouti (1981), Rwanda (1981), Eritrea (1993), Namibia (1993), Zimbabwe (1981), Ethiopia (1981), Somalia (1981), Madagascar (1993), Kenya (1981), Sudan (1981), Seychelles (1993), Lesotho (1981), Swaziland (1981)
SADC: Southern African Development CommunityAngola (1980), Lesotho (1980), Zimbabwe (1980), Botswana (1980), Swaziland (1980), Namibia (1990), Malawi (1980), Tanzania (1980), Mauritius (1995), Mozambique (1980), Zambia (1980), South Africa (1995)
CEMAC: Central African Economic and Monetary CommunityCongo, Republic of (1962), Central African Republic (1962), Gabon (1962), Chad (1962–68, 1984), Cameroon (1962), Equatorial Guinea (1983)
WAEMU: West African Economic and Monetary UnionBenin (1984), Burkina Faso (1962), Côte d’Ivoire (1962), Mali (1962), Niger (1962), Senegal (1962), Mauritania (1962–94), Togo (1994)
Rest of the world:
ASEANBrunei (1984), Cambodia (1999), Indonesia (1967), Lao PDR (1997), Malaysia (1967), Myanmar (1997), Philippines (1967), Singapore (1967), Thailand (1967), Vietnam (1995)
EUAustria (1995), Belgium (1958), Denmark (1973), Finland (1995), France (1958), Germany (1958), Greece (1981), Luxembourg (1958), Ireland (1973), Italy (1958), Netherlands (1958), Portugal (1986), Spain (1986), Sweden (1995), United Kingdom (1973)
United States-IsraelUnited States (1985), Israel (1985)
NAFTACanada (1989), United States (1989), Mexico (1994)
CARICOMAntigua and Barbuda (1974), Bahamas (1983), Barbados (1973), Belize (1974), Dominica (1974), Guyana (1973), Grenada (1974), Jamaica (1973), Montserrat (1974), Saint Kitts and Nevis (1974), Saint Lucia (1974), Saint Vincent and the Grenadines (1974), Suriname (1995), Trinidad and Tobago (1973), Haiti (2002)
PATCRAAustralia (1977), Papua New Guinea (1977)
ANZCERTAAustralia (1983), New Zealand (1983)
CACMCosta Rica (1963), El Salvador (1961), Guatemala (1961), Honduras (1961), Nicaragua (1961)
SPARTECAAustralia (1981), New Zealand (1981), Cook Islands (1981), Micronesia, Federated States of (1986), Fiji (1981), Kiribati (1981), Marshall Islands (1981), Nauru (1981), Niue (1981), Papua New Guinea (1981), Solomon Islands (1981), Tonga (1981), Tuvalu (1981), Vanuatu (1981), Samoa (1981)
MERCOSURArgentina (1991), Brazil (1991), Paraguay (1991), Uruguay (1991), Chile (1996), Bolivia (1996)
Table A4.Currency Unions in the Sample
Currency Union MembersEndCurrency Union MembersEnd
Antigua and BarbudaBarbados1975Central African RepublicCongo, Republic ofOngoing
Antigua and BarbudaDominicaOngoingCentral African RepublicCôte d’IvoireOngoing
Antigua and BarbudaGrenadaOngoingCentral African RepublicEquatorial GuineaOngoing
Antigua and BarbudaGuyana1971Central African RepublicGabonOngoing
Antigua and BarbudaMontserratOngoingCentral African RepublicGuinea1969
Antigua and BarbudaSaint Kitts and NevisOngoingCentral African RepublicGuinea-BissauOngoing
Antigua and BarbudaSaint LuciaOngoingCentral African RepublicMadagascar1982
Antigua and BarbudaSaint Vincent and the GrenadinesOngoingCentral African RepublicMaliOngoing
Antigua and BarbudaTrinidad and Tobago1976Central African RepublicMauritania1974
ArubaNetherlands AntillesOngoingCentral African RepublicNigerOngoing
ArubaSuriname1994Central African RepublicReunion1976
AustraliaKiribatiOngoingCentral African RepublicSenegalOngoing
AustraliaNauruOngoingCentral African RepublicTogoOngoing
AustraliaSolomon Islands1979ChadBeninOngoing
AustraliaTonga1991ChadBurkina FasoOngoing
AustraliaTuvaluOngoingChadComoros1994
BangladeshIndia1974ChadCongo, Republic ofOngoing
BarbadosDominica1975ChadCôte d’IvoireOngoing
BarbadosGrenada1975ChadEquatorial GuineaOngoing
BarbadosGuyana1971ChadGabonOngoing
BarbadosMontserrat1975ChadGuinea1969
BarbadosSaint Kitts and Nevis1975ChadGuinea-BissauOngoing
BarbadosSaint Lucia1975ChadMadagascar1982
BarbadosSaint Vincent and the Grenadines1975ChadMaliOngoing
BarbadosTrinidad and Tobago1975ChadMauritania1974
BelgiumBurundi1964ChadNigerOngoing
BelgiumCongo, Democratic Republic of1961ChadReunion1976
BelgiumRwanda1966ChadSenegalOngoing
Belgium-LuxembourgBurundi1964ChadTogoOngoing
Belgium-LuxembourgCongo, Democratic Republic of1961ComorosBenin1994
Belgium-LuxembourgRwanda1966ComorosBurkina Faso1994
BeninBurkina FasoOngoingComorosCongo, Republic of1994
BeninCôte d’IvoireOngoingComorosCôte d’Ivoire1994
BeninEquatorial GuineaOngoingComorosEquatorial Guinea1994
BeninGabonOngoingComorosGabon1994
BeninGuinea1969ComorosGuinea1969
BeninGuinea-BissauOngoingComorosMadagascar1982
BeninMadagascar1982ComorosMali1994
BeninMaliOngoingComorosMauritania1974
BeninMauritania1974ComorosNiger1994
BeninNigerOngoingComorosReunion1976
BeninReunion1976ComorosSenegal1994
BeninSenegalOngoingComorosTogo1994
BeninTogoOngoingCongo, Republic ofBeninOngoing
BhutanIndiaOngoingCongo, Republic ofBurkina FasoOngoing
BhutanPakistan1966Congo, Republic ofCôte d’IvoireOngoing
BotswanaLesotho1977Congo, Republic ofEquatorial GuineaOngoing
BotswanaSwaziland1977Congo, Republic ofGabonOngoing
BruneiMalaysia1971Congo, Republic ofGuinea1969
BruneiSingaporeOngoingCongo, Republic ofGuinea-BissauOngoing
MyanmarIndia1966Congo, Republic ofMadagascar1982
MyanmarPakistan1971Congo, Republic ofMaliOngoing
CameroonBeninOngoingCongo, Republic ofMauritania1974
CameroonBurkina FasoOngoingCongo, Republic ofNigerOngoing
CameroonCentral African RepublicOngoingCongo, Republic ofReunion1976
CameroonChadOngoingCongo, Republic ofSenegalOngoing
CameroonComoros1994Congo, Republic ofTogoOngoing
CameroonCongo, Republic ofOngoingCôte d’IvoireBurkina FasoOngoing
CameroonCôte d’IvoireOngoingCôte d’IvoireMadagascar1982
CameroonEquatorial GuineaOngoingCôte d’IvoireMaliOngoing
CameroonGabonOngoingCôte d’IvoireMauritania1974
CameroonGuinea1969Côte d’IvoireNigerOngoing
CameroonGuinea-BissauOngoingCôte d’IvoireReunion1976
CameroonMadagascar1982Côte d’IvoireSenegalOngoing
CameroonMaliOngoingCôte d’IvoireTogoOngoing
CameroonMauritania1974DenmarkFaroe IslandsOngoing
CameroonNigerOngoingDenmarkGreenlandOngoing
CameroonReunion1976DjiboutiBenin1949
CameroonSenegalOngoingDjiboutiBurkina Faso1949
CameroonTogoOngoingDjiboutiCameroon1949
Central African RepublicBeninOngoingDjiboutiCentral African Republic1949
Central African RepublicBurkina FasoOngoingDjiboutiChad1949
Central African RepublicChadOngoingDjiboutiComoros1949
Central African RepublicComoros1994DjiboutiCongo, Republic of1949

