1 Introduction

Since 2018 the European Commission has, on various occassions, expressed its ambition to strenghten the role of the common European currency in the world (Juncker 2018). Embedded in its strategy to “stimulate the openness, strength and resilience of the EU’s economic and financial system”, the European Commission mentions concrete measures to promote the international use of the euro (EUR). The European Commission in particular aims at establishing the EUR as a means of payment in the energy and the commodities sectors, at increasing depth and liquidity of the EU’s capital markets for example by issuing euro-demominated bonds, and at promoting sustainable finance. In addition, the European Commission is supportive of the introduction of a European digital currency, also to bolster the attractiveness of the EUR (European Commission 2021). Given the mixed performance of the EUR with regard to its perceived stability during the euro crisis and its decreasing relative importance as an international reserve currency, it is difficult to assess the prospects of success measured against these ambitions (Maggiori et al. (2019), The Economist (2019)).

Not only since Mr. Juncker’s announcement in 2018 international reserves have been of considerable interest for academics and policy makers. International currency reserves not only increasingly play an important role for international currency markets and exchange rates but also more broadly in terms of geopolitics and the global economy. During recent years, holdings of international reserves have grown at an extremely fast pace, as shown by Fig. 1. According to the IMF International Financial Statistics (IFS) international reserves have seen a more than eightfold increase over the last two and a half decades.Footnote 1 This development has largely been driven by a small group of countries, as the ten biggest holders of reserves account for 70% of worldwide international reserves. China and Japan alone managed around 39% of worldwide international reserves in 2016.Footnote 2

Fig. 1
figure 1

Volume of international reserves (in trillion USD). Notes: The line (left axis) corresponds to aggregated worldwide international reserves volumes without gold. The bars (right axis) display respective year-over-year growth rates. [Please use colors in print.] Source: IMF International Financial Statistics

Full size image

The academic literature dealing with international reserves can broadly be separated into research questions concerning what determines the size of a country’s overall reserve holdings and those concerning a country’s drivers for its currency composition. We focus on the latter. While empirical examinations of currency compositions are hampered by limited access to relevant IMF data on country-specific reserve compositions, we compile a new dataset based on all countries publicly disclosing their data.

Making use of country-specific data, we test the relevance of transaction motives in determining currency compositions. Transaction motives with respect to the currency compositions of foreign exchange reserves refer to bolstering a country’s preparedness for potential foreign exchange interventions, current account interventions, or temporary import financing. Our main findings suggest that currency compositions of international reserves under central bank management are influenced by currency pegs and trade patterns. Thus we are able to confirm previous academic findings (see Heller and Knight (1978), Dooley et al. (1989), Eichengreen et al. (2000) and recently Aizenman et al. (2019)). Unlike these papers, we place more focus on the specific role of the EUR, which since its introduction has been considered a serious competitor to the USD on international currency reserve markets by some. For instance, Bergsten (1997) calls the EUR the new rival to the USD and expects the establishment of a ‘bipolar currency regime’ and Chinn and Frankel (2008) describe scenarios in which the EUR will replace the USD as the main international reserve currency.

In particular, we aim to disentangle the impact of the euro crisis on the euro’s role as an international reserve currency. Splitting the sample into pre-crisis and crisis subsamples, we analyse the effect of the crisis on the determinants of currency compositions. We identify a significant and negative impact of the euro crisis on EUR holdings across countries. Additionally, our findings suggest that the euro crisis impacted on the effect of the determinants of currency compositions, rendering proxies for bilateral economic interdependence more relevant. We cautiously interpret this as a setback for the EUR as an international reserve currency. In a final exercise, we use our model to predict currency compositions of the three most important non-disclosing holders of international reserves, China, Japan, and Saudi Arabia.

The structure of this paper is as follows. In Sect. 2, we address the related literature. In Sect. 3, we present our data and descriptive findings. Section 4 is dedicated to analyzing determinants of country-specific reserve compositions, the role of the euro crisis and provides predictions of currency shares for non-disclosing countries. In Sect. 5, we summarize and discuss our findings.

2 Literature

The related literature can be divided into two fields. The first covers contributions analysing the general level of international reserve holdings, the second focuses on contributions dealing with the composition of reserves (see for example the introduction into these two strands by Beck and Weber (2011).

