Foreign Exchange Intervention: A Dataset of Public Data and Proxies
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Foreign exchange intervention (FXI) is a highly debated topic. Yet, comprehensive and comparable data on FXI is hard to find. This paper provides a new dataset of FXI covering a large number of countries over the period 2000-20 at monthly and quarterly frequencies. It includes publicly available data for about 40 countries and carefully constructed proxies for 122 countries. Proxies are focused on both spot and derivative transactions that alter the central bank’s foreign currency position and account for a wide range of central bank operations, including vis-à-vis residents, the first proxy to do so to our knowledge. The paper discusses the merits of the new proxy relative to coarser measures traditionally used like the change in reserves, and potential definitional differences with published data. The paper also presents stylized facts using our newly constructed FXI proxies.

Abstract

Foreign exchange intervention (FXI) is a highly debated topic. Yet, comprehensive and comparable data on FXI is hard to find. This paper provides a new dataset of FXI covering a large number of countries over the period 2000-20 at monthly and quarterly frequencies. It includes publicly available data for about 40 countries and carefully constructed proxies for 122 countries. Proxies are focused on both spot and derivative transactions that alter the central bank’s foreign currency position and account for a wide range of central bank operations, including vis-à-vis residents, the first proxy to do so to our knowledge. The paper discusses the merits of the new proxy relative to coarser measures traditionally used like the change in reserves, and potential definitional differences with published data. The paper also presents stylized facts using our newly constructed FXI proxies.

I. Introduction

Exchange rate policy and the use of foreign exchange intervention (FXI) are central aspects of policy making in open economies and highly debated topics. Yet, information on FXI remains scant for most countries. As a result, macroeconomic research on a wide range of aspects related to the use of FXI has often relied on coarse proxies—typically, changes in the stock of central bank’s reserves or reserve flows from balance-of-payment statistics1—including for analysis on exchange rate regimes (see, for example, Ilzetzky et al, 2017), currency manipulation (Bayoumi et at, 2015; Gagnon 2012 & 2013) and exchange rate management more broadly (Adler et al, 2019; Blanchard et al, 2015; Daude et al, 2016; Dominguez 2012 and 2020, etc.). These coarse proxies of FXI, however, are often polluted by valuation changes and investment income flows, as well as distorted by central bank foreign currency transactions vis-à-vis residents and nonresidents that affect the level of reserves but do not constitute foreign exchange intervention (i.e., exchange of local and foreign currency assets).

This paper provides a new dataset that compiles published and proxied FXI, for both operations in spot and derivative markets, covering 2000–20 and an ample set of countries. Specifically, the dataset includes:

  • 1) Published FXI data covering 39 countries at monthly and 43 countries at quarterly frequency.2

  • 2) Proxied FXI data for 122 countries at monthly and quarterly frequency.

To our knowledge, this is the most comprehensive dataset of FXI for macroeconomic analysis. Our proxies provide a more accurate measure than coarser measures traditionally used in macroeconomic analysis of transactions that alter the central bank’s foreign currency position. It does so by accounting for a wide range of central bank operations, including vis-à-vis residents and nonresident entities. This work complements the recent work by Fratzscher et al (2020), who focus on identifying periods when interventions took place through news searches for a smaller group of countries, although still relying on traditional coarse proxies of FXI to quantify the actual degree or magnitude of FXI.

Despite the growing transparency in FXI operations over time, the share of countries that publish such data remains low (Figure 1). Specifically, about 74 percent of (non-reserve currency issuing) central banks of advanced economies publish FXI data, while only 16 percent of those of emerging market and developing countries do it. Typically, countries that publish FXI data at least at quarterly frequency also do it on a monthly basis.

Figure 1.
Figure 1.

Coverage of publicly available FXI data

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: Various. See Appendix Table A1.1 for further details.Notes: Based on countries that report FXI data at least on a quarterly frequency. Statistics for AE and EMDE groups are based on quarterly FXI publication.

Moreover, a large number of countries that do not publish FXI data likely manage their exchange rates significantly, at least periodically, through foreign exchange interventions. This is apparent from Figure 2, which indicates that non-publishing countries display at least as much, if not more, variation in changes in reserves as countries that publish FXI data. Reliable proxies of foreign exchange intervention are, thus, key for any comprehensive cross-country analysis on exchange rate policy.

Figure 2.
Figure 2.

Change in Reserves and FXI Publication

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: IMF, International Finance Statistics; IMF, Information Notice Systems; International Reserves and Foreign Currency Liquidity Template; and IMF staff calculations.Notes: The figure reports the cumulative distribution of the country-specific standard deviation of changes in reserves, distinguishing between countries that reported FXI data for most of the sample period and those that did not. Based on countries that report FXI data at least on a quarterly frequency.

The rest of the paper is organized as follows: Section II explains in broad terms how our FXI proxy is constructed. Section III presents some stylized facts based on this proxy. Section IV describes our new dataset of publicly available FX interventions. Section V assesses the performance of our proxy relative to other proxies. Section VI concludes.

II. Constructing FXI Proxies

The literature on FXI is rich and long dated, yet the vast majority of the papers overlook defining FXI. We define FXI as ’any transaction that changes the central bank’s foreign-currency position’ This definition has 4 key elements:

  • i. A focus on active transactions, which means that changes in foreign currency positions that are passive, such as those arising from the interest income on reserve assets or from changes in asset prices are not considered interventions. Such variations in foreign currency positions are not driven by contemporaneous actions of the central bank.

  • ii. A focus on the central bank as the main entity conducting foreign exchange interventions. Thus, the definition excludes foreign currency operations by other public sector entities as they are normally not involved in exchange rate policy and for which data limitations are more severe. See further discussion on this in section V.D.

  • iii. A focus on the central bank’s foreign currency position, encompassing any exchange of foreign and domestic currency assets. This is consistent with the most relevant channel through which FXI can have macroeconomic effects—the portfolio balance channel.3 Whether FXI operations are sterilized or unsterilized—which is about the type of domestic currency asset (money or debt) supplied in exchange for foreign currency assets—is beyond the scope of this work.

