The Foreign Exchange Market

Annotation

This chapter examines the structure and functioning of the spot foreign exchange (FX) market. The market structure, which has become much more complex over the past three decades, has evolved largely endogenously, as the global FX market is subject to much less regulation than equity and bond markets in most countries. Large banks used to dominate liquidity provision, but in an increasingly fragmented electronic market, their role has been undermined by high-frequency trading firms. The information structure of the market has also changed. In this context, high-frequency cross-correlations of assets, especially with the futures market, have become increasingly important. The chapter also discusses the important role of the official sector in the foreign exchange market and highlights some special topics such as flash events and the currency fixing scandal. We conclude with some suggestions for future research. '

Keywords: Financial markets, foreign exchange, market microstructure, dealer intermediation, electronic trading, algorithmic trading.

JEL Classification: F31, G15

*This article was prepared for the Research Handbook on Financial Markets, edited by Refet Gurkaynak and Jonathan Wright.

We are grateful to Refet Gurkaynak and Jonathan Wright, our editors, and Angelo Ranaldo, our panelist, for valuable comments. We also benefited from conversations and exchanges with Mark Bruce, Alexei Djiltsov, Colin Lambert, Alexis Laming, Thomas Maag, Eugene Markman, Josh Matthews, James O'Connor, Roel Oomen, Dan Reichgott, Morten Salvesen, Merhrdad Samadi, James Sinclair, Nikolai Tarashev, and Clara Vega.

We thank EBS, Refinitiv and BIS for kindly providing the data. The views expressed in this chapter are solely those of the authors and should not be interpreted as reflecting the views of the Federal Reserve Board or the Bank for International Settlements.

"Foreign exchange spot is the simplest asset class to trade, yet the most complex."

Quote from the head of a major FX liquidity provider.

Introduction

The foreign exchange (FX) market, which determines the relative prices of the world's currencies, is essential for international transactions in goods, services and financial assets. In addition, the FX market is often regarded as an asset class in its own right. Therefore, the end users of the FX market are a wide variety of financial and non-financial customers around the world. The trading activities of these agents and their interaction with market intermediaries determine the exchange rate setting process, which influences virtually all international economic activity. As a result, the foreign exchange market is the largest financial market in the world. Trading volumes in the foreign exchange market, for example, far exceed those in the global equity market (King and Rime, 2010).

Foreign exchange trading takes place around the world and around the clock, with the weekly cycle beginning early Monday morning in Asia-Pacific and ending on Friday afternoon in the Americas. Trading activity often peaks when the day trading hours in London and New York coincide, and is relatively low during the so-called "witching hour" - late evening in New York and early morning in Asia.

More than 50 currencies are regularly traded in the market, but the U.S. dollar (USD) has long dominated as the transportation currency. The US dollar is a party to nearly 90% of all global currency transactions, with the euro (EUR) and Japanese yen (JPY) a distant second and third place (BIS, 2022).

Similarly, although the number of market participants around the world is very large, the majority of liquidity provision in the FX market is concentrated with a relatively small number of global banks and non-bank liquidity providers. Thus, despite the global and dispersed nature of the FX market, some aspects of the market exhibit a high degree of concentration.

The market structure supporting this activity is constantly evolving. Overall, the FX market is an over-the-counter market in which electronic trading has grown rapidly since the early 2000s. It used to have two completely different segments - inter-dealer and dealer-customer. Over time, these distinctions have become blurred: multiple trading venues have emerged, the variety of transaction execution methods is growing, and some non-bank participants have begun to act as liquidity providers alongside bank dealers.

This chapter examines the organization and functioning of the spot foreign exchange market, including the major participants, market structure, and the role of the official sector. Knowing how the foreign exchange market functions is critical to understanding how it converges to equilibrium exchange rates, but this chapter does not focus on exchange rates or exchange rate determination per se, which is the focus of an extensive literature (1).

(1) For a review of the theoretical and empirical literature on exchange rates and exchange rate determination, see, for example, Maggiori (2022) and the chapters in James, Marsh and Sarno (2012).

Note that the foreign exchange market also includes a large amount of derivatives trading, including FX swaps, currency swaps, forwards, futures and options. FX swaps and currency swaps, important instruments used primarily for financing and hedging purposes, are discussed in Angelo's chapter.