The literature addressing the question of why countries hold international reserves identifies three main motives: A precautionary motive, a mercantilist motive, and an investment motive. Relevant contributions are i.a. Heller (1966), Dooley et al. (2003), Rodrik (2006), Aizenman and Lee (2008), Calvo et al. (2012), Goldberg et al. (2013).

The issue of why countries amass foreign exchange reserves is linked to the question of how central banks choose the currency composition of their foreign exchange reserves. On this subject, the literature mainly distinguishes between transaction and classical portfolio investment considerations. Transaction motives with respect to the currency composition of the foreign exchange reserve portfolio arise from the central bank’s potential need to conduct foreign exchange interventions, current account interventions, or temporary import financing. The portfolio investment motive is characterized by a yield/risk trade-off.

However, empirical investigations trying to explain currency compositions are often hampered by poor data availability. One important source is the IMF’s COFER database (International Monetary Fund 2016). On very rare occasions, the IMF grants access to its confidential country-specific data. Influential contributions based on these data are Heller and Knight (1978), Dooley et al. (1989), Eichengreen et al. (2000). Other contributions rely on publicly accessible data, as for example Wong (2007), Chinn and Frankel (2008), Eichengreen et al. (2016), Aizenman et al. (2019), Ito and McCauley (2019).

Finally, while the previously mentioned contributions focus on economic explanations of international reserve compositions, political factors also play a role. Only few would deny that the unique role of the USD as an international reserve currency, and thus the United States’ so-called “exorbitant privilege”, is also the result of its political weight in the world. For example, according to Posen (2008) foreign reserve holdings “depend as much on foreign policy as economics”, while stressing the particular importance of security ties and pointing to high USD holdings of states as different as Saudi Arabia, Panama, and Taiwan. Frankel (2012) states that from a historical perspective the status of an international reserve currency has been associated with political and military strength and prestige. Another recent contribution to the literature focusing on political factors is Eichengreen et al. (2016). In this piece of work, the authors explain countries’ reserve currency preferences as a function of their dependence on the US for their security reasons.

3 Data and Descriptives

3.1 Data

Empirical work on the composition of international reserves has been somewhat limited by the discretion exercised by authorities in charge of reserve management. There are three main sources which provide insight into international reserves and its composition. The most general source is IMF’s International Financial Statistics (IFS), comprising the overall reserve holdings of most countries in the world. The IMF’s COFER database as another potential data source contains aggregated reserve holdings broken down by major international reserve currencies (International Monetary Fund 2016). It does not show individual reserve holdings and reporting is on a voluntary basis. Furthermore, of 145 reporting countries only 97 are listed by name on the IMF webpage. A third source consists of national data and is the foundation of this paper. To our knowledge, a total of 36 countries (plus the euro area) individually disclose the currency composition of their international reserves in different ways (mostly central bank annual reports). The advantage of this source is that it allows us to draw conclusions about the currency composition of worldwide total international reserves as well as about country-specific foreign reserve compositions. At the same time, this source is selective as the majority of central banks still prefer not to disclose currency compositions of their international reserves.Footnote 3 ,Footnote 4

The total volume of international reserves in our dataset accounts for more than one fifth of the total COFER reserves (22.2%) and for more than one fourth of the allocated COFER reserves (29.0%) in 2016. Table 7 in the Appendix compares our dataset with both the IFS and COFER datasets of the IMF in terms of scope and level of aggregation. Table 8 in the Appendix provides the coverage of countries of our data and the respective sources.

3.2 Descriptives

Our individual country data displays the same strong and parallel increase in international reserves and the domination of the USD over the last decade as shown by the IMF’s International Financial Statistics and the IMF’s COFER statistics. Figure 2 shows that the USD was able to successfully defend its positions as the world’s main international reserve currency. To cast light on the potential impact of the Covid-19 pandemic on the allocation of currency reserves, IMF reserve data are presented. A closer look a these data does not reveal clear patterns for volumes and growth rates of international reserves in 2020 and 2021 (see Fig. 3).