  • iv. The definition encompasses all transactions that meet the above criteria, irrespective of the stated intent (for example, to accumulate reserves without meaning to affect the exchange rate), as the intent is not verifiable and these operations can still affect the exchange rate.

The proxies of FXI developed in this paper follow these 4 key elements and encompass both operations in spot markets (“spot” proxy) as well as operations with derivatives (“derivative” proxy) as both affect the central bank’s foreign currency position. A “broad” proxy that includes both the spot and derivative transactions is also reported.

Spot interventions

Most of the literature on the macroeconomic implications of FXI has focused on coarse proxies, primarily based on the change in the stock of international reserves.4 Such measures, however, can deviate from the concept of FXI defined above in meaningful ways. In particular:

  • Valuation changes. Changes in the stock of reserves may reflect FXI but also valuation changes due to varying market prices of reserve assets or exchange rate movements among reserve currencies. These variations affect the central bank’s foreign currency position but are not a product of active transactions. As shown in the left panel of Figure 3 below, valuations changes can be a significant component of changes in stocks of reserves.

  • Investment income. Similarly, changes in the stock of reserves partly reflect the return on investments. While this component is slow-moving and smaller on average than valuation changes (Figure [3], right panel), it also needs to be stripped out from changes in reserves to capture more accurately actual foreign exchange transactions. This adjustment is particularly important when assessing whether FXI is one-sided or two-sided, as income flows are positive by definition and thus always contribute to increasing reserves. Such income flows can be particularly large in economies with large reserve holdings.

  • Other foreign currency assets and liabilities vis-à-vis nonresidents. Changes in reserves can also reflect reallocation of foreign assets, between reserve and non-reserve assets. This is particularly relevant for economies that have accumulated significant reserve assets and where reserve holdings may exceed levels consistent with precautionary liquidity buffers. In these economies, central banks may choose to invest part of their foreign currency holdings in less liquid (non-reserve) assets. Thus, portfolio management operations affect the level of reserves without altering the central bank’s foreign currency position. Similarly, reserve levels may vary as central banks borrow or repay loans from foreign entities (in foreign currency), without entailing interventions in the foreign exchange market or changes in the overall foreign currency position. The latter typically include loans from the International Monetary Fund and other official creditors.

  • Foreign currency assets and liabilities vis-à-vis residents. Changes in reserves (or foreign assets more broadly) may also reflect foreign currency operations vis-à-vis residents but do not necessarily affect the overall foreign currency position of the central bank. For example, operations on behalf of other domestic entities (e.g., the general government) entail changes in foreign assets but with an offsetting change in the liabilities vis-à-vis the domestic counterpart. Similarly, government foreign currency deposits (and withdrawals) at the central bank imply variations in reserves without changes in the foreign currency position. Changes in foreign-currency deposits of domestic (banking) institutions at the central bank also alter the latter’s gross foreign-currency assets without affecting its net foreign-currency position.

Figure 3.
Figure 3.

Changes in reserves, valuation changes and investment income

(in percent of previous quarter stock of reserves, simple average)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: IMF, Balance of Payment; IMF, International Finance Statistics; and IMF staff calculations.Note: Simple averages calculated across the country sample. Valuation changes calculated by subtracting quarterly balance-of-payments reserves flow from quarterly change in the stock of reserves. Reserve income flow is taken directly from balance-of-payments statistics for countries reporting it. Each chart uses only countries whose data are available for both variables, hence the difference in the variable “change in reserves” across the two charts.

Our “spot” proxy adjusts for all above-mentioned factors. In broad terms:5

  • The proxy uses available information on the composition of reserve assets to estimate valuation changes. Specifically, it relies on information on the type of assets from the International Reserves and Foreign Currency Liquidity Template6; and on the currency composition of individual countries from the International Reserves and Foreign Currency Liquidity Template, when available, or country-group aggregates from the Currency Composition of Official Foreign Exchange Reserves (COFER) dataset.7 Estimates of valuation changes are necessary for the FXI proxy at monthly frequency as the starting point for this proxy is the change in the stock of reserves, which is “polluted” by valuation changes. At the quarterly frequency, on the other hand, our proxy is based on balance-of-payments (BOP) data—which are transaction-based—and, thus, does not require an adjustment for valuation changes, except where BOP data is not available and the change in the stock of reserves is used also for the quarterly proxy.

  • Estimates of investment income on reserve assets are also based on the available information on the composition of assets, combined with information on market interest rates as well as interest payments by issuers of reserve currencies (to approximate average coupon rates on longer-term securities). At the quarterly frequency, we use investment income on reserves reported by some countries as part of their BOP statistics. Estimates of investment income are applied to both the monthly and quarterly FXI proxy.

  • The proxy adjusts for changes in other foreign assets and liabilities based on available information in the Standardized Report Forms of the IMF’s Monetary and Financial Statistics. This includes holdings of non-reserve foreign assets as well as liabilities vis-à-vis all foreign creditors, including IMF and other official creditors.

  • Finally, the proxy adjusts for variations in foreign currency assets and liabilities vis-à-vis domestic entities, also using information available from the IMF’s Monetary and Financial Statistics. As discussed below, this is a key improvement vis-à-vis proxies used in the literature up to now.8

FXI through Currency Derivatives

Operations with foreign exchange derivatives have also become increasingly important instruments for intervention in foreign exchange markets. With varying magnitudes, these have been commonly used by AEs and are increasingly used in EMDEs, as shown in Figure 4. Moreover, these numbers may underestimate the true frequency of FXI with derivatives as they are based on data reported to International Reserves and Foreign Currency Liquidity Template and some countries do not report data to it. Capturing interventions with derivatives is key as they may entail similar effects on exchange rates than interventions in spot markets (see Nedeljkovic and Saborowski, 2019; and Chapter 5 in Chamon et al 2019). It is also important because some central banks use FX swaps to manage FX market liquidity and smooth market functioning. These FX swaps are combinations of spot and mirroring forward interventions and do not necessarily affect the central bank’s FX position once both elements (the spot and forward legs) are taken into account.

Figure 4.
Figure 4.

