Cross-Border Currency Exposures
  • 1 https://isni.org/isni/0000000404811396, International Monetary Fund
  • | 2 https://isni.org/isni/0000000404811396, International Monetary Fund
  • | 3 https://isni.org/isni/0000000404811396, International Monetary Fund

Contributor Notes

Author’s E-Mail Address: ljuvenal@imf.org (corresponding author), dgautam@imf.org, benetria@tcd.ie, martin.schmitz@ecb.int

This paper provides a dataset on the currency composition of the international investment position for a group of 50 countries for the period 1990-2017. It improves available data based on estimates by incorporating actual data reported by statistical authorities and refining estimation methods. The paper illustrates current and new uses of these data, with particular focus on the evolution of currency exposures of cross-border positions.

Abstract

This paper provides a dataset on the currency composition of the international investment position for a group of 50 countries for the period 1990-2017. It improves available data based on estimates by incorporating actual data reported by statistical authorities and refining estimation methods. The paper illustrates current and new uses of these data, with particular focus on the evolution of currency exposures of cross-border positions.

1 Introduction

The analysis of the role played by different currencies in trade and finance has lately been at the center of the agenda in international economics (Gopinath and Stein, 2018). This paper contributes to this literature by focusing on the currency composition of countries’ external assets and liabilities. To this aim, we construct a new dataset on the currency composition of the international investment position (IIP) building on earlier estimates by Lane and Shambaugh (2010a) and Bénétrix, Lane and Shambaugh (2015).

A key refinement in our new dataset covering 50 countries over the period 1990–2017, centers on the expansion of available actual data, made possible thanks to a recent IMF survey on the currency composition of the main IIP components. Currencies of denomination were broken down into the five Special Drawing Rights (SDR) currencies (i.e. US dollar, euro, Japanese yen, Pound sterling and renminbi), domestic currency (if different from the five SDRs), and “other currencies” which bundle up all other foreign currencies not included in the previous two categories. This valuable new data obtained via the IMF survey are complemented with other sources of actual data: (i) data on the currency composition of portfolio equity and debt assets from the IMF’s Coordinated Portfolio Investment Survey (CPIS), (ii) portfolio debt data reported to the European Central Bank (ECB) and (iii) banks cross-border positions reported to the Bank of International Settlements (BIS) available through its Locational Banking Statistics (LBS). In addition, where actual data are missing, estimates for currency denomination based on well-known methodologies are used.1 In some cases, these estimates have been refined, for example, to allow for a different treatment of equity and debt in foreign direct investment (FDI).2

A comparison between actual and estimated data which can naturally only be done for the years for which both are available reveals that by and large the main patterns of actual data are captured by the estimates. However, some contrasts are present. For example, on the asset side, we observe that US dollar weights tend to be underestimated, in particular for portfolio equity and FDI equity. This likely reflects the fact that our estimates rely on the geography of cross-border positions as the closest predictor of currency of denomination. However, certain cross-border asset positions which do not involve the issuers resident in the US are denominated in US dollars, due to its widespread international usage.3

On the liability side, domestic currency weights for portfolio debt liabilities are underestimated in particular for emerging countries. The estimation of this item relies largely on the BIS international debt issuance statistics. One source of discrepancy between actual and estimated data could be driven by the lack of information on domestically issued debt securities held by non-residents which may be denominated in domestic currency.4 These issues highlight that better availability of actual data is key for further improving the accuracy of the dataset.

We use the dataset to construct a number of indicators to guide our study. The core of our analysis focuses on an indicator of aggregate foreign currency exposure, defined as the net foreign assets denominated in foreign currency as a share of total assets and liabilities (Lane and Shambaugh, 2010a). It measures the overall sensitivity of a country’s external balance sheet to a uniform movement of its domestic currency against all foreign currencies.

Our key findings are described as follows. First, the analysis of the long-term dynamics shows that the cross-country distribution of foreign currency exposures has shifted significantly towards long positions (i.e. larger gross foreign assets in foreign currency than liabilities in foreign currency) between 1990 and 2017. This pattern was driven by both improving net international investment positions (excluding reserves) and larger net liability positions in domestic currency (resulting from larger gross equity liabilities). Second, with respect to 2012, the last year reported in Bénétrix et al. (2015), the distribution of foreign currency exposures has shifted slightly towards longer positions between 2012 and 2017. Third, most of the adjustment in net foreign currency exposure took place ahead of the global financial crisis period, as the post-crisis period has been characterized by relatively persistent net foreign currency exposures. In the pre-crisis period not only overall net international investment positions improved, but also the currency composition of external balance sheets moved towards larger shares of liabilities denominated in domestic, rather than foreign currency (particularly for equity, but also debt). The trend towards larger shares of equity liabilities being denominated in domestic currency continued after the global financial crisis, but overall changes were limited due to the persistence in net international investment positions over the post-crisis period. Fourth, net foreign currency exposures continue to differ substantially across country groups. Advanced economies have a large share of their debt liabilities denominated in domestic currency, while for emerging economies debt liabilities are predominantly denominated in foreign currency, despite the increase in local currency debt issuance. Net short positions in foreign currencies continue to be more prevalent in emerging economies, which have a history of borrowing heavily in foreign currency, a phenomenon referred to as “original sin” (Eichengreen, Hausman and Panizza, 2003), thereby leaving several emerging economies exposed to adverse valuation effects if the domestic currency weakens.

The dataset is also used to compute financial exchange rates, which are an input for the calculation of valuation effects, and a measure of international financial integration broken down by currency. Exchange rate movements can have significant effects on the economy through a valuation channel which reflects capital gains or losses on the international balance sheet. These valuation effects on foreign currency exposures have been found to be sizable and are moreover a key channel for the international transmission of monetary policy, as changes in the monetary policy stance also have wealth effects via exchange rate movements (Georgiadis and Mehl, 2016). The valuation effect depends on whether a country is long or short on foreign currency. For example, if a country has a negative net position in foreign currency, a depreciation can generate significant negative wealth effects. Our findings confirm the results of Lane and Shambaugh (2010a) that trade weighted exchange rates are not the adequate measure to evaluate wealth effects of currency movements and financial exchange rates should be used instead. One of the tools we provide with this dataset is an updated measure of financial exchange rates which can be used to compute valuation effects.

The scale of exchange-rate induced valuation effects is determined by the interaction of the financial exchange rate index and the size of a country’s external balance sheet, which is given by the well-known measure of de-facto international financial integration (IFI) proposed by Lane and Milesi-Ferretti (2001). While this measure focuses on the size of the international balance sheet, little is known about its currency composition. Our data allow to look into the currency dimension of this IFI measure. A first glance at the aggregate measure confirms the main patterns documented in Lane and Milesi-Ferretti (2003, 2007 and 2018), who show that IFI increased from the 1990s until the global financial crisis and came to a halt afterward. On the currency dimension front, we find that: (i) euro-denominated cross-border holdings expanded rapidly between 1999 and 2007, but declined considerably after the global financial crisis; (ii) US dollar positions, after declining sharply in 2008, recovered quickly and have shown an upward trend since 2014, (iii) outside the US and euro area, there is an unambiguous dominance of cross-border holdings in US dollars throughout the whole sample period.5

The paper is organized as follows. Section 2 describes the data and its uses. Section 3 presents the building blocks to compute exchange rate valuation effects and the currency composition of the international financial integration measure. Section 4 focuses on the components and long-term dynamics of aggregate foreign currency exposures and shows our main results. Section 5 discusses the pre- and post- crisis adjustment of aggregate foreign currency exposures and Section 6 concludes. Details about data sources and estimation methods used are detailed in the Appendix.

2 Data

The dataset in this paper builds on the contributions by Lane and Shambaugh (2010a) and Bénétrix et al. (2015) who provide estimates of the currency composition of external assets and liabilities for a sample of 117 countries between 1990–2004 and 1990–2012, respectively.