Fig. 2
figure 2

Holdings in major reserve currencies (own data - in trillion USD). Notes: The graph shows aggregated reserve holdings for the five major currencies in our dataset. [Please use colors in print.] Source: Own data (mostly annual reports of central banks, see Table 8)

Full size image
Fig. 3
figure 3

Holdings in major reserve currencies (IMF data - in trillion USD). Notes: The graph shows reserve holdings for the five major currencies in the IMF COFER dataset. [Please use colors in print.] Source: IMF International Financial Statistics

Full size image

Our unbalanced panel consists of 23 countries from Europe (plus the euro area), six countries from the Americas, four countries from Asia, Mozambique as the only African country and Australia and New Zealand (Table 9 in the Appendix lists all countries in our sample). Following the IMF World Economic Outlook classification, 21 countries (plus the euro area) are considered industrial countries and 15 are classified as emerging countries. Despite an overrepresentation of European countries, our sample includes countries from all regions of the world with a considerable variance in terms of economic development. Still, we are aware that we have to be cautious with regard to the generality of our results. The decision to disclose currency compositions might not be orthogonal to the mechanism driving currency compositions.

Regarding our sample, Fig. 6 presents currency shares for industrial and emerging countries and shows that the USD is the dominant reserve currency in industrial countries. EUR and USD shares lie closer to each other in emerging countries than in industrial countries. Figure 6 also shows average currency shares when the sample is split according to membership in the euro area. For countries being part of the euro area, the USD makes up for the vast majority of currency holdings, showing unrivaled shares of above 80%. Looking at the development of the GBP and JPY, euro area countries exhibit a larger share of JPY most of the time compared to countries not being part of the euro area.

In total, our panel dataset contains the individual international reserve compositions of 36 disclosing countries (plus the euro area) and includes the eight major reserve currencies for different time spans between 1996 and 2016. In the euro area, both the national central banks and the European Central Bank (ECB) manage foreign reserves. In 2016, the ECB’s international reserves were composed of around 83% USD-denominated and 17% JPY-denominated assets. The eight major currencies in our sample are the USD, the EUR, the Japanese yen (JPY), the Canadian dollar (CAD), the Chinese renminbi (CNY), the British pound (GBP), the Deutsche mark (DEM), the Swiss franc (CHF), and the Australian dollar (AUD). The entire (unbalanced) panel roughly contains 4,300 observations. In our empirical analysis, we will focus on the four most important international reserve currencies, the USD, the EUR, the JPY and the GBP which all have a substantial average share throughout the given time period.

In order to discuss a few trends, we observe developments for a total of 30 central banks which are consistently part of our panel for the period 2006-2016.Footnote 5 Our data shows the direction in which the percentage shares of the eight major reserve currencies have evolved. Data gaps prevent doing so for the other central banks of the sample. For instance, 16 countries increased the proportion of USD in their international assets, while 12 countries reduced their exposure to the USD. Only seven countries increased their EUR share and a total of 17 countries decreased their EUR share. This is partly due to some countries introducing the EUR, which implied that EUR-denominated assets are no longer considered part of their international reserves. For the period under review, the AUD and the CAD experienced increased demand as no country or only a few reduced their reserve assets (0 and 2 respectively) while many countries increased their reserve shares (11 and 8 respectively).

In terms of (unweighted) aggregated reserves in our sample, the share of the USD slightly increased from 52.8% to 53.6%, whereas the share of the EUR decreased from 33% to 30.2%. Shares of the AUD and the CAD increased from 0.1% to 1.6% and 0.4% to 2.1% respectively. The CNY, often regarded as the rising international reserve currency, was certainly no international reserve currency by 2006, but by 2015 it had become part of the official international reserves of Australia, Italy, and New Zealand even though its share of the total international reserves has so far remained negligible (0.1% in 2016). The JPY and the GBP have remained rivals for the position of the world’s third international reserve currency, with shares fluctuating between 4% and 5%. Furthermore, it seems that countries increasingly diversify their foreign reserves portfolio. Indications hereof are a higher number of reserve currencies over time in our panel and also in the IMF COFER database, which was extended in the last years by the AUD, the CAD (both 2013), and the CNY (2016). While one could expect that higher national reserve holdings are also more diversified (i.e. higher number of different reserve currencies in the portfolio), our data does not support this notion. For example, the U.S. reserves are not more diversified than the reserves of several smaller countries. Table 1 presents aggregated currency shares in our sample for all countries, industrial and emerging countries respectively and shows that the EUR and the GBP gained shares in the early stage after the introduction of the EUR, especially in emerging countries. Table 10 in the Appendix reports country specific international reserve compositions.