Frequency of Changes in FX Derivative Positions

(percent of group sample)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: IMF, International Reserves and Foreign Currency Liquidity Template (IRFCLT); and IMF staff calculations.Note: Based on reported changes in short and long positions in forwards and futures in foreign currencies vis-à-vis the domestic currency (including the forward leg of currency swaps), and financial instruments denominated in foreign currency but settled by other means. These may underestimate the true extent of FXI with derivatives if economies that do not report to the IRFCLT also use FX derivatives.

The dataset includes a proxy of FXI with derivatives—“derivative” proxy, reported separately—that encompasses changes in the aggregate short and long positions in forwards and futures in foreign currencies vis-à-vis the domestic currency (including the forward leg of currency swaps), and financial instruments denominated in foreign currency but settled by other means (e.g., in domestic currency), as reported in the International Reserves and Foreign Currency Liquidity Template.9

III. Stylized Facts from our FXI Proxy

Our dataset allows for a comprehensive view on the use of FXI as a policy instrument over the last two decades, and unveils some interesting stylized facts:

  • The notion that EMDEs tend to intervene more heavily in FX markets than AEs does not always hold. As a group, EMDEs, relied significantly more on FXI than AEs before the global financial crisis (GFC), and this holds even after excluding China (see Figure 5). In the post -GFC period, however, the extent of foreign exchange intervention in EMDEs has been comparable to that observed in (non-reserve issuer) AEs. The FXI patterns in the latter group have been dominated by economies with large financial sectors which have deployed FXI in significant degrees to cope sizable capital flows (e.g., Hong Kong SAR, Singapore, Switzerland). 11

  • Interventions are mostly in spot markets and asymmetric (with a bias towards FX purchases) in both income groups. Figure 6 further compares the frequency of FXI in spot and derivative markets between non-reserve issuer AEs and EMDEs, as a function of their overall size. Again, we find that both EMDEs and AEs intervene frequently, especially in spot markets, with both conducting spot interventions of absolute size of at least 0.25 percent of GDP per quarter about ¾ of the time. Our dataset also reveals that, even after adjusting for investment income, there is significant asymmetry in the use of FXI, with purchases of foreign currency being significantly more frequent than sales. This is visible for EMDEs as well as AEs. The use of derivatives, on the other hand, remains limited in comparison to spot interventions, although it is more prevalent in AEs that EMDEs.12 Within AEs, financial centers display a particularly extensive use of FXI, especially through derivatives (see Appendix Figure A1.1). The patterns of FXI for other AEs are similar to those observed in EMDEs.

  • While there is significant variation within groups, the degree of exchange rate management in EMDEs and AEs is similar. Using our proxy of FXI, we can characterize countries’ degree of exchange rate management computing an index of the following form:

Figure 5.
Figure 5.

Foreign Exchange Intervention by Group

(broad proxy, in percent of GDP)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: authors’ estimations.Note: Estimated average quarterly foreign exchange intervention (using “broad” proxy=spot plus derivatives) weighted by 3-year moving average nominal GDP.
Figure 6.
Figure 6.

Size and Symmetry of Foreign Exchange Intervention

(frequency)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: authors’ estimations.Notes: Based on quarterly observations and using our “broad” proxy=spot plus derivatives transactions.
ρi=σiFXIσiFXI+σiNEER(1)

where σiFXI is the standard deviation of our broad FXI proxy (in percent of GDP) and σiNEER is the standard deviation of the change in the nominal effective exchange rate, both for the full sample period. The index takes the value 0 for countries that do not intervene (pure floating) and 1 for those that intervene and whose nominal effective exchange rate does not move (de facto pegs).13,14 Figure 7 displays the values of equation (1) for AEs and EMDEs. While on average AEs allow for slightly greater exchange rate flexibility in comparison to EMDEs, the distributions are not very different and there is considerable variation within both groups. Among AEs, economies with large financial sectors (like Hong Kong SAR, Iceland, Singapore, Switzerland, etc.) tend to display a higher degree of exchange rate management than other AEs. Within EMDEs, there is also a wide range of values, with some large EMDEs displaying flexible regimes comparable to the most flexible AEs (Mexico, South Africa, Colombia, Brazil, Turkey), while others displaying highly managed regimes (including because they have de jure fixed exchange rate regimes, like Bulgaria and Saudi Arabia).

Figure 7.
Figure 7.

Degree of Exchange Rate Management, 2000–20

(Index)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: IMF, International Finance Statistics; IMF, Information Notice Systems; International Reserves and Foreign Currency Liquidity Template; and IMF staff calculations.Notes: The figure reports the index ρ as defined in equation (1) computed over the full sample period (thus, the index may not reflect the most recent pattern of exchange rate management). It varies from 0 (pure floating) to 1 (peg). Major reserve-issuing countries (Euro area economies, Japan, the U.K. and the U.S.) are excluded.
  • Interventions remain dominated by transactions in spot markets. While interventions in foreign currency derivatives have become more frequent, as indicated by Figure 4 above, their size remains smaller than interventions in spot markets. At the same time, we find that derivatives are used significantly during financial crises, as shown by the spikes in the aggregate series during the global financial crisis and the European crisis (Figure 8). This allows central banks to preserve liquidity buffers while still providing foreign currency hedging during episodes of heightened exchange rate market pressures.

  • Countries that intervene less are more inclined to publish their FXI operations. As shown in Figure 9, our data indicate that countries that publish FXI data also intervene less in foreign exchange markets, as illustrated by the greater mass in the density function around zero, in comparison to countries that do not publish FXI data. That is, the lack of publication does not indicate a lack of use of the instrument, but the contrary.

Figure 8.
Figure 8.

Spot versus Derivatives

(absolute value, percent of GDP)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: International Reserves and Foreign Currency Liquidity Template; and authors’ calculations.Notes: Spot and Derivatives denote quarterly sum of the respective absolute values across countries as a share of their (3-year moving average) group GDP.
Figure 9.
Figure 9.