Our dataset offers a number of improvements with respect to those efforts. Most notably, it makes use of actual data reported to the IMF by national statistical authorities. This, together with new access to more granular information, means a substantial reduction in the dependence on estimated currency weights. Previous vintages relied more heavily on dimensions such as geography, gravity models or bond issuance to obtain the currency composition of each component of the IIP. In some cases estimates have been refined, for example, to allow for a different treatment of equity and debt in foreign direct investment (FDI). This is a crucial improvement in the light of the increasing relevance of FDI debt positions since the global financial crisis (Blanchard and Acalin, 2016; Damgaard, Elkajer, and Johansen, 2019; and Lane, 2015).

Our approach consists of taking actual data whenever available and extend its coverage by using estimated data based on improved methodologies. More precisely, we complete the coverage of actual data with estimated currency weights for each component of the IIP.6 This allows for filling the gaps in actual data and expand its coverage over the period 1990–2017.7 Given the limited availability of actual data, we focus on a sample of 50 countries, which are part of the IMF External Balance Assessment and/or External Sector Report.8 Taken together, these countries account for over 90 percent of world’s GDP.

The Appendix provides a detailed description of the methods and data sources used to estimate the currency composition of international balance sheets. In what follows we provide a summary which emphasizes the improvements with respect to earlier work.

2.1 Actual Data

One of the main contributions of this dataset, which distinguishes it from earlier work, consists on the incorporation of actual data on the currency composition of the IIP. These are obtained from a variety of sources: (i) a survey sent by the IMF to country authorities; (ii) the Coordinated Portfolio Investment Survey (CPIS) (Table 2) dataset; (iii) European Union (EU) countries’ data reports to the ECB and; (iv) the Locational Banking Statistics (LBS) from the Bank of International Settlements (BIS).

Table 1:

Correlations of Exchange Rate Indices, 1990–2017

article image
Notes: The table shows the the means and medians of within-country correlations between the percentage change in exchange rate indices. Sample includes annual data over the period 1990–2017. A and L, refer to asset and liability weighted indices, T denotes trade-weighted index and I is the financial exchange rate index.
Table 2:

Foreign Currency Exposure (FXAGG) and Subcomponents

article image
Notes: The table shows the mean, median, and the interquartile range of the foreign currency exposure indicator (FXAGG) and its main subcomponents for 2017.

The main source of actual data is a survey sent to authorities by the IMF Research Department in collaboration with the Statistics Department. The survey requested data from 1990 on the main components of the IIP broken down into the five SDR currencies (i.e. US dollar, euro, Japanese yen, Pound sterling and renminbi), domestic currency (when different from the five SDRs), and “other currencies” which bundle up all the other foreign currencies not included in the previous two categories. Country authorities responded to the survey on a voluntary basis. For recent years, around 52 percent of countries reported some data but as we go back in time the coverage is more limited. Section A.1.1 in the Appendix provides more details about the survey.

We complemented the survey data with Table 2 from the CPIS, which includes the currency composition of portfolio equity and portfolio debt assets, and ECB data on the currency of denomination of portfolio debt assets and liabilities which Euro Area countries are required to submit. In addition, we use the currency of denomination of cross-border positions of banks sourced from the BIS LBS. Tables A.3 and TA.4 in the Appendix describe the coverage of actual data for each country.

Our final dataset extends the coverage of actual data with estimated currency positions. A summary of the estimation methods is summarized below, while the Appendix explains in detail the methods used to estimate each item.

2.2 Foreign Assets

The asset side of a country’s international investment position contains five main items: portfolio equity, foreign direct investment (equity and debt), portfolio debt, other investment (mainly bank-related), and reserves.

The sources of actual data for portfolio equity assets are the IMF survey and CPIS Table 2. In order to fill the data gaps going backwards we use the method based on geography as in Lane and Shambaugh (2010a) and Bénétrix et al. (2015). The CPIS dataset provides the geographical location of portfolio equity asset holdings for 82 reporting countries and 220 host countries since 2001. The use of geographical data to calculate currency weights relies on the assumption that equity issued by a country is denominated in the currency of that country.9

Actual data from the survey revealed that the debt component of FDI has a very different currency composition with respect to equity. Therefore, we estimate their currency of denomination separately. This is an improvement with respect to Bénétrix et al. (2015) which treat FDI as equity only. We obtain the share of equity in FDI from the IMF International Financial Statistics (IFS). For the cases in which actual data are missing we use estimated data on FDI equity as follows. Between 1990 and 2008 we use the data from Bénétrix et al. (2015). For the years between 2009 and 2017 we use data from CDIS which include outward and inward stocks of direct investment for 108 reporting countries. As in the case for portfolio equity assets, we assume that direct investment is denominated in the currency of the host country. For FDI debt, whenever available, we used actual data from the IMF survey. Our synthetic data were obtained using the currency weights of portfolio debt assets.

The sources of actual data for portfolio debt are the IMF Survey, CPIS Table 2 and ECB data. We expand the coverage using the estimation method by Lane and Shambaugh (2010a) which is based on combining the geography of portfolio debt assets positions from CPIS with the currency of denomination of host countries’ bonds issued in international markets.

Actual data for Other Investment is sourced from the IMF survey and extended backwards based on the BIS LBS since banking assets are the largest component of this item.

Data on the currency composition of reserves from 1990 to 2012 are sourced from Bénétrix et al. (2015). In order to update the data up to 2017 we construct a second dataset, which covers the period 2010–2017, by obtaining or estimating reserve weights from non-confidential sources such as Central Banks or Ministry of Finance Publications and publicly available IMF Currency Composition of Official Foreign Exchange Reserves (COFER) data (details on data sources are detailed in the Appendix). We merge the series from 1990–2012 with the ones from 2010–2017 using an interpolation method.

2.3 Foreign Liabilities

The liabilities side of a country’s international investment position contains four main items: portfolio equity, foreign direct investment (equity and debt), portfolio debt and other investment (mainly bank-related).

Consistent with the treatment on the asset side, portfolio equity and FDI equity liabilities are assumed to be in the currency of the host country, which implies exposure in domestic currency.10 Whenever available, we used actual data on FDI debt from the IMF survey and extend its coverage backwards using as synthetic weights the currency breakdown of portfolio debt liabilities.

Actual data on portfolio debt are obtained from the IMF Survey and ECB data. For most countries, synthetic data for the currency breakdown are obtained from the BIS international debt issuance statistics. The dataset covers all debt securities issued by non-residents and includes a comprehensive breakdown by currency. However, for emerging markets this dataset tends to underestimate the share of domestic currency issuance, as the BIS does not report the currency composition of domestically issued debt securities and there is no information of the proportion of domestically issued securities which is held by non-resident investors. Due to the importance to correct for this, for a sample of 19 countries we use the share of central government securities held by foreigners and denominated in domestic currency from Arslanalp and Tsuda (2014).11 As in the case of other series, we used actual data whenever available and extend the series using synthetic data. Finally, the assembly of other investment liabilities proceeds analogously with the one for other investment assets.

2.4 Comparison: Actual and Estimated Data

The estimation of the currency composition of the different components of the IIP requires making a number of assumptions and in some cases data availability is limited. Section A.4 of the Appendix provides a comparison between actual and estimated currency weights. This can naturally only be done for the years for which actual data are available. In that case, estimated data are used for comparison purposes but are not part of the final dataset since the hierarchy of sources gives priority to actual data.

Overall, the main patterns of actual data are captured by the estimates. However, some contrasts are present. On the asset side, we observe that US dollar weights are underestimated, in particular for portfolio equity and FDI equity for which synthetic data rely on the geography of cross-border positions as the closest predictor of currency denomination. However, given the dominance of the US dollar, it is likely that certain cross-border positions which do not involve the issuers resident in the US are denominated in US dollars and these would not be captured by geography. Another source of discrepancy could be driven by the currency of denomination of positions vis-à-vis Special Purpose Entities (SPEs) which may not be captured by geography.

On the liability side, domestic currency weights of portfolio debt liabilities in domestic currency are underestimated in particular for emerging countries. This happens because the BIS does not report the currency composition of domestically issued debt securities and there is moreover no information of the proportion of domestically issued securities which is held by non-resident investors. Other investment liabilities exhibit substantial dispersion with US dollar weights revealing a slight underestimation in the early to mid-2000s. As in the case of the asset side, this item is proxied using banking data only while other components are not estimated. This could partly explain the differences between actual and estimated data.