Table 1 Share of currencies in total reserve holdings (percent)
Full size table

In our empirical analysis we include as control variables dummies denoting currency pegs to the two main reserve currencies USD and EUR and variables measuring bilateral economic dependencies between holding countries and the currency-emitting countries. We concretely proxy bilateral relations via trade, foreign currency public debt and liabilities in currency reserves of the private banking sector. Table 2 presents descriptive statistics for the main currency shares and the control variables. Further, Table 11 in the Appendix reports the sources of our set of explanatory variables. In our sample, the USD on average exhibits the highest share with 46% followed by the EUR with 38%. Countries in our sample import substantially less from the US than from the euro area (8.6% vs. 29%), while external debt (59% vs. 26%) and liabilities (56% vs. 14%) are on average more likely to be denoted in USD than in EUR.

Table 2 Descriptive statistics
Full size table

4 Empirical Analysis

4.1 Model

We follow Dooley et al. (1989) and Eichengreen et al. (2000) relating currency shares to its potential determinants. We estimate three models with a different set of independent variables separately for four reserve currencies as dependent variables. Our model specifications separately test for the effect of imports (Model A),Footnote 6 foreign debt (Model B) and foreign currency claims (Model C) on currency reserves, while all models include currency peg dummies. As some covariates are not available for the entire set of countries in our sample, the number of observations strongly depends on the model we choose. Furthermore, as altering the model also changes the composition of the countries in the sample, models cannot directly be compared with each other. The censored nature of the data (values for the dependent variable lie within the range of 0 and 1) makes the use of the Tobit estimator preferable (Beck and Weber 2011). As compared to linear models, in non-linear models the interpretation of the coefficients is less intuitive, for which reason we provide estimates of the average marginal effects if necessary. Table 12 in the Appendix reports country composition and number of observations across our model specifications.

4.2 Estimation

Exploiting the panel structure of the data, Table 3 presents the results from random effects Tobit estimations for Model A, which includes currency pegs and import variables as dependent variables. The actual magnitude of implied changes in currency holdings can be interpreted in the average marginal effects rather than in the coefficients. The coefficients typically overestimate the average marginal effects, especially for the JPY and the GBP regressions with a substantial amount of left-censored observations (i.e. currency share equal to 0). The coefficient of the USD peg is significantly positive in the regression for the USD and significantly negative in the regression for the EUR. On the contrary, the coefficient for the EUR peg is significantly negative in the USD regression and significantly positive in the EUR regression. We suspect a stronger impact of a USD peg to the USD share given a presumed different rationale of pegging to the USD. Pegging to the EUR might be more driven by historical ties with western European economies. For the case of eastern European countries, we consider the integration in western European value chains and the wish to join the euro area in the future explaining factors for an anchorage of a national currency to the EUR (however, joining the euro area in the future requires exchange rate stability in general and not increased levels of EUR reserve holdings in particular). The fact that several former French colonial countries in Africa have replaced pegs to the French franc with EUR pegs, when the latter was introduced, speak for this assumption of historical ties.

Indeed in the EUR regression the effect of the EUR peg is economically much less pronounced when compared to the effect of the USD peg on USD holdings. Generally speaking, countries using a currency peg want to hold relatively larger reserves in the given currency in order to back the exchange rate to which they committed. At the same time, relative holdings of the other main currencies are reduced. The average marginal effect of the USD peg is 0.365. All other things equal, the introduction of a USD peg corresponds to an increase of the USD share by 36.5 percentage points. The peg to the EUR marginally increases EUR holdings by 4 percentage points.

Table 3 Estimation results: Model A, Tobit
Full size table

Concerning trade, we find a significantly negative coefficient for imports from the euro area in the USD regression but we do not find imports from the US to be significantly correlated to EUR shares. In terms of average marginal effects, a 1 percentage point increase of imports from the euro area is correlated to a decrease of 0.15 percentage points in the USD share. Imports from Japan and the UK have a significantly positive influence on the share of the JPY and the GBP, respectively. On average, a 1 percentage point increase in imports from Japan and the UK relate to increases of 0.44 and 0.14 percentage points of the JPY and GBP share respectively. Central banks might to some degree be obliged to finance foreign trade with accumulated reserves (Soesmanto et al. 2015), thus rationalizing the finding that increased imports from the euro area or Japan reduce a central bank’s propensity to hold USD relative to other currencies. Contrarily, we do not find imports from the US and the euro area to be associated with increases in the USD share and the EUR share respectively. Our interpretation is that the transaction motive seems to hold for the two large currencies in terms of the effect of currency pegs, but less so for trade patterns. For the two smaller reserve currencies JPY and GBP the trade relations with the emitting countries appear to be more relevant in terms of the accumulation of reserves.