FXI Proxy for Publishing and Non-publishing Economies

(percent of GDP, broad FXI proxy)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: Country authorities and authors’ calculations.Note: Based on broad FXI proxy (spot plus derivatives) at quarterly frequency. The support of the distributions is limited to +/-.05, and thus absolute intervention beyond that value appear as a mass in the extremes of the chart. Based on countries that report FXI data at least on a quarterly frequency.

IV. Published Data

In addition to our FXI proxies, the accompanying dataset includes time-series of publicly available information on FXI. To our best knowledge, this is the most comprehensive dataset of published FXI. Putting these data together entailed a multi-stage effort. First, we searched through central bank websites, aided by the information reported in the IMF’s AREAER.15 The format of available data varies significantly across countries as some central banks provide time-series directly in a usable format while others report scattered FXI data information through separate reports, for example monthly or quarterly reports on monetary and exchange rate policies. Thus, a second effort entailed hand-collecting information from such reports. Country-specific details, including sources, are available in Appendix Table A1.1. Finally, the collected information was shared and discussed with various IMF country teams. In some cases, central banks were contacted to corroborate the collected information.

The dataset includes published FXI for 43 countries at quarterly frequency and for 39 countries at monthly frequency. 17

As discussed in detail in Section V.D, an important issue regarding published FXI data is that there is no uniform definition or template for reporting FXI and, in some cases, reported series may not correspond to the concept of FXI of interest (i.e., transaction-based changes in the foreign-currency position of the central bank). In fact, in most cases, there is no easily accessible and clear definition of the published FXI series (some exceptions are Australia, Costa Rica and Mexico). Thus, it is unclear whether the treatment of certain FX transactions is uniform across countries. For example, differences in the treatment of operations done with or on behalf of the government or state-owned enterprises,18 or the treatment of exporter surrendered requirements may give rise to significant differences in what is reported as FXI across countries. Also, in some cases, reported series may only encompass operations in FX markets (excluding bilateral transactions with other, private or public, entities) while in others the series may report any exchange of foreign and domestic currency assets.

V. Comparing Proxied and Published FX Interventions

This section compares published FXI data with commonly used measures of FXI and our spot proxy to assess whether the latter approximates published FXI better.20

However, the analysis should be interpreted with caution. It is based on the premise that published FX series encompass all transactions of the central bank that alter its foreign exchange position (consistent with the definition of FXI used in this paper), although in some cases reported series may be defined differently. To mitigate this discrepancy, published FXI data were not used for the analysis in this section where definitional differences between our proxy and the published FXI data were likely, i.e. where stated definitions clearly departed from the FXI concept in this paper. Specifically, two main criteria were used to exclude published FXI data from the analysis below. Firstly, countries classified as freely floating for at least 10 years in the AREAER dataset (which overwhelmingly report zero interventions in most periods) were excluded. This encompassed United States, United Kingdom, Canada, Japan, Euro area, Australia, New Zealand, and Mexico. Secondly, Angola, Bolivia, Nicaragua, Nigeria, and Turkey were also excluded on account of large discrepancies between published FXI and the change in reserves (e.g., published interventions are mostly one-sided while the stock of reserves drifts in the opposite direction over time). Users of our dataset are advised to treat the data for these countries with particular caution as proxies and published FXI series may correspond to different FXI definitions.21

A first glance at the data indicates that our spot proxy for FXI matches published interventions— where available—more closely than the change in the stock of reserves (see Figure 10 for a comparison at the monthly frequency). The raw correlation of proxied and published FXI is 0.61 while that of changes in reserves and published FXI is 0.54. Moreover, proxied and published interventions rarely have different signs, with only about 0.5 percent of country/month observations having a meaningfully different sign.22 The relatively low correlation is driven by a few outliers for which published FXI is reported to be zero but our proxies assign it a non-zero value, with the difference possibly reflecting a narrower definition of FXI used by some of these central banks.23

Figure 10.
Figure 10.

Correlation with published FXI. Spot proxy vs. change in reserves

(share of GDP, monthly frequency)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: IMF International Financial Statistics and authors’ estimates.Note: Published data includes only spot interventions.

In what follows, we compare our spot proxy as well as other coarser proxies with published FXI data more formally to learn about the information content in these different measures, and to identify which specific adjustments to the coarse measures are useful to provide a more accurate picture of FXI.

A. Comparing Proxies: Univariate Regressions

We start by using a classic measurement error framework (see, for example, Greene, 2000) to compare more formally the accuracy of our spot proxy versus coarser measures traditionally used. Specifically, consider the following measurement error equation:

xi,t=αi+βxi,t*+εi,t(2)

where xi,t* is published FXI data, xi,t is the FXI proxy for country i and period t, both expressed in percent of GDP24 and εi,t~N(0,σe). In this specification, αi captures any systematic deviation between proxied and published data and the error term is assumed to be random measurement error. If proxies are polluted only by classic measurement error, αi ≈ 0, β ≈ 1. In addition, if there is no measurement error, then σε ≈ 0.25

The coefficient β above is estimated using a panel fixed effects estimator for the changes in reserves as well as for our spot proxy. These comparisons are done both for the monthly and quarterly proxies as the construction of each of these proxies varies (due to availability of balance of payments data, which are purged of valuation effects, at different frequencies) as explained before.

The spot proxy displays a coefficient closer to 1, including statistically, compared to the regressions using the change in reserves or BOP flows, although the difference is small (see Table 1). The r-squared of the regressions using the spot proxy are also considerably higher, pointing to greater accuracy of the latter vis-à-vis the other, coarser, measures. Moreover, the constant is several times smaller in the case of the spot proxy, indicating that the latter displays significantly smaller systematic deviations on average26 from the published data than the other proxies, partly due to the fact that the spot proxy strips out investment income from the growth in international reserves.

Table 1.

Published and Proxies of FXI. Full Sample (2000–19)

article image
Source: authors’ estimationsNotes: Standard errors clustered at the country level in parentheses. *** p<0.01, ** p<0.05, * p<0.1Source: authors’estimations.

