One money, one market: the effect of common currencies on trade
📄 Summarized from the full manuscript · Human-reviewed for faithfulness before publication
In brief
Does sharing a currency really boost trade? Using a gravity-model panel of 186 countries across five decades, Rose finds that countries using the same currency trade over three times as much as otherwise-similar countries with separate currencies -- an effect an order of magnitude larger than eliminating exchange-rate volatility, and one that survives extensive robustness checks. The paper is candid that it cannot explain why the effect is so large, nor convert it into a welfare number, but concludes the trade benefits of currency unions had been substantially underestimated.
What this paper finds — and why it matters
This 2000 Economic Policy paper by Andrew Rose asks a simple question – does sharing a currency boost trade, and by how much – using a large cross-country panel rather than the handful of time-series studies that had previously found little effect of exchange-rate variability on trade. Rose estimates an augmented gravity model of bilateral trade, adding income, population, distance, a common land border, a common language, a regional trade agreement, colonial history, and common nationhood to the usual gravity regressors, then adds a currency-union dummy and a measure of bilateral exchange-rate volatility, on a panel of 186 countries and dependent territories spanning five benchmark years (1970, 1975, 1980, 1985, 1990) and over 33,000 country-pair observations, including more than 300 observations in which both countries use the same currency. The headline finding is that exchange-rate volatility has the expected small negative effect on trade, but sharing a currency is associated with a far larger, economically enormous increase in trade: the point estimate implies that two countries using the same currency trade over three times as much as two otherwise-identical countries with separate currencies, an effect Rose shows is more than an order of magnitude larger than the estimated gain from eliminating exchange-rate volatility entirely. The result survives an extensive battery of robustness checks – alternative samples (excluding intra-developing-country pairs, excluding major “key-currency” countries, excluding whole regions, excluding highly asymmetric income or size pairs), alternative measures of currency union and exchange-rate volatility, alternative distance and language measures, ten different sets of additional controls for omitted variables, alternative estimators (Tobit, weighted least squares, Heckman selection, random effects, quantile and robust regression), a trade-growth specification, and an instrumental-variables approach using inflation-rate instruments to address possible reverse causality from trade to exchange-rate volatility. Rose is explicit that the paper cannot explain why the estimated effect is so large relative to the small transaction-cost savings usually assumed to be at stake, and that as a reduced-form gravity exercise it cannot deliver a structural welfare number – but concludes that the trade, and likely the welfare, benefits of currency unions have been substantially underestimated in the existing literature.
Summary of a classic paper, AI-assisted and human-reviewed. See the linked original for the authoritative claims and full conditions.
Questions & answers
Q1. What is the augmented gravity equation the paper estimates, and what does each new term add?
Rose estimates ln(X_ijt) = beta0 + beta1 ln(Y_iY_j)_t + beta2 ln(Y_iY_j/Pop_iPop_j)_t + beta3 lnD_ij + beta4 Cont_ij + beta5 Lang_ij + beta6 FTA_ijt + beta7 ComNat_ij + beta8 ComCol_ij + beta9 Colony_ij + gamma CU_ijt + delta V(e_ij)_t + error, where X is bilateral trade, Y is real GDP, D is distance, and the added dummies control for a common land border, common official language, shared regional trade agreement, common nationhood, shared colonizer post-1945, and colonial relationship (Sec. 4.1, “Mortar,” p. 9). The two coefficients of central interest are gamma, the effect of sharing a currency (CU, a 0/1 dummy), and delta, the response of trade to exchange-rate volatility V(e) – the latter measured as the standard deviation of the first-differenced log monthly exchange rate over the preceding five years.
Q2. What is the headline point estimate for the effect of a currency union on trade, and how large is it in context?
The benchmark pooled OLS regression (Table 1, with year effects) gives gamma approximately 1.21, so that exp(1.21) is approximately 3.35 – Rose’s point estimate is that “countries with the same currency trade over three times as much with each other as countries with different currencies” (Sec. 5.1, “The Killer App,” p. 13). The estimated volatility effect, delta approximately -0.017, implies that hypothetically eliminating exchange-rate volatility entirely (reducing it from its 7% sample mean to 0%) would raise log bilateral trade by only about 0.12, or roughly 13% – so “entering a currency union delivers an effect that is over an order of magnitude larger than the impact of reducing exchange rate volatility from one standard deviation to zero” (Sec. 5.1, p. 14).
Q3. How does the data set and sample size compare with what earlier studies used?
The panel covers 186 countries, dependencies, and territories for which the UN Statistical Office collects trade data, pooled across 1970, 1975, 1980, 1985, and 1990, giving 33,903 usable bilateral trade observations in total, of which 330 involve a pair that shares a currency (Sec. 4.2, “Bricks,” pp. 10-11). Trade data come from the World Trade Database (Feenstra, Lipsey and Bowen 1997, covering an estimated 98% of world trade); income and population come from the Penn World Table 5.6, supplemented by World Bank WDI data; distance, language, and colonial-history data come from the CIA World Factbook. Rose notes explicitly that many (though not all) of the 330 currency-union observations involve small or poor economies unlike the (then) eleven euro-area members, so he cautions against taking the exact point estimate too literally when extrapolating to EMU.
Q4. What happens to the raw, unconditional correlation between trade and sharing a currency once the standard gravity controls are added?