Given that there are some differences between actual and estimated data, it is very important to incorporate as much actual data as possible.

2.5 Data Uses

The dataset we construct and make publicly available contains asset, debt asset, liability, and debt liability weights. These are obtained combining the shares of each component of the IIP from the External Wealth of Nations dataset by Lane and Milesi-Ferretti (2007) with the corresponding currency weights which we calculate using the methods described in this section as well as in the Appendix. Debt assets include portfolio debt, FDI debt, other investment and reserve assets, while debt liabilities include portfolio debt, FDI debt and other investment.

The dataset can be used to compute a number of indicators which will guide our analysis. In particular, we calculate a measure of valuation effects in response to exchange rate movements which contains two main ingredients: financial exchange rates and a measure international financial integration. After introducing this indicator we zoom into the currency composition of the stock of external assets and liabilities and compute US dollar and euro measures of international financial integration. The core of our paper centers around a measure of aggregate foreign currency exposure.

3 Exchange Rate Valuation Effects: Building Blocks

Exchange rate movements can potentially have significant effects on the economy through a financial and a trade channel. The financial channel refers to capital flows and capital gains or losses related to external balance sheets. The trade channel operates through the effects on exports and imports.

In order to assess the impact of currency movements on gross stocks, Lane and Shambaugh (2010a) propose the following measure of valuation effects:

VALi,t+1XR=%ΔIi,t+1FIFIi,t,(1)

where VALi,t+1XR indicates the currency-induced valuation change related to currency movements for country %ΔIi,t+1F is the percentage change in the net financial exchange rate index during the period t + 1 and IFIi,t denotes the size of the external balance sheet, measured by the international financial integration index proposed by Lane and Milesi-Ferretti (2007). This is calculated as IFIi,t = (Ai,t + Li,t/GDPi,t, where Ai,t and Li,t are the end-of-period t gross stock of external assets and liabilities expressed in current US dollars.

3.1 Financial Exchange Rate Indices

One of the building blocks for the calculation of valuation effects is the financially weighted exchange rate index based on net foreign currency exposures. This provides a measure of the sensitivity of country’s external balance sheets to currency movements and is given by

Ii,t+1F=Ii,tF(1+ωi,j,tF%ΔEi,j,t).(2)

%ΔEi,j,t is the percentage change in the bilateral end-of-period nominal exchange rate between the currency of country i and the foreign currency j between t and t + 1. ωij,tF is the net financial weight of country i to currency j in period t. This is calculated as

ωi,j,tF=ωi,j,tAsi,tAωij,tLsi,tL,(3)

where ωij,tA(ωij,tL) is the proportion of assets (liabilities) denominated in currency j, Si,tA=Ai,tAi,t+Li,t is the share of assets in the country external balance sheet and si,tL=Li,tAi,t+Li,t is the corresponding share for its liabilities. By construction si,tA+si,tL=1.

While a depreciation of the domestic currency tends to be associated with a net financial outflow (via interest parity conditions) and thus an improvement in the net IIP, that same exchange rate movement could amplify or mitigate that flow effect via valuation changes in gross external positions. The direction of the latter depends on whether the country has a long or a short position in foreign currencies. The magnitude, depends on the size of the external balance sheet.

Traditionally, the effect of exchange rates on external positions and its sustainability has been studied by using trade-based exchange rate indices (originally designed to study trade flows as the outcome variable) and their effect on the balance of payments counterpart of the financial account, i.e. the current account or its main component, the trade balance. Although this approach could be a fair approximation of the effects of a uniform change in a currency value on financial flows and external positions, it has its limits. First, it is based on a “proxy” for the financial account: the current account. While conceptually, one should be a close reflection of the other, a more precise assessment should focus on the financial account itself.12 Second, it is based on the geography of trade, which does not necessarily match the counterparts of international investment positions and the currencies in which these are denominated. Third, it does not provide a one-to-one link between the change in value of the external positions – a dimension increasingly relevant in the light large scale of international balance sheets, as documented by Lane and Milesi-Ferretti (2001, 2007) – and exchange rate movements. Given these fundamental differences, it is reasonable to expect that financial and trade-based exchange rate indices exhibit heterogeneous dynamics.

In fact, the trade-weighted and financial exchange rates can move in opposite directions, making the latter a crucial tool for the assessment of the sustainability of external imbalances and for surveillance purposes. Therefore, as a byproduct of this dataset we provide an updated measure of financial exchange rates to calculate valuation effects.13

The findings in Lane and Shambaugh (2010a) reveal that there is considerable heterogeneity in the comovement between the trade-weighted and financial exchange rates and conclude that trade-weighted exchange rates are not the adequate measure to evaluate the wealth effects of currency movements. When we compute financial exchange rates for our sample of 50 countries between 1990 and 2017 and compare them with trade-weighted indices we confirm their results.14

Figure 1 shows examples of financial exchange rates vis-à-vis trade weighted exchange rates for a subsample of countries. The case of the US is interesting since the financial exchange rate index shows much less volatility than the trade-weighted index. As documented in Lane and Shambaugh (2010a) this is due to the fact that the foreign currency component of US external liabilities is low and that the trade index gives more weight to trading partners which are not necessarily major destinations of US investment (see also Tille, 2003).

Figure 1:
Figure 1:

Financial and Trade weighted exchange rates

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: This figure shows a comparison between financial and trade weighted exchange rate indices for the USA, Colombia and Germany. The Financial Exchange Rate Index is calculated using weights of US dollar, euro, Pound sterling, Japanese yen and domestic currency as described in this working paper. These weights are then applied to changes in bilateral exchange rates in these currencies for each country, keeping 1990 as base year.

Figure 1 also shows the financial and trade weighted exchange rate indices for Colombia. It is interesting to see that both indices show a high degree of comovement but the financial index exhibits substantially lower volatility since the 2000s, reflecting the increasing importance of domestic currency in total liabilities.

Finally, the Figure displays the comparison for Germany. In line with a typical case for an advanced economy, we observe that the financial exchange rate index exhibits lower volatility than the trade-weighted index, which indicates the presence of domestic currency liabilities and the fact that Germany’s investment partners (and the investment currencies) may not necessarily be aligned with its trading partners.

Table 1 shows the mean and median within-country correlations of the different indices. The asset and liability indices exhibit a high correlation but it is smaller for emerging and developing economies. We then calculate the correlation between each of these two individual indices and the trade-weighted index and observe that the correlations are weak. Finally, we compute the pairwise correlation between the financial and trade-weighted indices and show that the correlations are small, but slightly larger for the group of advanced economies. The volatility of the financial exchange rate is significantly lower than that of the trade-weighted index. However, the financial index for emerging and developing economies is more volatile than that of advanced economies.15

After computing the financial exchange rates we can calculate the valuation effects. As an example, in Figure 2 we plot our measure of valuation effects for Germany and compare it with a measure of revaluations of external net financial assets and liabilities due to exchange rate changes sourced from the Deutsche Bundesbank.16 Our measure tracks the valuation changes extremely well. The calculation of V ALXR allows us to see the valuation loss experienced in the aftermath of the global financial crisis and the subsequent recovery.

Figure 2:
Figure 2:

Valuation Effects: Comparison of VALXR for Germany

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: This figure compares our measure of valuation effects (VALXR) for Germany with the one sourced from the Deutsche Bundesbank.

The calculation of financial exchange rates and valuation effects are a useful tool to quantify the effects of exchange rate movements on the external balance sheet. For countries that have a negative net position in foreign currency, a depreciation of the domestic currency can generate significant negative wealth effects. This is the prototype example of a developing country. However, many developing countries have increased the share of foreign liabilities in domestic currencies mainly through an increase in equity and in some cases also through an increase in debt denominated in domestic currency (see IMF External Sector Report, 2019). These shifts can have substantial implications in the wealth effects from currency movements. The formula for valuation effects in Equation 1 can be applied to any country of our sample in a straightforward way.