As shown in Table 2, the advantage of the spot proxy vis-à-vis the coarser measures is even more evident when the sample period is restricted to the last decade (2010–19). This is visible in the larger differences in the β coefficient and the r-squared, and reflects the fact that, with growing stocks of reserves (and foreign currency assets more broadly) over time, valuation changes have become larger and, thus, leading to a lower accuracy of the change in the stock of reserves as a proxy for FXI. A similar notion applies to the investment income component, although the secular decline in global interest rates has reduced the importance of this component over time (see also Appendix 1). Finally, the adjustments made due to changes in central banks’ FX positions are also typically available only after 2002, and for some cases later, making the spot proxy more accurate in more recent years.

Table 2.

Published and Proxies of FXI. Recent years (2010–19).

article image
Source: authors’ estimations.Notes: Standard errors clustered at the country level in parentheses. *** p<0.01, ** p<0.05, * p<0.1

B. A Horse Race: Spot Proxy vs. Changes in Reserves

An alternative approach to evaluate the usefulness of our proxy vis-à-vis changes in reserves is through ‘horse race’ regressions of the following type:

xi,t*=ai+γ1xi,t1+γ2xi,t2+ξi,t(3)

where xi,t* is published FXI data; xi,t1 is the change in reserves; and xi,t2 is our spot proxy, all expressed in percent of GDP. For quarterly data, we also compare changes in reserves against reserve transactions as recorded in the balance of payments.

The results point to the superiority of our spot proxy over the change in reserves (Table 3). Specifically, our spot proxy adds significantly to the r-squared of the regression at the monthly frequency (going from column 1 to 2). Also, its coefficient in column 2 is positive and significant at 1 percent level, meaning that an upward change in our spot proxy correlates with purchases reported in published FXI, controlling for changes in reserves. That is, there is significant additional informational content in our proxy in relation to the information included in the changes in reserves. The same holds for quarterly data, where also reserve transactions as recorded in the BOP seem to improve on the change in reserves (column 4 compared to 3). Our spot proxy, though, dominates the BOP reserve flow since its coefficient becomes insignificant in a three-way regression (column 5). The r-squared again is also considerably larger when the spot proxy is introduced.

Table 3.

Horse Race: Changes in Reserves vis-à-vis Spot Proxy (2000–19)

article image
Source: authors’ estimations.Notes: Standard errors clustered at the country level in parentheses. *** p<0.01, ** p<0.05, * p<0.1

C. Dissecting the Spot Proxy

An approach similar to the above can be used to assess the value of specific elements of our proxy. This entails dissecting the spot proxy into its different components in order to assess their individual contributions. To this end, we construct “intermediate” proxies to be included in equation (3), one for each main adjustment made to the changes in reserves as part of the construct of our spot proxy, introducing adjustments incrementally. Such a horse race allows to evaluate the individual importance of each adjustment in terms of providing information that correlates with published FXI data that is not already embedded in the change in reserves. That is, a significant coefficient on any one of these intermediate proxies indicates that the adjustment that makes this intermediate proxy different has additional information and is helpful in getting our spot proxy closer to published FXI data. Moreover, changes in the value and statistical significance of the coefficient of the change in reserves serve as a test on whether the spot (intermediate or final) proxies subsume the information embedded in the change in reserves.

Table 4 displays horse races at monthly frequency across the different adjustments made to arrive at our final proxy, confirming that several of our adjustments to changes in reserves improve the FXI proxy, bringing it closer to the published FXI data. First, we evaluate the adjustments for investment income and valuation changes (column 2). Although the coefficient of the corresponding intermediate proxy that adjusts for income and valuation effects is not significant at the 10 percent level, partly reflecting collinearity, it is significant at 15 percent level and it increases the r-squared, suggesting that while the contribution may be small, it adds information. Adjusting for the position of the central bank vis-à-vis the IMF (column 3) moves the FXI proxy substantially closer to published data as indicated by the highly significant coefficient of the intermediate proxy and the higher r-squared.27 Column (4) shows that adjustments to the net position of the central bank vis-à-vis other nonresidents do not seem to add much on their own (the coefficient is insignificant and the r-squared is broadly unchanged). Finally, the adjustment for central bank’s positions vis-à-vis residents (column 5), is found to have the largest impact, with a coefficient that is highly significant and a material impact on the r-squared. Note that, in this case, the coefficient on the adjustment vis-à-vis other nonresidents becomes negative (and highly significant), denoting some offsetting between the two adjustments for central bank positions relative to non-residents and residents. Finally, note that the coefficient on the change in reserves is always insignificant in columns (2)-(5) indicating that all the relevant information in the change in reserves (and more) is contained in our proxy. Appendix Table A1.3 shows similar findings for the proxies at quarterly frequency, as appendix Tables A1.4 and A1.5 focused on most recent years (2010–19).

Table 4.

A horserace between proxies. (2000–19, monthly frequency)

article image
Source: authors’ estimations.Notes: Standard errors clustered at the country level in parentheses. *** p<0.01, ** p<0.05, * p<0.1

D. Possible Future Refinements

While the proxy of FXI presented in this paper entails a refinement relative to previous measures, there is scope for further improvements along some dimensions, if more granular data become available. Specifically:

  • Currency composition: A limitation in the computation of valuation changes arises from the limited reporting of country-specific information on the currency composition of central banks reserves (and foreign assets more broadly). The available information is currently provided by country authorities to the IMF as part of their submissions to Currency Composition of Official Foreign Exchange Reserves (COFER) database, but the individual country data is confidential. Some progress has been made with individual reporting and publication of data under the International Reserves and Foreign Currency Liquidity Template, although the current set of countries remains small. Thus, valuation changes for most countries are estimated on the basis of aggregate currency shares. For many countries and in early years, this assumption is likely innocuous as the US dollar dominates reserve holdings. For countries with close economic ties to economic areas that issue other reserve currencies, however, other reserve assets may play a greater role (e.g., Eastern European economies may hold larger than average reserve shares in Euro-denominated assets).28 In addition, while still relatively low, some currencies (like the Swiss Franc, Canadian dollar, Australian dollar, Renminbi) have gained some ground in reserve holdings over time.