Table 4 displays results for both Models B and C, incorporating debt and liabilities of domestic banks denominated in foreign reserve currencies respectively. We focus on regressions with the shares of the USD and the EUR as dependent variables. Using these alternative explanatory variables reduces the number of observations substantially as most countries in our sample are either not indebted in foreign currencies or do not report so. Similarly, for foreign currency liabilities the sample size is reduced compared to Model A. Column 1 shows that debts denominated in both USD and EUR have a significantly negative association with the USD share. Column 3 reports the same relationship for liabilities denominated in both main reserve currencies and the USD share respectively. The negative coefficients for debt and liabilities denominated in USD can be interpreted similarly to the insignificant effect of imports from the US on USD shares reported earlier. USD shares might not necessarily be driven by bilateral proxies of economic interdependence but instead by a premium stemming from its role as the prime reserve currency.

Table 4 Estimation results: Models B and C, Tobit
Full size table

4.3 Robustness Checks

In this section, we discuss the robustness of our findings related to Model A. In order to control for country heterogeneity we employ in total three region fixed effects for the Americas, Europe and Asia. Adding country fixed effects overfits the model as the specification is not left with sufficient variation over time to identify the effects of our covariates. We present the results of estimating our benchmark Model A including region fixed effects in Table 13 in the Appendix. In our benchmark models we normalize imports to 0 for imports from emitting countries in regressions analyzing the determinants of other currency shares (i.e. imports from the euro area are set to 0 for euro area countries in the USD regression). Doing so prevents further sample attrition, particularly in the USD, JPY and GBP regressions as euro area countries are retained in the sample. As a robustness check we add currency emission fixed effects to our benchmark model. The regressions from Model A are estimated including a dummy for the United States, the UK and one dummy for all euro area countries. We present the results of this specification in Table 14 in the Appendix. The results for these two alterations remain qualitatively very similar compared to our benchmark model.

Additionally, we alter estimation methodologies for sensitivity analyses. We confront the results of the Tobit approach with results of random effects OLS panel regressions and Seemingly Unrelated Regressions (SUR) pioneered by Zellner (1962). SUR appear to be a plausible choice as by definition there is a dependence between the respective currency shares (shares add up to 1). Concretely, SUR allows the estimation of a system of equations with correlated error terms. Taking into account such correlations increases efficiency compared to the estimation of separate equations using OLS. For instance, Soesmanto et al. (2015) advocate the use of the SUR technique when examining determinants of international reserve holdings of the Reserve Bank of Australia (RBA). We focus on comparing the results from the different regression techniques for the two main currencies USD and EUR and for our main Model A.

Table 15 in the Appendix contrasts the results for regressions explaining the USD share. The system of equations we estimate with the SUR estimator consists of three linear equations explaining the USD share, the EUR share and the aggregated share of all remaining currencies. By subsuming all other currencies in one share we want to reduce the number of coefficients to be estimated and focus on explaining the choice between the two leading currencies. For USD shares in Model A, OLS and SUR estimates broadly confirm our findings from Tobit estimations. In the regressions from all three methodologies, coefficients of the currency pegs are economically relevant and highly significant. The previous finding of trade not being decisive in the accumulation of USD shares is confirmed in the SUR regression which even features a negative coefficient on imports from the US. Coefficients on other trade variables, however, are intuitive. Unlike the other regressions, the SUR regression results yield negative and statistically significant estimates on imports from all other three reserve currency countries, suggesting the relevance of trade patterns. Table 16 in the Appendix compares the results across regression techniques for the EUR. Similarly to the USD regressions, the qualitative results concerning the relations between currency pegs and currency shares are supported in the two alternative regressions techniques. Further, the SUR regression yields supportive results for the impact of trade patterns on the accumulation of the EUR currency share as the coefficient on imports from the euro area is positive and statistically significant.