Rose notes that the simple bivariate correlation between log bilateral trade and the currency-union dummy is “small and negative,” and that the correlation between trade and exchange-rate volatility is similarly small and negative (Sec. 4.2, p. 11) – it is only after conditioning on income, population, distance, and the other gravity-model controls that the currency-union coefficient becomes “positive and both economically and statistically large,” since currency-union pairs in the raw data disproportionately involve small, poor, or geographically close economies that would trade heavily (or little) for other reasons.
Q5. What sample perturbations does the paper use to check that the result is not driven by a narrow subgroup of observations, and what does it find?
Table 2’s sensitivity checks exclude, in turn: all purely intra-developing-country trade; all observations involving the five “key-currency” countries whose money is used elsewhere (Australia, France, New Zealand, the UK, and the US); all African trade (and so the CFA franc zone); all trade involving Europe, the Antipodes, and the Pacific; all trade involving the Americas and Caribbean; highly trade-concentrated pairs; and pairs with very large income or income-per-capita disparities or involving very poor (sub-$1,000 GDP per capita) or very small (under one million population) countries (Sec. 5.2, “Really?,” pp. 15-16). Both gamma and delta “retain their economic and statistical significance throughout these perturbations.”
Q6. How does the paper address the concern that currency-union status might itself be driven by trade (reverse causality / simultaneity)?
Rose argues the simultaneity concern is more plausible for exchange-rate volatility (countries may stabilize their exchange rate because they trade a lot) than for currency-union membership, since decisions to join or leave a currency union are rare and historically appear to be politically rather than commercially motivated (he cites Ireland’s departure from the sterling peg and CFA-franc-zone entry/exit decisions as examples) – but he nonetheless instruments for exchange-rate volatility (and, in a check, jointly for volatility and currency-union status) using three terms built from the two countries’ inflation rates (Sec. 5.3, “But…,” pp. 19-21). With exchange-rate volatility alone instrumented, gamma remains positive and significant; when currency-union status is also treated as endogenous, standard Hausman tests cannot reject exogeneity of CU at the 5% level (though they can at 1%), and Rose concludes he “treats these results with a grain of salt” given the difficulty of finding good instruments for currency-union incidence specifically.
Q7. Besides cross-sectional levels, what does the paper find using trade growth rates?
Regressing the 1970-1990 annualized growth in bilateral trade on the growth in output, output per capita, and FTA status plus the currency-union dummy gives a coefficient of 0.16 (s.e. 0.03, N=2989, R-squared=0.47): currency-union partners’ bilateral trade grew about sixteen percentage points faster per year than non-currency-union pairs, holding other factors constant (Sec. 5.2, p. 18-19) – a result Rose offers as a check that does not rely on the exact cross-sectional specification or on any particular year’s sample.
Q8. What channels does the paper offer – and explicitly decline to adjudicate between – for why the currency-union effect is so large?
Rose is candid that “the puzzle in this paper is that the effect seems to be so enormous. Why does sharing a currency have such a big effect on trade? The short answer is: I don’t know.” He lists candidate channels without resolving between them: a currency union may signal a serious, long-term commitment to integration that in turn encourages private-sector trade; it may induce greater financial integration; and money facilitates trade in its roles as unit of account and medium of exchange, so that price transparency is clearly higher within a currency union, “though it is wisest to conclude that we simply don’t know why” by how much (Sec. 6.2, “Ignorance,” pp. 22-23). Because the gravity model is explicitly a reduced form that is not tied to any particular structural model of trade, the paper also cannot provide a quantitative welfare estimate of currency union from the trade effect alone (Sec. 6.3, “Bliss,” p. 24), though Rose notes separate estimates (Frankel and Romer 1999) that a one-percentage-point rise in the trade/GDP ratio raises income per person by one-half to two percent, suggesting the implied welfare gains could be large even without the full tripling of trade.
Key terms in this paper
Definitions below follow the paper's own usage.
- Augmented gravity equation
- The paper's empirical specification (Sec. 4.1, eq. p. 9) for bilateral trade: the standard gravity regressors (the product of GDP, the product of GDP per capita, log distance) plus a common land border, common language, shared regional trade agreement, common nationhood, shared colonizer, and colonial-relationship dummy, to which a currency-union dummy (CU) and a measure of bilateral exchange-rate volatility (V(e)) are added as the variables of central interest.
- Currency-union trade multiplier
- The paper's headline point estimate (Sec. 5.1, p. 13): a currency-union coefficient gamma of approximately 1.21 in the benchmark pooled specification, implying via exp(1.21) approximately 3.35 that two countries sharing a currency trade just over three times as much with each other as two otherwise-comparable countries with separate currencies -- an effect the paper finds to be an order of magnitude larger than the estimated impact of eliminating exchange-rate volatility.
- Home-bias benchmark
- The paper's framing reference point (Sec. 3, citing McCallum 1995 and Helliwell 1996): intranational trade between two regions of the same country is more than twenty times larger than trade between comparable regions on opposite sides of an international border, even for well-integrated areas; the paper treats a shared currency as one measurable component of the broader bundle of commercial, legal, and cultural commonalities that produce this home-bias gap, and interprets its currency-union estimate within that context rather than as a stand-alone structural effect.