  • Equity holdings. As foreign currency holdings grow, surpassing desirable levels of liquidity buffers for precautionary motives in some economies, investment diversification towards higher return and less liquid assets-including equity--is likely to take place. Consequently, valuation changes may become more volatile and difficult to estimate unless granular data on such investments are made available.

  • Coverage. In line with the rest of the literature, our work focused on foreign currency transactions by the central bank. In some cases, however, operations by other public entities (like the central government, public banks, sovereign wealth funds or state-owned enterprises) share some of the key characteristics of central bank interventions (Bayoumi et al 2015). Limited data on the balance sheets of these public entities, especially at high frequency and breaking down positions by currency, precludes a broader coverage of foreign currency transactions at this point.

  • Consistency in published data. The comparisons made in Section V.A-C assumed, except for a few cases, that published FXI data reflects the concept of FXI of interest closely. As mentioned before, however, there is limited information on the definition of FXI used by central banks and, thus, it is possible that the definition of reported series may differ across countries. Converging on a commonly agreed definition will help ensure that published data is consistent, increasing their usefulness for direct cross-country analysis as well as for evaluating the accuracy of available proxies.

VI. Conclusions

Exchange rate management remains a central topic of international economics and a key element of policy making. Yet, data on central bank interventions in foreign exchange markets remain scant.

This paper provides a new database of publicly available data and carefully constructed proxies for FXI—for both operations in spot and derivative markets, at monthly and quarterly frequencies-focusing on transactions that alter the central bank’s foreign currency position. The dataset significantly expands available FXI data, and improves on previously used measures by accounting for a wide range of central bank operations, including, crucially, operations vis-à-vis residents. These data are particularly useful for rigorous and comprehensive cross-country macroeconomic analysis aimed at shedding light on the merits and downsides of FXI as a policy tool—a key aspect of the IMF’s Integrated Policy Framework.

Appendix 1: Additional Tables and Figures

Appendix Table A1.1.

Countries that Publish Foreign Exchange Intervention Data

article image
Sources: Various.Notes: “Time series complied from multiple reports” indicates that information on FXI was collected from multiple, scattered sources. Typically, these are monthly or quarterly reports on monetary and exchange rate policies. This table reflects our best knowledge as of December 2020 and may not reflect all countries that publish FXI data.
Appendix Table A1.2.

Country-by-country Regressions of Published on Proxied FXI

article image
Notes: Constant and slope coefficients of country-by-country regressions of proxy on published FXI are reported. A star denotes p-value smaller than 0.05.
Appendix Table A1.3.

A horserace between proxies. Full sample (2000–19, quarterly frequency)

article image
Source: authors’ estimations.Notes: Standard errors clustered at the country level in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Appendix Table A1.4.

A horserace between proxies. (2010–19, monthly frequency)

article image
Source: authors’ estimations.Notes: Standard errors clustered at the country level in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Appendix Table A1.5.

A horserace between proxies. (2010–19, Quarterly frequency)

article image
Source: authors’ estimations.Notes: Standard errors clustered at the country level in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Figure A1.1.
Figure A1.1.

Size and Symmetry of Foreign Exchange Intervention. Financial Centers and other AEs.

(frequency)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: authors’ estimations.Notes: Based on quarterly observations and using our “broad” proxy=spot plus derivatives transactions.

Appendix 2: General Methodology for Building the Proxies

A. Conceptual Framework

As discussed in the main text, we are interested in a concept of foreign exchange intervention that captures central bank’s transactions that alter its net foreign currency position. To derive a proxy for such a concept, consider a simplified version of a central bank’s balance sheet (expressed in local currency):

RES + OF A + FCDA + LCDA = FL + FCDL + LCDL + EQ

where RES denotes foreign currency reserves; OFA are other foreign assets; FCDA denotes foreign currency domestic assets; LCDA are Local-Currency Domestic Assets; FL are foreign liabilities; FCDL are foreign currency domestic liabilities; LCDL are domestic currency domestic liabilities; and EQ denotes equity or net worth. This expression can also be written in terms of the net foreign and local currency positions, as:

NFCP = NLCP + EQ

where NFCP = RES + OFA + FCDA – FCDL – FL and NLCP = LCDA – LCDL. It follows that:

dNFCP = dNLCP + dEQ

Our interest lies only in changes of NFCP arising from transactions and thus have a counterpart on change in the NLCP (not in the equity position, EQ):

dNFCPFXI = dNLCPFXI

In many cases, only the overall change in the net foreign currency position, which includes investment income and valuation changes, is observed:

dNFCP = FXI + NFCP_InvIncome + NFCP_Val

Thus, an estimate of FXI can be built from the observed change in NFCP and the estimated investment income and valuation changes.

Est[FXI] = dNFCP – Est[NFCP_InvIncome] – Est[NFCP_Val]

Since reserve assets tend to dominate NFCP, valuation changes and income flows will be primarily driven by reserve assets. This means that FXI can be estimated as:

Est[FXI] = dNFCP – Est[RES_InvIncome] – Est[RES_Val]

This can also be written as:

Est[FXI]=dRES+d(OFAFL)+d(FCDAFCDL)Est[RES_InvIncome]Est[RESVal]

The latter equation highlights how a refined proxy of FXI departs from the coarse measure of changes in reserves (dRES) to include changes in other assets and liabilities, changes in foreign assets and liabilities vis-à-vis residents; and estimated investment income and valuation changes. These adjustments are key as changes in reserves can vary for a number of reasons without entailing a transaction that alters the central bank’s NFCP (i.e., FXI):

  • Changes in reserves can reflect reallocation between reserve and non-reserve foreign assets. This is particularly relevant in economies with significant reserve assets—where reserve holdings may exceed levels consistent with precautionary liquidity buffers and, thus, central banks may choose to invest part of their foreign currency holdings in less liquid assets.

  • Similarly, reserve levels may vary as central banks borrow from or repay loans to foreign entities, without entailing changes in the foreign currency position.

  • In some economies, the central bank conducts operations with foreign exchange on behalf of other domestic entities (most commonly the central government). This can temporarily affect the holdings of reserve and non-reserve foreign currency assets, but they also entail a corresponding foreign currency liability vis-à-vis those domestic entities.