3.2 International Financial Integration

Another building block to compute valuation effects is the IFI measure, which does not contain a currency dimension, but once multiplied by the change in the financial exchange rate index we obtain the valuation effect. However, our dataset allows us to break down the IFI measure by currency and analyze the main trends.

The literature pioneered by Lane and Milesi-Ferretti (2003) and Lane and Milesi-Ferretti (2007) has analyzed the evolution of de-facto IFI based on cross-border assets and liabilities positions relative to GDP. The authors have documented the large increase in international financial integration from the 1990s to the global financial crisis and noted that since the 1990s the pace of integration has been more gradual for emerging markets than for advanced economies. More recently, Lane and Milesi-Ferretti (2018) highlight that the growth in international financial integration came to a halt in the aftermath of the global financial crisis. This trend is mainly attributed to the decline in cross-border activity by banks in advanced economies (McQuade and Schmitz, 2017). While the IFI measure focuses on the size of the international balance sheet and its decomposition into financial instruments, little is known about the currency composition of the stocks of external assets and liabilities.

Our dataset allows us to look into this dimension and analyze how the trends in international financial integration are reflected in the currency breakdown. Of particular interest is to assess the role of the US dollar relative to the euro since these are the two dominant currencies in international finance and trade.

Figure 3 reports the measure of IFI denominated in US dollars (in blue) and in euros (in red) for all the countries in our sample. In addition, we include a measure of IFI in US dollars (dotted blue) and euros (dotted red) for a sample of all countries excluding the US and the Euro Area. Figures 4 and 5 show the same information for advanced and emerging economies, respectively. In all cases we include the sum of external assets and external liabilities scaled by the weighted average of each country’s GDP.17

Figure 3:
Figure 3:

International Financial Integration: All Countries

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: This figure shows the measure of international financial integration for all countries.
Figure 4:
Figure 4:

International Financial Integration: Advanced Economies

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: This figure shows the measure of international financial integration for advanced economies.
Figure 5:
Figure 5:

International Financial Integration: Emerging and Developing Economies

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: This figure shows the measure of international financial integration for emerging and developing economies.

From the figures we observe some interesting patterns. The overall trends described in Lane and Milesi-Ferretti (2003), Lane and Milesi-Ferretti (2007) and Lane and Milesi-Ferretti (2018) are clearly visible in our dataset, demonstrating the large role the countries in our sample play in global finance. In particular, the IFI measure shown in Figure 3 has increased two-folds from the early 1990s levels to 360 percent of GDP by 2017 and exhibits a decline in the aftermath of the global financial crisis. For reference, as of 2017, around seventy percent of the cross-border holdings are denominated in US dollars or euros.

In the sample including all countries we observe that euro-denominated cross-border holdings expanded rapidly between 1999 and 2007, but declined considerably after the global financial crisis. US dollar positions on the other hand dropped in 2008, but quickly recovered showing an upward trend and gaining a clear predominance over the euro since 2014. Note that this pattern is not only evident for the US or Euro Area countries. In fact, when we exclude the US and the Euro Area countries (dotted lines) the shift away from the euro and into the dollar is still present.18 However, in this case we observe an unambiguous dominance of cross-border holdings in US dollars throughout the whole sample period. The difference between the IFI measure (dotted lines) in dollars (in blue) and in euros (in red) has widened since 2007 and US dollar cross-border positions were three times larger in 2017.

Several factors played a role in driving this pattern. On the one hand, Euro Area banks have persistently deleveraged from cross-border positions since the crisis.19 Moreover, as documented in Maggiori, Neiman and Schreger (2019b), the uncertainty triggered by the Euro Area sovereign debt crisis led investors to shift away from euro positions. In addition, the US dollar appreciation and the high liquidity of dollar assets during the peak of the crisis has reinforced the dominant role of the US dollar, in particular as there is a lack of supply of safe euro-denominated assets, relative to the US dollar (Ilzetzki et al., 2019).

Figure 4 shows that the IFI measure in US dollars for the sample of advanced economies (solid blue line) is very similar to the one of advanced economies excluding the US and the Euro Area (dotted blue line). The overall trend of increased financial integration and subsequent decline in the aftermath of the global financial crisis described in Lane and Milesi-Ferretti (2018) is also very clear from this graph. As in the sample which includes all countries, Figure 4 shows that this pattern is accompanied by a substantial decline in positions in euros, especially in the sample which excludes the US and Euro Area, in line with Maggiori et al. (2019a) and Maggiori et al. (2019b).

Further analysis, focusing on the sample of all advanced and emerging economies excluding the US and the Euro Area, reveals that both external assets and liabilities denominated in euros declined over the past decade, although the reduction in assets contributed slightly more to the decline. For the full sample, US dollar denominated liabilities are still slightly below their peak, but in the sample of advanced economies both assets and liabilities exceed previous peaks. In addition, the decline in euro denominated IFI is largely driven by the debt component with equity showing an upward trend for advanced economies. By contrast, we observe an increase in both equity and debt cross-border holdings denominated in US dollars in the aftermath of the global financial crisis.20

Emerging markets certainly account for a small but growing share of cross-border holdings (Figure 5). For this group of countries, the euro plays a muted role since the majority of cross-border holdings of assets and liabilities are denominated in US dollars. Figure 5 shows the disproportionate role of the dollar in cross-border holdings for emerging economies.

The finding that the dollar dominates the holdings of cross-border positions is linked to the role of the dollar in trade invoicing as documented in Goldberg and Tille (2008), and in trade and finance as described in Gopinath and Stein (2018).

3.3 Foreign Currency Exposures

The core of the analysis in the rest of the paper is centered around a measure of foreign currency exposure as a crucial building block. As common in the literature on foreign-currency exposures, we use an indicator which is intuitively defined as the net foreign assets denominated in foreign currency as a share of total assets and liabilities. Following Lane and Shambaugh (2010a) we calculate the aggregate foreign currency exposure indicator as

FXi,tAGG=ωAi,tFsi,tAωLi,tFsi,tL,(4)

where si,tAandsi,tL are defined in equation (3) and their relative size captures whether the country exhibits a creditor or debtor external position.21 ωAi,tFandωLi,tF are the proportions of asset and liabilities denominated in foreign currency, respectively.

This indicator ranges between -1 (case where all liabilities are denominated in foreign currency and all assets are in domestic currency) and 1 (case where all assets are denominated in foreign currency and all liabilities in domestic currency). A country is “long on foreign currency” if FXAGG is positive and “short on foreign currency” if it is negative. More generally, this indicator captures the sensitivity of a country’s external position to a uniform appreciation or depreciation of its currency vis-à-vis all other currencies.22 Note that if we were interested in calculating the impact of a uniform change in the value of the domestic currency against all foreign currencies, the uniform valuation effect (VALXR,U ) would be given by

VALi,t+1XR,U=FXi,tAGGIFIi,t%Ei,t+1U,(5)

where %ΔEu denotes a uniform shift in the value of the domestic currency against all foreign currencies. Our interest in FXAGG extends beyond its use for the calculation of valuation effects. In the remainder of the paper we build our analysis around this indicator and its main subcomponents. Due to international data limitations our dataset does neither capture the currency denomination of cross-border financial derivatives nor the extent to which various types of financial derivatives are used to hedge cross-border currency mismatches. However, Bénétrix et al. (2015) point out that this is not a major drawback to estimates of foreign currency exposures due to several factors. First, hedging activities tend to be mainly concentrated in advanced economies and on the liability side. As advanced economies tend to show long foreign currency exposures, this implies that with the inclusion of hedging activities, these positions would even be longer in foreign currency (i.e. not hedged away). Moreover, for most countries net financial derivatives positions tend to be a minor component of the external balance sheet (less than 1 percent of GDP), which mechanically limits their potential to provide large valuation gains or losses.

While the focus of our approach is on cross-border positions, one implication of financial globalization is that foreign currencies may also be used in “local” positions (i.e. between residents of the same economy). Data on the currency denomination of these positions is very sparse, but available for the banking sector of a number of countries in the BIS Locational Banking Statistics. For euro-US dollar positions, Bénétrix and Schmitz (2019) find suggestive evidence that cross-border currency exposures and local currency exposures may partly serve as hedges for each other. However, as this type of hedging is only partial in nature and confined to the banking sectors of a few advanced economies, it has only a very limited impact on the country-level exposures reported in our dataset.