Market size and its main participants

Measuring global trading activity is challenging because the global FX market is obviously not under a single jurisdiction. However, a comprehensive and authoritative source of information, albeit infrequent, is the central banks'Triennial Survey of Foreign Exchange and Derivatives Market Activity ("Triennial").

The Triennial provides an indication of daily trading activity in the FX market every three years in April (2). Data for the Triennial Survey are collected by central banks from dealer banks in their jurisdiction and then aggregated, analyzed and published by the Bank for International Settlements (BIS). More frequent estimates can be obtained using data from the biannual surveys conducted by the foreign exchange committees (FXCs), industry groups sponsored by central banks in different countries.

(2) At the time of writing this chapter, the most recent triennial survey was conducted in April 2019.

Daily trading volumes and trading geography

Average daily trading volume in the global spot FX market has long been on an upward trend (Figure 1).(3) According to the latest Triennial data, in April 2022, daily trading volume in the global spot FX market averaged $2.11 trillion.(4)

Illustrating the global role of the US dollar as the primary carrier currency, the top three most traded currency pairs were EURUSD with 23% of trading volume, followed by USDJPY with 14% and GBPUSD with 10%. (5)

Figure 1 shows that spot currency trading is concentrated in a few major financial centers. London alone accounted for 38 % in 2022, and the combined share of the four largest trading centers, which also include New York, Singapore and Hong Kong, accounted for 74 % of global spot FX turnover (BIS, 2022). (6)

The major FX trading centers are not necessarily located in countries that dominate global commodity trade, as FX trading for financial motives, such as investing in foreign currency-denominated securities, far exceeds the volume of transactions associated with international trade. In addition, the main data centers that host the matching systems that power electronic trading platforms in the FX market are a critical part of the global FX market infrastructure. This encourages some key players to locate their trading activities near these centers. (7)

Notes: Semi-annual FXC data used to create the benchmark series are collected in April and October. Turnover from the China Foreign Exchange Trading System (CFETS) and Hong Kong Treasury Markets Association survey have been added to the summarized bars from April 2015 and April 2017, respectively.

(3) The notable spike in trading volume in 2014 was associated with monetary easing by the Bank of Japan, while the contraction in 2015 is attributed to the de-risking following the Swiss National Bank's (SNB) unexpected abandonment of the minimum rate on the Swiss franc against the euro (Moore, Schrimpf, and Sushko, 2016).

(4) The average daily trading volume of all OTC currency instruments in April 2022 was $7.5 trillion. Exchange-traded futures and options accounted for less than $0.2 trillion.

(5) Each currency is assigned a three-letter code (code "ISO 4217"). Exchange rates are then represented by a currency pair, with the base currency indicated first. For example, GBPUSD, "sterling-dollar", is quoted in dollars per British pound sterling, and USDJPY, "dollar-yen", is quoted in yen per dollar.

(6) The latter two Asian financial centers have gradually overtaken Tokyo as major centers of trade.

(7) Some of the major data centers used for currency trading are London (LD4), New York (NY4), Tokyo (TY3) and, more recently, Singapore (SG1).

The main types of counterparties

Figure 2 shows the share of trading volume of dealer banks in the foreign exchange market ("reporting" dealers) with the three broad categories of counterparties covered by the central bank and BIS surveys: other dealer banks, other financial institutions and non-financial clients.

Non-financial customers, primarily corporations, use the foreign exchange market to support their core activities, especially international trade.

The broad category of "other financial institutions" traditionally represents the financial customers of reporting foreign exchange dealers. This includes institutional investors, asset managers, hedge funds, commodity trading advisors (CTAs), small banks that are not currency dealers, and central banks.

More recently, principal trading firms (PTFs), often referred to as high-frequency traders (HFTs), have become major players in the "other financial institutions" category.

Over the past 20 years, the share of global FX transactions conducted by non-financial clients has declined from 20% to less than 10%. In contrast, the share of financial customers, which used to be close to that of non-financial customers, has risen to more than 50%. This again reflects the fact that currency trading is increasingly driven by the needs of financial clients rather than by needs arising directly from international trade. (8)

For a long time, dealer banks represented the only sector that provided liquidity and kept risk for the rest of the market. Dealers were the archetypal liquidity providers (LPs) for financial and non-financial customers, who were the archetypal liquidity consumers (LCs).