Figure A1.Main African Regional and Subregional Economic Integration Arrangements (as of August 2005)

References

Charalambos Tsangarides is an economist in the IMF’s Research Department; Pierre Ewenczyk is a senior economist with the IMF’s Offices in Europe; Michal Hulej is an economist at the National Bank of Poland; and Mahvash Saeed Qureshi is an economist in the IMF’s African Department. The authors thank the editor, the anonymous referees, Xavier Debrun, Anne-Marie Gulde, Catherine Pattillo, Andrew Rose, Sylvana Tenreyro, and João Santos Silva for helpful comments and suggestions.

Initially, these models were criticized for lacking a proper theoretical justification. Anderson (1979) and Bergstrand (1985) were the first formal attempts to address this criticism and derived the gravity equation theoretically. Deardorff (1998) provides a comprehensive overview of the gravity model and shows that a variety of theoretical models can be tied to it. Similarly, Feenstra, Markusen, and Rose (2001) argue that gravity models are consistent with several theoretical models of trade.

See Baldwin (2005) for a comprehensive review of the literature and relevant issues on currency unions and trade.

Figure A1 lists African regional economic integration arrangements. See Masson and Pattillo (2004) for a detailed discussion of currency unions in Africa.

However, he acknowledges that the effect may be smaller for modern industrial countries; most currency unions in Rose (2000) comprise small or poor countries or both.

Rose (2004) performs a “meta-analysis” of the currency union effect by combining estimates from 34 other studies and estimates a range of 30 to 90 percent for the currency union effect.

However, the currency union variable used by Masson and Pattillo (2004) uses the FTA definition from Glick and Rose (2002), which does not distinguish between FTA and currency union effects. We overcome this limitation by constructing separate variables for FTA and currency unions, and identify their impacts separately.