4 Components and Long-Term Dynamics of FXAGG

This section describes the different components of aggregate foreign currency exposures, presents some descriptive statistics, a snapshot of the long-term dynamics and a regression analysis to identify the extent to which these are associated with relevant macroeconomic variables.

4.1 Components

Given that FXAGG accounts for the currency exposure on the net aggregate IIP, all asset classes are incorporated in this measure on both sides of the balance sheet. This implies that long and short foreign currency positions are affected by the currency composition of the different assets classes as well as the overall size and sign of the net international investment position, regardless of its underlying currency of denomination. More precisely, if a country has 60% of both its assets and liabilities denominated in foreign currency (ωAi,tF=ωLi,tF=0.6), the value of FXAGG will depend on the net IIP only. If the country is a net debtor, say (A – L)/(A + L) = -0.6, FXAGG will be negative and equal to -0.36. The only case where the currency of denomination will be the sole determinant of FXAGG is when a country’s IIP is zero (i.e. balanced), a very unlikely scenario.

In order to understand the relative role of its different components, we break FXAGG down into two components:

FXAGG=(AL)(A+L)+FXoAGG,(6)

where the first component the net foreign asset position (A – L) scaled by the size of the external balance sheet (A + L), and FXoAGG measures the net liability position in domestic currency (DC) (LDC – ADC) as a proportion of the external balance sheet. A positive value of this indicator implies that the proportion of liabilities denominated in domestic currency is greater than the proportion of assets denominated in domestic currency in relation to (A+L). FXoAGG can be interpreted as the aggregate foreign currency exposure evaluated at a zero net foreign asset position. If A – L = 0, a positive value of FXoAGG would imply that the proportion of assets denominated in foreign currency is larger than the proportion of liabilities denominated in foreign currency in relation to (A+L). From here onward, we refer to FXAGG as the foreign currency exposure and to FXoAGG as the foreign currency mix.

Table 2 reports the summary statistics of this decomposition and its subcomponents for the full sample of 50 countries as well as for the advanced and emerging economy groups in 2017 (the final year of the dataset).

The mean and median in all country groupings show a long foreign currency position, FXAGG > 0, ranging from 0.11 to 0.18. This is also observed in the interquartile range of the full sample and in the advanced group. However, some countries show a short position in the emerging economies group. Note that a negative value of FXAGG implies balance sheet losses in case of a depreciation and gains in case of an appreciation.

Considering the two main components of the exposures, Table 2 reveals that there are two counterbalancing forces in most of the cases. On the one hand, many countries exhibit a debtor position with (A-L)/(A+L) ranging from -0.13 to -0.2 across mean and median countries. On the other hand, all statistics show a long position for the foreign currency mix, FXoAGG>0. More precisely, long foreign currency exposures in 2017 were mostly the result of the currency mix of assets and liabilities rather than net creditor positions. In terms of the dispersion, the interquartile range shows more cross-country variation in (A – L)/(A + L) than in FXoAGG.

In order to have a deeper understanding of these two items, we break them down into their relevant subcomponents. First, we decompose (A – L)/(A + L) into the net foreign asset position excluding foreign exchange reserves and the reserves component as follows:

(AL)(A+L)=(ANRL)(A+L)+FXR(A+L).(7)

ANR are foreign assets excluding reserves and FXR denotes the foreign exchange reserves. As expected, the net position is dominated by the non-reserve component. In addition, Table 2 indicates that it is negative for most countries, in particular for emerging economies. While advanced economies mean value of (ANRL)(A+L) is -0.04, it is -0.33 for emerging countries. By definition, foreign currency reserves are always positive, but we observe that they represent a larger share of external assets in emerging countries. In fact, foreign reserves in these countries have been assessed to exceed the IMF’s reserve adequacy guidelines (Alfaro and Kanczuk, 2019).

Second, we decompose FXoAGG into the following terms:

FXoAGG=PEQLDC+FDILDC(A+L)+DEBTLDC(A+L)ANRDC(A+L),(8)

where PEQLDC denotes portfolio equity liabilities, which are denominated in domestic currency, and FDILDC is the equity component of direct investment liabilities denominated in domestic currency. DEBTLDC are debt liabilities denominated in domestic currency and ANRDC are non-reserve assets denominated in domestic currency.

Note that for advanced economies FXoAGG is mainly driven by debt liabilities in domestic currency followed by equity liabilities in domestic currency, including both FDI equity and portfolio equity. By contrast, the key driver of FXoAGG for emerging economies is the equity component of liabilities. Compared to advanced countries, the debt component in domestic currency is small but still positive. It has been documented that during the last decade there has been a remarkable change in emerging market government finance as governments in these economies have increasingly borrowed in their own currency (Alfaro and Kanczuk, 2019). While our analysis does not break debt down between public and private sectors, we confirm this trend when we analyze the evolution of portfolio debt liabilities. However, as external debt liabilities also include FDI debt and other investment, which in emerging markets continue to be dominated by foreign currencies, the domestic currency component in debt instruments remains relatively small.

Table 3 reports the mean, median and interquartile range of the correlation coefficient across all these items for the period 1990–2017. For FXAGG, the decomposition shows that (A-L)/(A+L) and FXoAGG are positively correlated with FXAGG. The correlations suggest that FXAGG is more responsive to changes in (A-L)/(A+L) than in FXoAGG. The relation between these differs depending on the summary statistic of the correlation distribution that we analyze. On the one hand, the mean and median correlations are very small with negative and positive signs, respectively. On the other hand, the interquartile range is wide and includes large negative and positive correlations.

Table 3:

FXAGG decomposition, correlations 1990–2017

article image
Notes: This table presents the mean, median, 25th and 75th percentile of the cross-country correlation coefficient between FXAGG and its sub-components. Correlation coefficients are computed based on the full time span in our data set: 1990–2017.

As expected, there is a strong positive correlation between (A – L)/(A + L) and (ANR -L)/(A+L). By contrast, the correlation between (A-L)/(A+L) and FXR/(A+L) is positive at the mean and median but negative correlations emerge in the interquartile range indicating that (A – L)/(A + L) and FXR/(A + L) move in opposite directions for some countries.

When the focus is on the foreign currency mix, FXoAGG, Table 3 shows that the equity liability components, (PEQL+FDIL)/(A+L), are highly correlated with FXoAGG. High mean (0.76) and median (0.97) correlations highlight the relevance of changes in equity liabilities for overall exposures. Conceptually, a positive relation is also expected for debt liabilities in domestic currency, DEBTLDC/(A + L), and FXoAGG. Although it is positive, this relation is not strong at the mean or median of the correlation distribution. The interquartile range is wide with correlations from -0.25 at the 25th percentile to 0.74 at the 75th percentile. By construction, the correlation between non-reserve assets denominated in domestic currency, ANRDC/(A+L), and the foreign currency mix, FXoAGG, should be negative. For the mean and median correlations we find that this is the case, although the correlations are small in absolute value.

For completeness, we also present correlation matrices for the advanced and the emerging country groups. As in the full sample, (A – L)/(A + L) and FXoAGG are positively correlated with FXAGG in both groups. Interestingly, advanced countries show a negative link between (A-L)/(A+L) and FXoAGG indicating that debtor countries are able to “hedge” their overall negative net positions with a long foreign currency mix. In these cases, exchange rate movements will improve the net external position of the country. However, this is not the case for emerging countries as they show a positive correlation between these items for at least half of the countries in the group. For them, the exposure associated with a net debtor position is exacerbated by a short foreign currency mix. This means that a depreciation of the exchange rate would act as a destabilizing force increasing a net debtor position further. A reason for this result could be that emerging countries with negative net external positions are deemed as inherently more risky, which in turn reduces their ability to issue liabilities in domestic currency. Other differences between these country groups include the role of foreign exchange reserves which, as expected, are more strongly associated with (A – L)/(A + L) and FXoAGG in emerging countries.