  • Similarly, in many emerging and developing economies, banks are allowed to take on deposits in foreign currency, with associated reserve requirements. This can translate in changes in foreign currency (reserve or non-reserve) holdings of the central bank with corresponding foreign currency liabilities vis-à-vis domestic banks.

B. Constructing the FXI proxy

This section describes how different sources of data are used to construct a proxy of FXI with the adjustments described above.

Non-reserve foreign currency asset and liabilities

Information on reserve assets comes from the Fund’s International Financial Statistics (IFS). Non-reserve foreign assets and liabilities as well as foreign currency position vis-à-vis residents are obtained from the IMF’s Monetary and Financial Statistics (MFS). Appendix Table A2.1 provides details of the data coverage.

Valuation changes and investment income

Estimating investment income and valuation changes entails using available information on the structure of assets and liabilities in terms of their currency and asset composition. In order to understand how different sources of information are combined, consider the stock of reserves which can be written as the sum of its components as follows:

RES=ΣcΣaScPacHac

where Sc denotes the exchange rate US dollar per currency c; Pac is the price of asset type a expressed in currency c; and Hac is the (quantity) holding of asset type a denominated in currency c.

The investment income on reserves can, therefore, be written in terms of in investment income on the individual components:

RES_InvIncome=ΣcΣaiacwac

where wac=scPacHacΣcΣaScPacHac is the share of currency – asset type a in total reserves.

Appendix Table A2.1.

Coverage of Central Banks’ Foreign Currency Positions Data

article image
article image
Source: Monetary and Financial Statistics Dataset

The International Reserves and Foreign Currency Liquidity Template (IRFCLT) provides information on the composition of assets. Information on the currency structure is available from the Currency Composition of Official Foreign Exchange Reserves (COFER) dataset and IRFCLT. However, for most categories of asset types, there is no information of the currency structure within the asset type. The only exceptions are the IMF position and SDR holdings, which are denominated in SDRs (and gold holdings, which can be valued at the price of gold in US dollars). Thus, for all other asset types, the currency structure is assumed to be invariant. That is, ScPacHacScPcHc, which implies that scPacHacΣcscPacHac=wc and therefore wac=wcwa.

The above expression for the income on reserves can, thus, be written as:

RESInvIncome=ΣcΣaiacwcwa

In turn, valuation changes arising from both exchange rate and asset price movements can be expressed as:

RESVal=ΣcdScScwc+ΣcΣadPacPacwcwa

In order to estimate these components, currency weights of the five major currencies (USD, Euro, Yen, Pound, Renminbi) as well as the SDR are considered as the account for the vast majority of reserves. Country-specific currency weights (wc) are taken from the Reserve Template, when data are reported. Otherwise, group averages from COFER are used. As mentioned, individual country information is confidential for all COFER reporting economies.

Asset type weights are based on the country-specific information available in the IRFCLT. The key categories included there are: (i) Currency and Deposits; (ii) Securities; (iii) Position at the IMF; (iv) SDR holdings; (v) Gold; (vi) Other foreign currency assets. As shown in Appendix Figure A2.1, the bulk of foreign assets is held in the form of Securities and Currency & Deposits, which have relatively stable shares over time, although there is considerable cross-sectional heterogeneity. Other forms of reserves, like claims on the IMF, SDRs, Gold and Other Assets represent a small fraction of total reserves for most countries.

Appendix Figure A2.1.
Appendix Figure A2.1.

Weights by asset type

(share of total foreign currency assets)

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: International Reserves and Foreign Currency Liquidity Template (IRFCLT), and authors’ calculations.Notes: Cross-section percentiles 25, 50 and 75 among economies reporting to IRFCLT (and with GDP larger than US$10 billion in 2019) are depicted.

Data on the asset structure of reserves is available for about 50–60 percent of the countries in the sample from the IRFCLT (see Appendix Figure A2.2). We complement that information with partial information available in MFS. Concretely, MFS reports gold and SDR holdings of central banks, which we use to calculate the weight of those two items in countries’ reserves.

Using this information, asset weights are constructed as follows: (i) based on country-specific shares reported in the IRFCLT, when available; (ii) when country-specific information is only available for some (most recent) periods, shares are spliced backwards using median weights observed for other countries to account for broad trends in the asset structure, while maintaining heterogeneity in levels; (iii) when no country-specific information is available in IRFCLT, we use MFS based data for the weights on gold and SDR assets and use the median breakdown of foreign currency assets from IRFCLT for the rest of the assets; and (iv) finally, if no information is available from either MFS or IRFCLT the median of the weights observed for all countries in IRFCLT are used.

Appendix Figure A2.2 illustrates the coverage of available country-specific asset-type weights, which has grown over time but remains only slightly above half of the sample of countries of analysis. For most AEs and some EMDEs, there is ample information since the early 2000s, while data for smaller EMDEs and LICs remains limited.

Appendix Figure A2.2.
Appendix Figure A2.2.

Coverage of data on asset structure

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: International Reserves and Foreign Currency Liquidity Template (IRFCL), and authors’ calculations.

Investment income

Some countries report investment income on reserves as part of the income balance in their balanc of-payments statistics, although in many cases providing short time series (Appendix Figure [A2.3]) This provides an exact measure of the investment income on reserves that needs to be netted out. When available, such series are used as a measure of income on reserves.

Appendix Figure A2.3.
Appendix Figure A2.3.

Investment Income on Reserves

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: International Reserves of Foreign Currency Liquidity, Balance of Payments Statistics, author’s calculations.Notes: Left panel depicts cross-section percentiles 25, 50 and 75 among economies (with GDP larger than US$ 10 billion in 2019) reporting to IRFCLT and BOP. Right panel shows the share of countries in the sample with reported BOP data on investment income on reserves.

Otherwise, the interest on reserves is estimated using available information on currencies and asset types combined with estimates of the coupon rate on the different instruments as follows:

Appendix Table A2.2.

Interest rates (iac)

article image

Effective interest rate paid by the corresponding reserve-issuing countries. Source: Authors’ calculations.