In sum, results are mostly robust for currency peg and trade variables and can explain international reserve compositions reasonably well. Models B and C acknowledge the importance of international debt and international bank claims but present mixed results. In the following sections, we rely on our baseline Model A as a workhorse model.

4.4 Effect of the Euro Crisis on Euro Holdings

In the first decade after the introduction of the common European currency, the EUR share experienced a steady increase in international reserve holdings. The outbreak of the euro crisis in 2010 marked a turning point, as since then the EUR share has constantly lost ground. IMF COFER data document a peak in 2009 with a share of 28% of total allocated foreign reserves. Our more eurocentric data suggest the year 2008 was already a break date and document a EUR share in total foreign reserves of nearly 38% for that year. In both data sets, the EUR shares have decreased by 9 and 8 percentage points, respectively, since the respective peak years. Not surprisingly, scenarios of the EUR surpassing the USD as the world’s main reserve currency usually date from the years prior to the euro crisis (see Chinn and Frankel (2008)). Figure 4 illustrates the initially increasing and subsequently decreasing importance of the EUR as an international reserve currency, with 2007 as a base year (2007 = 100). In terms of the euro’s share in international reserves, 2015 was the currency’s temporary nadir according to our data since 2001.Footnote 7

Fig. 4
figure 4

EUR share in total reserves. Notes: IMF COFER. This graph shows the EUR share in international reserves according to our data (bars) and to the IMF COFER data (line). 2007 is defined as base year (=100%) to illustrate the parallel evolution. The vertical red line separates the sample into pre-crisis and post-crisis subsamples. [Please use colors in print.] Source: IMF COFER, own data (mostly annual reports of central banks, see Table 8)

Full size image

We use Chow tests to verify a structural break for EUR shares triggered by the euro crisis in 2010. To further examine effects of the euro crisis, we add a crisis dummy to capture a direct effect and interaction terms to capture an indirect effect to the baseline regression (see Table 5). Crisis years are marked by a dummy variable (1 for all years after 2010, 0 otherwise). Interaction terms are constructed as products of the dummy variable and the six independent variables of the baseline regression. We find a significant and negative coefficient for the crisis dummy in columns 2 and 3, which traces the reduction in EUR holdings post-2010. The coefficient of −0.08 in column 2 corresponds to a marginal reduction of EUR shares of around 6 percentage points. Interestingly, the interaction terms of the crisis dummy and covariates displayed in columns 3 and 4 suggest that channels based on transaction motives are strengthened after the euro crisis. For instance, the effect of a currency peg to the USD is more positive, the effects of imports from the US and the UK are more negative and the effect of imports from the euro area are more positive after the crisis. For the latter, note that the total effect of imports from the euro area after the crisis is still negative and insignificant. Yet, the tendency for estimates after the crisis suggests the interpretation that the euro crisis has damaged the attractiveness of the EUR on international reserve markets, causing a flight to quality out of the EUR. Consequently, after the crisis currency shares of the EUR might be explained more by transaction motives rather than stemming from the status of being a prime reserve currency.Footnote 8

Table 5 Estimation results with crisis interactions
Full size table

4.5 Actual and Model Predicted Currency Compositions

In this section, we confront model predicted and actual currency shares. We restrict ourselves to the two main international reserve currencies the USD and the EUR and make use of our baseline model, i.e. Tobit-type panel regressions with currency pegs and import shares. Figure 5 plots predicted and actual shares for both currencies for 2016. The majority of shares predicted for the USD and the EUR lie in the range between 40 and 60%. The model provides a reasonable fit on average as it is capable of assigning larger currency shares for those countries featuring larger actual shares. The correlation coefficients between actual and predicted shares are 0.64 for the USD and 0.48 the EUR respectively. Comparing actual to predicted currency shares further reveals interesting patterns about the fit of our model. The parsimony of the model does not capture country idiosyncrasies driving extreme choices (i.e. currency shares close to 0 or 1) in currency compositions. For instance, an important number of Latin American countries seemingly have a strong preference for holding USD-denominated assets while EUR shares are regularly larger than predicted in neighboring countries of the euro area. Potentially, other factors beyond trade patterns and currency pegs, such as geopolitical factors, are relevant for these countries. Further, many emitting countries exhibit large holdings of the respective other main currency, including the US with a large share in EUR-denominated assets due to a missing choice between two rivaling currencies.