In sum, the mean and median countries show long foreign currency positions in 2017, driven by a positive foreign currency mix FXoAGG. In advanced economies this is mainly due to debt liabilities, while for emerging economies the key driver is the equity liabilities component. Moreover, we uncover a very interesting stylised fact. While net international investment positions and long foreign currency mix exhibit a negative correlation in advanced economies, indicating currency hedging of the liability positions, emerging economies show a positive correlation. Even if there has been a movement toward a long foreign currency exposure via long foreign currency mixes, the way in which these two subcomponents interact suggests that exchange rate movements are more likely to worsen than to improve external positions via valuation effects in emerging countries.

4.2 Long-term dynamics

Figure 6 presents the cumulative cross-country distribution of FXAGG positions at the start and end of our sample period for advanced (in red) and emerging economies (in blue). Consistent with the evidence provided in Lane and Shambaugh (2010a) and Bénétrix et al. (2015), the curve has shifted significantly towards long positions in foreign currency since 1990 in the context of a surge in global financial flows. While 60 percent of countries in our sample exhibited net negative positions in foreign currencies in 1990, this proportion declined to 20 percent in 2017.23

Figure 6:
Figure 6:

FXAGG long-term dynamics: 1990 vs 2017

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: Cumulative distribution for net aggregate foreign-currency exposures defined as FXi,tAGG=ωAi,tFsi,tAωLi,tFsi,tL,wheresi,tA(si,tL) is the share of total assets (liabilities) in the sum of total assets plus liabilities. ωAi,tFandωLi,tF are the proportions of total assets and liabilities denominated in foreign currency, respectively. FXi,tAGG is measured on the horizontal axis and ranges between -1 and 1. The vertical axis measures the cumulative distribution, or the proportion of countries, below each FXAGG value in the horizontal axis. We include 50 countries. For Russia and Czech Republic, we use FXi,tAGG for 1993.

Net negative positions in foreign currencies continue to be dominated by emerging market economies, which have a history of borrowing heavily in foreign currency, a phenomenon referred to as “original sin” (Eichengreen, Hausman and Panizza, 2003), raising questions about their vulnerability to external shocks, particularly those associated with large currency movements. Our analysis reveals that there has been an improvement in their net position, partly driven by a change in the currency composition of foreign liabilities away from foreign currency and toward local currency instruments as well as a sustained accumulation of foreign currency assets. While emerging economies accounted for 22 out of the 29 countries with short foreign currency positions in 1990, they represent 9 of the 10 countries short in foreign currency in 2017. In 2017 the only advanced economy with negative positions is Greece.

In what follows we will decompose the different elements of FXAGG which will help us understand the role played by the different subcomponents in driving the movement in the curve.

4.2.1 Decomposition

Figure 7 displays the cumulative distribution of the components of FXAGG as described in Equation 6. It shows that the remarkable shift towards long foreign currency positions in the distribution of FXAGG between 1990 and 2017 was mainly driven by the improvement in the net international investment positions, (A – L)/(A + L). In fact, 35 percent of the countries in the sample were net creditors in 2017, compared to 20 percent in 1990. Moreover, the size of net debtor positions (scaled by total external assets and liabilities) shrank substantially as visible in the downward shift of the distribution in the chart on the bottom left. In addition, the foreign currency mix, FXoAGG, moved towards longer positions since 1990, thereby also contributing to the rightward shift in FXAGG.

Figure 7:
Figure 7:

Long-term dynamics: FXAGG decomposition

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: FXAGG=(AL)(A+L)+FXoAGG as defined in main text.

Figure 8 shows the cumulative distribution of the components of (A-L)/(A+L) following Equation 7, which breaks down the net international investment position into a term that includes foreign assets without reserves, (ANR – L)/(A + L), and foreign exchange reserves component, FXR/(A + L). The Figure reveals that the improvement in net foreign assets (scaled by A + L) since 1990 was overwhelmingly driven by the non-reserve components of the IIP. Note that emerging economies experienced an expansion in foreign reserves since the mid-1990s, partly driven by an increase in financial development, as described in Obstfeld, Shambaugh and Taylor (2010). Interestingly, the accumulation of reserves by these countries has been so large that it is hard to reconcile from the lens of theoretical models which feature insurance against a “sudden stop” (Jeanne, 2007). However, Figure 8 shows that this item played a limited role in the overall improvement of the net IIP.

Figure 8:
Figure 8:

Long-term dynamics: (A — L)/(A + L) decomposition

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: (AL)(A+L)=(ANRL)(A+L)+FXR(A+L) as defined in main text.

Finally, Figure 9, displays the cumulative distribution of the components of FXoAGG, described in equation 8. In this Figure, FXoAGG is broken down into domestic currency (portfolio and FDI) equity liabilities, domestic currency debt liabilities and external assets denominated in domestic currency (all scaled by A + L). While the distribution of the asset component in domestic currency has not changed much for most of the sample since 1990, it increased quite substantially for the Euro Area countries. The adoption of the euro meant for these countries that a big part of their cross-border assets were “redenominated” from foreign to domestic currency. Therefore, the overall right-ward shift (i.e. increase) was mainly driven, in a rather uniform way, by larger equity liabilities denominated in domestic currency. The importance of this item is even more marked for emerging economies. This is in stark contrast with the distribution of debt liabilities in domestic currency which looks very similar in 1990 and 2017 for around 80 percent of the countries in the sample.

Figure 9:
Figure 9:

Long-term dynamics: FXoAGG decomposition

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: FXoAGG=PEQL+FDIL(A+L)+DEBTLDC(A+L)ANRDC(A+L) as defined in main text.

We find that the improvement in the foreign currency exposure in emerging economies is dominated by an increase in equity liabilities in domestic currency and to a lesser extent an increase in portfolio debt liabilities in domestic currency. This result is related to the findings in Haussman and Panizza (2011), who suggest that the reduction in “original sin” for emerging economies is small and very specific to a limited number of countries. When we look at overall debt, the predominance of domestic currency borrowing is more tenuous given that external debt is defined as the sum of FDI debt, other investment and portfolio debt with the first two being predominantly in dominant currencies. By contrast, some advanced economies, notably the Euro Area countries, experienced a strong expansion in debt liabilities denominated in domestic currency. The adoption of the euro allowed Euro Area countries to issue more debt instruments in domestic currency.

Overall, the decomposition analysis reveals that the shift towards foreign currency positions since 1990 was mainly driven by improving net international investment positions (excluding reserves) and to a lesser extent also due to larger net liability positions in domestic currency (driven by larger equity liabilities).

4.3 Regression Analysis

In this section we assess the extent to which foreign currency exposures are associated with a set of relevant macroeconomic variables. To this end, we present a series of parsimonious regressions models following the approach in Lane and Shambaugh (2010b). Our goal is to study the importance of trade and financial openness, macroeconomic risk, exchange rate regimes, country size and development on aggregate foreign currency exposure and its subcomponents.

Table 4 shows pooled regression models including three years of data: 1997, 2007 and 2017. The set of core covariates includes trade openness (calculated as the sum of exports and imports scaled by GDP and denoted by Trade), macroeconomic risk indicators captured by the volatility of GDP (vol(GDP)), volatility of inflation (vol(π)) and financial exchange rates (computed as shown in equation (2) and denoted by I).24

Table 4:

FXAGG determinants. All countries

article image
Notes: Robust standard errors in parentheses. ***, **, and * denote, respectively p < 0.01, p < 0.05, and p < 0.1.