Dealers then traded among themselves to hedge their positions and rebalance their holdings, which generated a significant amount of trading volume in the FX market. Figure 2 shows that inter-dealer trading accounted for about two-thirds of all spot trading in the FX market in the late 1990s.

Currently, the share of interdealer trading has declined significantly. This is partly because dealers have become less reliant on interdealer trading for inventory control, and partly because the line between traditional LP and LC has become blurred.

In particular, as PTFs have grown in importance in the FX market, some have begun to play a dual role as both LPs and LCs, displacing some of the transactions between dealers and clients. For example, according to Euromoney's widely acclaimed annual FX market study, PTFs will account for nearly a third of electronic FX trading with clients by 2022. (9)

(8) Some retail investors also participate in FX trading, often trading with high leverage through retail margin brokers. But in general they account for a very small proportion of the trading activity of the "wholesale" market participants discussed above. Japan may be an exception, as margin trading in the foreign exchange market by retail investors has become more substantial there (Mukoyama, Kikuta, and Washimi, 2018).

(9) The results of these annual FX market surveys, which Euromoney magazine has been publishing for over 40 years, are a very useful source of information about the global FX market, including the relative importance of different liquidity providers. However, as many companies actively seek to be recognized in the survey, the results are best viewed as indicative.

Trading environment

The organization of the FX market can best be understood in light of three key economic constraints: credit risk management, inventory risk management, and asymmetric information.

Along with technological advances, these constraints have shaped the evolution of foreign exchange market structure over time. Spot foreign exchange transactions can be for very large amounts, and are settled two business days after the transaction is concluded by transferring bank balances in the respective currencies (T+2 settlement, discussed in Section 6.3).

Thus, the foreign exchange transaction has historically been a bilateral extension of credit, which naturally led to a market structure with banks at its core (Lyons, 2002).

The introduction of prime brokerage (PB) in the late 1990s was a significant advance in credit management. The service, first offered to hedge funds by large foreign exchange dealers, allowed non-banks to transact directly with a wide range of counterparties under the umbrella of their service provider. This allowed a hedge fund to access prices and liquidity from a large number of dealer banks solely on the basis of its lending relationship with the prime broker.

Over time, prime brokerage has become widespread in the foreign exchange market. It allows various types of non-banks, including PTFs, to trade on a large number of venues, including those previously considered exclusively "inter-dealer".

Thus, prime brokerage has helped expand the scope of liquidity provision beyond dealer banks and probably also increased the number of trading venues.

Managing foreign exchange inventory imbalances arising from trading with clients used to be a key driver of inter-dealer trading. The repeated transfer of inventory imbalances between dealers in the market, dubbed "hot potato trading" (Lyons, 1996), contributed to the large share of interdealer transactions in total volume observed in the early years in Figure 2.

However, the way dealer banks managed their inventory began to change in the early to mid-2000s when the largest banks began to "internalize" some of their trades, waiting for an offsetting customer trade instead of immediately hedging a position in the interdealer market (Butz and Oomen, 2018).

This was likely facilitated by technological advances as well as a fairly high concentration of trading volume among the largest dealers at the time. Currently, the share of internalization in major currency pairs is estimated to have risen to 80% or more among the largest dealers (Moore et al., 2016; Schrimpf and Sushko, 2019). (10)

Access to information determines trading in all assets, and the structure of each market affects the speed of information aggregation. In the foreign exchange market, relevant information is dispersed among market participants and key intermediaries need to aggregate it to set prices.

Examples of relevant dispersed information include institutional investors' decisions about the allocation of international portfolios, firms' real-time observations of the state of the economy (e.g., based on their knowledge of certain import and export flows), or changes in risk preferences.

Dealers then learn bits and pieces of disparate information by observing the order flow of their various customers. Customers are not equally informed, and dealers profile them to better learn the information content of their trades. Large banks with a broad base of financial customers are better informed about currency transaction developments than other banks (Bjønnes, Osler, and Rime, 2021; Menkhoff, Sarno, Schmeling, and Schrimpf, 2016; Ranaldo and Somogyi, 2021). It is also recognized that dealers bring their own independent information to the market (Moore and Payne, 2011).