As shown in Appendix Table A2.2, the interest rate on short-term assets like currency and deposits can be estimated fairly accurately with the prevailing market rate. Estimating the coupon rate on longer-term securities, however, is more challenging as the holdings of this type of securities may encompass a portfolio of different maturities with coupon rates that may be different from the prevailing market rates. To overcome this obstacle, the effective rate paid by reserve currency-issuing countries, available from their public accounts, is used. A synthetic portfolio combining 2-year and 10-year bonds purchased over the last 2 and 10 years, respectively, which are shown to track the effective interest rate paid closely over the sample period (Appendix Figure A2.4).

ι¯¯tc=(0.7)*Σι=110(110)itιc,10y+(0.3)*Σι=12(12)itιc,2y
Appendix Figure A2.4.
Appendix Figure A2.4.

Effective interest rate and 10-Year Treasury Bond Yield

Citation: IMF Working Papers 2021, 047; 10.5089/9781513566672.001.A001

Sources: World Economic Outlook, Haver Analytics, and authors’ calculations.Note: Effective interest rate is imputed by dividing general government interest expense by general government gross debt. Synthetic portfolio composed of 70 percent of 10-year maturity bonds purchased over the last 10 years and 30 percent of 2-year maturity bonds purchased over the last 2 years.

Valuation changes are estimated with the following assumptions for asset prices:

Appendix Table A2.3.

Asset Prices (Pac)

article image
Source: Author’s calculations.

Currency and Deposits, the SDR and IMF position are taken at nominal value, implying that only exchange rate changes lead to valuation changes for this component. Consistent with the estimation of interest income, security holdings are assumed to combine 2- and 10-year bonds purchase in previous 2 and 10 years respectively. The market price of the basket of securities in currency c can, thus, be estimated as:

Ptc=(0.7)*ΣRM=110(110)Pc,10,RMY+(0.30)*ΣRM=12(12)Pc,10,RMY

where Ptc,M,RMY is the price of a bond of original maturity M and residual maturity in years RMY that pays out interest ∝c,M on a quarterly basis, is given by:

Ptc,M,RM=c,M[1(1+iRM4)RM]iRM/4

where iRM is the observed M-quarter maturity (residual) market yield and RM is the number of quarters of residual maturity. Information on the yield curve at each point in time is used to estimate changes in these prices. This entail a refinement with respect to previous estimations (e.g., Dominguez, 2012) which assumed a 10-year residual maturity bond was systematically held.

Gold holdings are valued at end of period market rates. Exchange rates are end-of period nominal rates vis-à-vis the US dollar (e.g., USD/Euro), as reported in International Financial Statistics.

VII. Adjustments at Different Frequencies

The requirements for the computation of the FXI proxy depend on the frequency of interest.

Quarterly frequency. The computation of FXI proxies at quarterly frequency makes use of the data from balance-of-payments statistics, which report changes in reserves on a transaction basis and, thus, are not polluted by valuation changes. Thus, quarterly proxies based on BOP flows only require adjusting for investment income flows.

Monthly frequency. Monthly proxies present additional challenges as transaction-based data (akin to those provided by balance-of-payment statistics) are rarely available. Thus, proxies rely on changes in stocks, which not only contain investment income flows but also exchange rate and asset prices changes. Thus, monthly proxies need to be adjusted for valuation changes as well.

Appendix Tables A2.4 and A2.5 provide details on how different country-specific information is used to estimate valuation changes and investment income for specific countries and periods of time, at quarterly and monthly frequency, respectively.

Appendix Table A2.4.

Basis for Valuation and Income Adjustments. Quarterly Frequency

article image
article image
Appendix Table A2.5.

Basis for Valuation and Income Adjustments. Monthly Frequency

article image
article image

Foreign Exchange Derivatives

Operations using foreign exchange derivatives include changes in aggregate short and long positions in forwards and futures in foreign currencies vis-à-vis the domestic currency (including the forward leg of currency swaps), and financial instruments denominated in foreign currency but settled by other means (e.g., in domestic currency), as reported in the International Reserves and Foreign Currency Liquidity Template. Some clarifications are important:

  • - These notional amounts are not reported as part of the central bank’s balance sheet, which only includes the market value of derivative positions (i.e., in- and out-of-the money positions) and, thus, may under or overestimate the actual FX derivate position.

  • - While changes in the notional amount properly capture new FXI operations through derivatives (e.g., a sale of FX forward), they would also capture the unwinding of derivative positions when they come due (in the previous example, this would result from the actual provision of FX as dictated by the previously sold forward FX contract). However, the unwinding of positions would normally be mirrored by other components of the central bank’s balance sheet (drop in FX holdings in the example). Thus, a broad measure of FXI that encompasses spot and derivatives could adequately capture the initial FXI operation with derivatives while netting out their automatic unwinding.29

  • - Some central banks use FX swaps to manage liquidity and smooth market functioning. These FX swaps are combinations of spot and mirroring forward interventions and do not necessarily affect the central bank’s FX position and, thus, do not represent FXI as defined in this paper. The spot and forward legs of these swaps are captured in our proxies for spot and derivative FXI, respectively. For countries using FX swaps frequently, a broad FXI measure that combines (adds) the spot and derivative FXI proxies provides a more accurate read of actual FXI operations

  • - Swap lines among central banks are also, in general, of different nature. While they provide foreign currency liquidity, they do not alter any of the central banks’ net foreign currency positions. Specifically, when swap lines are first established, they entail only contingent credit lines with no impact on the balance sheet or derivative positions of either central bank. When the swap lines are activated or drawn, there is an increase in both central banks’ foreign assets and liabilities. Despite the fact that a draw down entails a swap of currencies, there is no change in net foreign currency positions of either central bank as the exchange of currencies only serve the purpose of providing collateral. Any variation in the value of one currency relative to the other triggers an additional provision of the depreciated currency to maintain the parity in the value of currency holdings. Thus, there is no currency risk in the swap lines for any of the participating central banks.30

Appendix 3: AD-HOC Adjustments to Proxied FXI

article image
article image

Appendix 4: Summary Metadata for Published FXI

article image
article image
article image
article image
article image
article image
article image
article image
article image
article image