Foreign currency exposures can have stabilizing or destabilizing effects as a result of exchange rate movements. Countries long in foreign currency will exhibit a net capital gain on their external financial position as a result of a uniform depreciation of their currency. If this takes place in a recession context, that capital gain will act as a buffer or a stabilizing force. On the contrary, if the country is short foreign currency, the effects of a uniform depreciation in recessions will be exacerbated by a capital loss in the external position. The latter was a common feature in emerging market economies unable to issue debt in domestic currency. To proxy these stabilizing/destabilizing effects, we include the correlation coefficient between GDP and the financial exchange rate index. When this correlation is positive, recessions are associated with an exchange rate depreciation and vice-versa. Thus, a positive coefficient for corr(GDP,I) in the FXAGG regression model can be interpreted as evidence that the currency exposure in the external position is acting as a stabilizing force.25

In addition to the above explanatory variables, we include indicators for the degree of financial account openness, exchange rate regime classification, EMU membership, as well as country size and development. We use the updated Chinn and Ito (2006) de jure measure of financial openness (KA open) while the exchange rate regime classification is obtained from Shambaugh (2004) (the variable Peg takes value of one if the country is classified as having a fixed exchange rate). EMU membership is a dummy variable for the sample of countries in our data that belong to the European Monetary Union: it takes a value of zero for all countries in 1997 and one for member countries in 2007 and 2017. Country size and level of development are proxied by the logarithm of population (denoted by Size) and GDP per capita (GDPpc), respectively. Finally, we control for long-term movements in currency exposure distributions by including two dummy variables for the years 2007 and 2017.26

As in the previous section, we break FXAGG down into its subcomponents to study how the above variables affect each of them. This is important to uncover counterbalancing relations at the subcomponent level. We report the regression estimates for (A – L)/(A + L) and FXoAGG in columns (2) and (3), respectively. Then, we break the former down into (ANR – L)/(A+L) and FXR/(A + L). Regression results for this split are reported in columns (4) and (5), respectively. FXoAGG is divided into (PEQLDC + FDILDC)/(A + L), DEBTLDC/(A + L) and ANRDC/(A+L) with regression outputs shown in columns (6), (7) and (8), respectively.

Table 4 shows that our model is able to explain more than 50 percent of the cross-country variation in aggregate foreign-currency exposures. However, its explanatory power varies considerably across subcomponents, with R2’s ranging from 0.40 to 0.90.

In line with the work of Lane and Shambaugh (2010b), we find that trade openness is positively associated with long positions in foreign currency.27 This is driven by a positive effect in the net external positions, (A – L)/(A + L), instead of the foreign currency mix, FXoAGG. The former is, in turn, unrelated to trade openness through foreign exchange reserves, as shown in column (5). The absence of a significant link between trade and FXoAGG relates to its balancing effect on debt liabilities and non-reserve assets denominated in domestic currency, shown in columns (7) and (8). The link with debt liabilities in domestic currency is negative, meaning that countries which are more open are less likely to issue debt in domestic currency, possibly to hedge trade positions. On the other hand, countries that trade more have larger non-reserve assets in foreign currency. This positive association can be reconciled with the idea that the home bias in equity portfolio assets is smaller (that is, agents hold more foreign denominated portfolio assets) the higher the share of imports in domestic consumption is (Obstfeld and Rogoff, 2011). The empirical specification does not uncover a relation between trade openness and equity liabilities in column (6).

In relation to the country risk proxies, GDP volatility is positively associated with a long position in foreign currency as shown in column (1). This relation presumably indicates that a higher volatility in domestic wealth leads to the use of the balance sheet to hedge against risk. We find that this relation is driven by longer foreign currency positions in non-reserve assets (column (8)). The second proxy of country risk, given by inflation volatility, does not have a significant effect on aggregate exposure and the main subcomponents. Our model suggests that there is a small positive link between inflation volatility and the size of foreign exchange reserves, implying that countries that face higher inflation volatility tend to accumulate larger foreign exchange reserve assets. By contrast, exchange rate volatility has a significant effect on the aggregate foreign currency exposure in our sample. Higher exchange rate volatility is associated with short aggregate positions in foreign currency. Unfortunately, our model is not able to show clearly whether the net position, (A – L)/(A + L), or the foreign currency mix, FXoAGG, dominate this relation. However, the model uncovers a negative link between exchange rate volatility and both domestic currency denominated debt liabilities and non-reserve assets, in columns (7) and (8).

In terms of the potential buffer/amplifier effects of currency exposures via valuation effects, our model shows that the correlation between GDP and the financial exchange rate is positively associated with a long aggregate position in foreign currency. This means that countries in recession with depreciating currencies would also exhibit a capital gain on the net external position that would act as a buffer. This effect is driven by the response of the net position (A – L)/(A + L) instead of the foreign currency mix FXoAGG. The former channel is mostly explained by the effect on the net position excluding reserves (column (4)).

When we focus on the additional controls, our model suggests that the de jure measure of financial openness is not associated with the aggregate foreign currency exposure due to potentially counterbalancing effects on (A – L)/(A + L) and FXoAGG. The KA open indicator is negatively and statistically related with the latter only. In terms of the additional subcomponents, our model yields mixed results. Financial openness is negatively associated with reserve assets and equity liabilities (columns (5) and (6)). By contrast, the link with debt liabilities and non-reserve assets in domestic currency is positive (columns (7) and (8)). In turn, the exchange rate regime is not statistically significant in the FXAGG and main subcomponents regressions (columns (1)-(3)). However, our model shows a positive link between Peg and reserves, and a negative link between Peg and equity liabilities in domestic currency. Interestingly, EMU membership plays a strong role for aggregate currency exposures (column (1)). Our model shows that it is associated with short aggregate foreign-currency positions, driven by negative links with (A-L)/(A+L) and FXoAGG. The former effect is driven by lower reserve assets of Euro Area countries (column (5)). For the latter, the model shows statistically significant links with all subcomponents (columns (6)-(8)). The effect on equity liabilities in domestic currency is smaller than for debt liabilities. Importantly, the overall impact is driven by larger external assets denominated in domestic currency (i.e. the euro), thereby leading to lower net liabilities in domestic currency, which highlights the relevance of EMU membership on the currency of cross-border positions.

In relation to the country size and development controls, we find that these are positively associated with long aggregate foreign currency exposures through their link with net positions (A – L)/(A + L) and not with the foreign currency mix. These variables are also positively associated with the non-reserve net position. Only GDP per capita is negatively related with the stock of foreign exchange reserves. Larger and richer countries hold more debt liabilities and non-reserve assets denominated in domestic currency.

Finally, our year dummies shed light on the impact of time conditioning on the factors described before. In line with previous evidence in Lane and Shambaugh (2010a), Lane and Shambaugh (2010b) and Bénétrix et al. (2015) these show a rightward shift in the aggregate foreign currency exposures distribution. Here, we confirm this movement toward a longer foreign currency position unconditionally, as in the previous studies, and conditional on our set of covariates. Interestingly, the conditional movement in aggregate currency exposures is explained by a foreign currency mix effect and not by the net foreign asset position for the full sample including advanced and emerging countries. This is shown in column (3). Moreover, 2007 is associated with a reduction in the non-reserve net position while 2017 with an increase in foreign exchange reserves. Our model also uncovers a positive association between the year dummies and equity liabilities in domestic currency, and a negative link between the year dummies and external assets denominated in domestic currency.

4.3.1 Country Group Comparison: Advanced vs. Emerging Economies

While looking at the full country sample is a reasonable first step to study the relationship between a number of key variables and currency exposures, it is also relevant to study how these links may change when we focus on advanced and emerging countries separately. One natural reason to do so is that the former includes the issuers of the dominant foreign currencies in which the majority of external positions are denominated. More generally, there may be more fundamental differences between these groups such as those related to institutions, legal systems, economic policy, history, not captured by our previous approach.

With this in mind, we report the regressions for the two country groups in Tables 5 and 6. The results show that the positive effect between trade openness and long foreign currency exposures reported before is dominated by advanced economies since it is not significant for emerging markets (column (1)).

Table 5:

FXAGG determinants. Advanced

article image
Notes: Robust standard errors in parentheses. ***, **, and * denote, respectively p < 0.01, p < 0.05, and p < 0.1.
Table 6:

FXAGG determinants. Emerging countries

article image
Notes: Robust standard errors in parentheses. ***, **, and * denote, respectively p < 0.01, p < 0.05, and p < 0.1.