Fig. 5
figure 5

Actual and model predicted shares of USD and EUR (2016). Notes: This table displays model predictions and corresponding actual values for USD shares (blue) and EUR shares (red) by country for 2016. For prediction we use our benchmark model with pegs and imports as explanatory variables (Model A). [Please use colors in print.] Source: Own data (mostly annual reports of central banks, see Table 8)

Full size image

Lastly, we predict currency compositions of the three largest - non-disclosing - reserve holders China, Japan, and Saudi Arabia, which together account for nearly 45% of worldwide international reserves in 2016 (see Table 6). Substituting respective import shares and currency pegs into our baseline estimation equation for USD, EUR, JPY and GBP we obtain the following predicted shares. For China, our model predicts a composition of its international reserves of 39% USD, 37% EUR, 11% JPY and 7% GBP assets. The model further suggests that Japanese international reserves are invested to 47% in USD, to 40% in EUR and to 7% in GBP. For Saudi Arabia, the US peg translates itself via an economically significant coefficient into a highly USD-dominated, predicted reserve composition, despite lively trade with the euro area. The predicted foreign reserves portfolio consists of 63% USD, 13% EUR, 11% JPY and 8% GBP-denominated assets.Footnote 9

For the three countries of interest, these estimation results are difficult to compare to actual compositions about which little is known. Nevertheless, possible bands within which currency shares may fluctuate can be derived from statements by economists and analysts. For China, estimates assume a USD share between 60% and 70% of its international reserves, a EUR share between 20% and 30%, and JPY and GBP shares together between 5% and 10% (see Hu et al. (2010), pp. 8-9, Morrison and Labonte (2013), p. 5, Wildau (2014), Neely (2017), p. 1).Footnote 10 For Japan, a high USD share in the country’s international reserves appears realistic. For instance, Wong (2007) is among the few committing herself to an, in our opinion, plausible USD share between 83% and 89%. For Saudi Arabia, which is regularly among the largest holders of foreign reserves, market observers assume central bank foreign reserves to be denominated almost exclusively in USD (Torchia 2015). Bearing in mind the challenge of predicting extreme currency shares and the role of political factors which our model omits, our predictions seem to follow broadly the tendencies of USD dominance for these three countries.

5 Conclusions

In this paper, we examine determinants of the composition of international reserves and aim to disentangle the impact of the euro crisis on currency compositions. We compile a new dataset based on publicly available central bank data on international reserves to overcome the lack of comprehensive data in this area. Relating these data to a selection of country-specific factors reveals that trade patterns and currency pegs are primary determinants of foreign currency holdings. Our analysis suggests the importance of transaction motives in determining the composition of currency shares. In particular, we find correlations between currency pegs and imports. Thus, we can confirm findings from papers based on access to confidential IMF data (see Heller and Knight (1978), Dooley et al. (1989) and Eichengreen et al. (2000)).

We further show that the euro crisis caused a break in the rising relative significance of EUR holdings. In pre-crisis years and since the EUR introduction in 1999, the EUR was able to establish itself as the world’s second reserve currency after the USD and was even considered a rival to the USD as the world’s main international reserve currency. Since 2010, however, EUR shares have been declining by trend. Additionally, the euro crisis potentially strengthened transaction motives for the determination of EUR shares, which we cautiously interpret as a setback for the EUR as a rival the USD as the main international reserve currency.

Additionally, this paper argues that a simple model can, on average, explain currency compositions reasonably well, whereas deviations from predicted shares in some cases suggest that additional factors such as geopolitical factors might play a role. We use our model to predict currency shares for China, Japan, and Saudi Arabia, the largest - non-disclosing - reserve holding countries. Our model predicts USD dominance in these countries’ reserve holdings.

The development of international reserves remains of large general interest for academics and policy makers due to its relevance for currency markets, exchange rates, geopolitical factors, and the global economy as a whole. Academic research can further contribute to explain both recent and future dynamics. New developments such as the rise of digital currencies and the continued efforts of the European Union and China to strenghten their respective currencies motivate further work in this field.