The definition of “currency union,” following Glick and Rose (2002), implies that money is interchangeable between the two countries at a 1:1 par for an extended period of time, so that there is no need to convert prices when trading between a pair of countries. Under this definition, hard fixes are not identified as currency unions. Further, the definition of currency union is transitive: if country pairs X, Y and X, Z are in a currency union, then Y and Z are in a currency union.

Because the sample period of our data set begins in 1948, we ignore years spent in a currency union before 1948.

For details on the “classic gravity model mistakes,” see Baldwin (2005) and Frankel (2005).

Anderson and van Wincoop (2003) use the national price indices (Pi and Pj) to account for “multilateral resistance” between countries i and j, which can be estimated using an iterative process. However, because the estimation process is complex, they propose an alternative method that is preferable for empirical work: namely, estimating implicit prices by fixed effects, that is, by including country-specific dummy variables.

However, the instrument applied by Alesina and Tenreyro (2002) is not designed for multilateral currency unions.

The CFA franc zone and the Common Monetary Area were formed, in large part, due to the political self-interest of the major powers (France in the former case, and South Africa in the latter).

The issue of nonlinearities is also discussed in Baldwin (2005). See Frankel (2005) for a justification of using the “pooled” export-import specification. The treatment of zero-trade observations in the estimation is discussed in detail in Santos Silva and Tenreyro (2006).

Greene (1981) shows that the size of the “truncation bias” when the variables are distributed normally is inversely proportional to the “proportion of nonlimit observations in the sample,” but this bias decreases when the fit of the model improves or the regressors have a skewed distribution.

Jensen’s inequality implies that even if the expected value of the error term obtained from equation (1) is 0, E[log Xij|Zij] is not essentially equivalent to exp(E[Xij|Zij]). The possible bias in the presence of heteroskedasticity can be mitigated with the use of heteroscedasticity-robust standard errors in the estimations.

Political units include overseas territories, parts of kingdoms, possessions, self-governing territories in free association with another country, unincorporated territories, and crown dependencies.

Henceforth, unless stated otherwise, our panel and PML estimates include country-pair fixed effects along the lines of Glick and Rose (2002), as well as time effects. In addition, country fixed effects were also used to account for the Anderson and van Wincoop (2003) “multilateral resistance” factor in pooled OLS estimations and are available from the authors on request. In the interest of clarity, we do not show results of all the pooled OLS country fixed effects estimates in Tables 13, but some are summarized in Table 4.

Masson and Pattillo (2004) also estimate the gravity equation for Africa. However, their results may not be directly comparable to ours because they do not take into account free trade areas operating in Africa.

Although we estimate both fixed effects “within” and random effects, we rely on the robust fixed effects within estimator as suggested by the Hausman test.

To explore this issue in detail, we “decompose” the trade-generating coefficient for the world sample into the coefficients of two homogeneous groups—the industrialized and nonindustrialized countries—using fixed effects within estimator and PML. The results suggest that the PML technique is more sensitive to sample homogeneity than the other estimator. Thus, for example, the estimated coefficient for the currency union variable obtained from PML (fixed effects) is 0.14 (0.37) and 1.07 (0.51) for the industrialized and nonindustrialized groups, respectively. The world weighted average of 0.15 under PML appears to be influenced by the industrialized group average, and hence is smaller in magnitude. It is worth pointing out here that some other studies that estimate the gravity model using PML also obtain smaller estimated coefficients vis-à-vis other techniques, such as Tobit (see, for example, Amurgo-Pacheco and Pierola, 2008).

However, when we decompose the world sample into homogeneous groups we find that for the nonindustrialized group the trade creation effect is significantly positive (0.22), but a trade diversion effect exists for the industrialized group. As discussed in footnote 21, it appears that the PML result for the world in this case is also influenced by the industrial group observations.

To compare our results with those obtained by Alesina and Tenreyro (2002) we set the constant equal to 100. The dependent variable therefore becomes log(100 + Xij).

See Amemiya (1984) for a detailed discussion on this issue.

See Tables A2A4 for the number of identified currency union dissolutions for the world and Africa samples.

The horizontal lines in Figures 1 and 2 correspond to the estimate of the coefficient of the currency union dummy (namely, the g coefficient in equation (1) extended to include the vector of lagged variables described in the text) for the world and Africa, respectively. The 95 percent confidence intervals are also included.

The equations are estimated using the average values of the period 1948–2003 and pooled OLS. Because we cannot use panel estimation here, we account for country fixed effects as in Anderson and Van Wincoop (2003).

As discussed in Alesina and Tenreyro (2002), increased interindustry trade may stimulate sectoral specialization and lead to less output comovement, but intra-industry trade is likely to increase their comovement.

For trade stability, we estimate the model with various measures of stability, including the maximum absolute value, mean absolute value, and standard deviation of the residual from a conventional gravity equation of exports in levels. Results are available on request.

Other Resources Citing This Publication