Unfortunately, the country group strategy does not reveal which one dominates the relation between GDP volatility and foreign currency exposures. The point estimates in column (1) of Tables 5 and 6 are both statistically insignificant. The sign is negative in advanced and positive in emerging economies, as in the full sample. When we look at the subcomponents for the advanced economies group, GDP volatility yields a negative and statistically significant coefficient for (A-L)/(A+L) and a positive one for FXoAGG. In turn, higher GDP volatility is associated with smaller net creditor positions (column (4)) and it yields no significant relation with reserves (column (5)). By contrast, GDP volatility is associated with more positive external positions in the emerging countries group (column (2)). Interestingly, GDP volatility yields a positive and statistically significant effect on non-reserve domestic assets denominated in domestic currency (column (8)).

Inflation volatility is associated with a long foreign currency mix, higher reserves and larger equity liabilities in the advanced economies group (columns (3), (5), and (6)). In the emerging economies group there is also a positive association between inflation volatility and reserves (column (5)).

When the focus is on exchange rate volatility the country split shows that this variable is only significant for the emerging markets group, suggesting that the effects for the full sample may be driven by this group. Our model indicates that the relation comes from the link between exchange rates and reserve assets (column (5)). In relation to the stabilizing role of foreign currency exposures, the split reveals a strong effect in emerging countries that is driven by the effect on the net position (column (2)). For advanced countries, this variable does not show a statistically significant relation.

The link between aggregate foreign currency exposures and country characteristics (Size and GDP per capita) is present in both country groups. GDP per capita is associated with a long foreign currency position in advanced economies driven by the effect on net positions (column (2)), which is dominated by the impact on (ANR – L)/(A + L) (column (4)). The effect on the foreign currency mix is explained by the negative association between GDP per capita and debt liabilities in domestic currency (column (7)). In the emerging countries group, the positive relation uncovered with FXAGG is driven by the net position and is also unrelated to reserve assets.

The conditional dynamics, captured by the 2007 and 2017 dummy variables yields no statistically significant relation for the aggregate and net positions in the advanced economies group. However, a movement toward a long foreign currency mix FXoAGG>0 is present for both groups in 2017 and only for emerging economies in 2007. For advanced economies, we find a negative link between the 2017 dummy and non-reserve net positions (column (4)). For emerging economies, the 2007 and 2017 dummy variables are positively associated with equity liabilities in domestic currency (column (6)).

We checked the sensitivity of our results to alternative models which encompassed the inclusion of country fixed effects, a different proxy for financial integration given by the IFI indicator, proxies for financial deepening such as stock market capitalization, credit and deposits, the inclusion of a financial center dummy following the classification of Lane and Milesi-Ferretti (2018), alternative measures of volatility based on growth rates and appreciation rates. Overall, the results remain robust to different specifications.28

Our regression analysis shows that trade openness (driven by advanced economies) and output volatility are associated with longer positions in foreign currency, while exchange rate volatility (due to emerging economies) and EMU membership are associated with shorter aggregate positions in foreign currency. We also find among emerging economies that countries in recessions with depreciating currencies tend to exhibit capital gains on the net external position, thereby providing a hedge against domestic output fluctuations.

5 Pre- and post-crisis adjustment of FXAGG

The goal of this section is to document the heterogeneity in the long-term adjustment of currency exposures before and after the global financial crisis. As the post-crisis period was marked by a broad-based and persistent fall in international capital flows (Milesi-Ferretti and Tille, 2011; McQuade and Schmitz, 2017), it is of high interest to analyze how currency exposures adjusted in such an environment. To this aim, we describe the main patterns of the data and a regression analysis.

5.1 Descriptive Evidence

We build on the stylized fact of a rightward shift in the distribution of FXAGG documented in the previous section. While Bénétrix et al. (2015) show that this shift was associated with a reduction of large initial short foreign-currency exposures in the run-up to the crisis, we analyze the adjustment in the post-crisis period in which a sharp deceleration was observed (Figure 10). While 48 percent of countries had negative FXAGG in 1997, only 14 percent showed negative values in 2007. The curve changed direction in a minor way in 2012 (last year of Bénétrix et al., 2015) and 2017 (last year in the current dataset). In fact, 24 percent of countries displayed negative values in 2012 and 20 percent in 2017. Therefore, with respect to 2012 the curve moved slightly to the right.

Figure 10:
Figure 10:

FXAGG long-term dynamics: 1997, 2007, 2012, 2017

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: Cumulative distribution for net aggregate foreign-currency exposures defined as FXi,tAGG=ωAi,tFsi,tAωLi,tFsi,tL,wheresi,tA(si,tL) is the share of total assets (liabilities) in the sum of total assets plus liabilities. ωAi,tFandωLi,tF are the proportions of total assets and liabilities denominated in foreign currency, respectively. FXi,tAGG is measured on the horizontal axis and ranges between -1 and 1. The vertical axis measures the cumulative distribution, or the proportion of countries, below each FXAGG value in the horizontal axis. We include 50 countries. For Russia and Czech Republic, we use FXi,tAGG for 1993.

This change in dynamics is also evident from the scatter plots presented in Figure 11. In the pre-crisis period, the relation between initial foreign-currency exposures and the subsequent change in the run-up to the crisis is negative for the full sample as well as for the advanced and emerging country groups individually. All countries with short positions in foreign currency in 1997 exhibited improvements in FXAGG between 1997 and 2007, with the exception of Sri Lanka. This pattern holds in particular for emerging market economies which generally started with substantially larger short positions, but also for advanced economies.

Figure 11:
Figure 11:

FXAGG dynamics pre- and post-crisis

Citation: IMF Working Papers 2019, 299; 10.5089/9781513522869.001.A001

Notes: Net aggregate foreign-currency exposures defined as FXi,tAGG=ωAi,tFsi,tAωLi,tFsi,tL,wheresi,tA(si,tL) is the share of total assets (liabilities) in the sum of total assets plus liabilities. ωAi,tFandωLi,tF are the proportions of total assets and liabilities denominated in foreign currency, respectively. FXi,tAGG on the horizontal axis refers to 1997 (top panel) and 2007 (bottom panel). The vertical axis shows the change in FXi,tAGG between 1997 and 2007 (top panel) and 2007 to 2017 (bottom panel).

For countries that exhibited long positions in 1997, the ensuing changes are rather heterogeneous. All emerging economies move towards longer positions, while advanced economies movements go in both directions. Importantly, all advanced economies with long positions in 1997 still exhibited long positions in 2007.

The second panel of Figure 11 provides novel evidence of a marked change in the relation between the initial currency exposure (i.e. at the eve of the crisis in 2007) and the subsequent adjustment until 2017. Among the advanced group, most economies move towards larger long positions, with only Greece ending up in a slightly negative position in 2017. However, as indicated by the flat trend line, there is no significant relation between the initial level and the subsequent change. Among emerging countries the relation between the initial exposure and the change in the following ten years turns slightly positive. This reflects two developments. First, most emerging economies that achieved a positive FXAGG position in 2007 moved further towards longer foreign currency exposures, with a notably large change recorded for the Philippines. Second, those countries with negative FXAGG in 2007 moved further into shorter positions. The latter group of countries, which includes Turkey, Sri Lanka, Pakistan and Tunisia, worsened their foreign currency exposures since the crisis. Notably also Egypt and Morocco recorded short foreign currency positions in 2017, despite starting with long exposures in 2007.29 The pre-crisis period was marked by strong adjustments in the direction of longer FXAGG positions, while the post-crisis period is characterized by persistence in currency exposures.

5.2 Regression Analysis

Building on this evidence, we further study the structure of these dynamics by looking into the various components in a regression framework. We use a variety of specifications that allow us to study these bi-variate relations, while controlling also for several conditioning factors.

Table 7 shows the coefficients for the initial value of FXAGG and subcomponents as explanatory variables in the regression models, with the dependent variable being the change of FXAGG (or the respective subcomponents) between 1997 and 2007 or between 2007 and 2017. For each of these items we follow three approaches.

Table 7:

Ten-year changes pre- and post-crisis vs. initial values. FXAGG and subcomponents.

article image
Notes: Robust standard errors in parentheses. ***, **, and * denote, respectively p < 0.01, p < 0.05, and p < 0.1.