The desire of market participants to manage the disclosure process in the market has likely been a factor behind the recent growth in the number of trading venues and available execution protocols. As we discuss, these trading venues may have significant differences in the extent of counterparty disclosure, the ability to assess a counterparty's price impact, or the ability to match potential counterparties to a particular subset of market participants.

(10) Butz and Oomen (2018) study how dealers adjust bid and ask quotes to affect the direction and speed of client order entry and estimate that the average holding period before an offsetting trade is executed (the "internalization horizon") can be as short as one minute for a liquid pair such as EURUSD. Barzykin, Bergault, and Gueant (2021) derive rules for a dealer's choice between internalization and externalization

Evolution of the trading environment in the foreign exchange market over time

Figure 3 attempts to capture the evolution of the foreign exchange market structure over the past three decades. The top panel of Figure 3a shows a stylized picture of market structure in the early to mid-1990s, Figure 3b in the early to mid-2000s, and Figure 3c in the last decade. The shaded area indicates the inter-dealer market, and the surrounding area indicates the dealer-to-customer market segment. Red arrows indicate voice execution and blue arrows indicate electronic execution.

Currency trading in the early period was often described as a simple two-tier structure (Sager and Taylor, 2006). At the "outer tier," customers traded directly with dealer banks, often by telephone or telex. Dealers were compensated in the form of bid-ask spreads and received private information from their customers' trades (Lyons, 1996). Trading between dealers, either for inventory risk management or speculation, constituted the "inside level." Inter-dealer trading could take both direct (bilateral) and indirect (brokerage) forms, initially only through voice brokers (VBs) (11).

The emergence of two electronic brokers, Reuters (now Refinitiv) and Electronic Broking Services (EBS), in the interdealer market in the early 1990s made the process of currency risk management and price discovery much more centralized and efficient. These two electronic brokers, organized as central limit order books (SLOBs), quickly became the primary sources of price discovery and reference prices for the entire FX spot market, and thus came to be referred to as the "primary market" (we will refer to them as "primary SLOBs" hereafter) ( 12).

Notes: Figure based on King, Osler, and Rime (2012); EB = electronic broker; MDP = multi-dealer platform; ECN = electronic communication network; PB = prime broker; PTF = principal trading firm; SDP = single-dealer platform; VB = voice broker.

The early 2000s opened up new opportunities for electronic trading in the dealer-to-customer segment, as shown in the middle panel (Figure 3b). Multi-dealer platforms (MDPs), arguably the most important innovation of the time, allowed customers to submit a request for quote (RFQ) to multiple dealers**(13)** simultaneously.

MDPs greatly expanded choice for customers and introduced an important element of competition among dealers. In response, a number of individual FX dealer banks invested in their own direct electronic trading platforms (SDPs). Over time, these SDPs have replaced the telephone or telex in most bilateral trading.

The bottom panel, Fig. 3c, provides a stylized view of the current market structure, which continues to grow in complexity. Remnants of the old two-tier structure can be seen, but it has been disrupted by two important innovations. First, the dealer-to-customer segment has seen the emergence of multiple venues offering multiple types of execution protocols. This allowed LPs to provide liquidity to clients by streaming prices or placing limit orders in the order book, which resembles the trading environment of the interdealer market. We will refer to these venues collectively as electronic communication networks (ECNs) or "secondary ECNs"(14). Second, PTFs have challenged domestic banks, both as LPs and as LCs on primary CLOBs. More recently, they have also become active in the external tier, where they transmit prices to clients via ECNs(15). PTFs are thus firmly entrenched in parts of the FX market that were previously the exclusive domain of dealer banks. At the same time, PTFs trade in the foreign exchange market through prime brokerage agreements with dealer banks, including some of the same banks with which they now compete in the market.

(11) In addition to telephone and telex, an electronic messaging platform called the Reuters Direct Dealing System was also widely used for direct dealing. According to 2022 Triennial, 23% of spot trades in the foreign exchange market took place using a direct voice method, such as Bloomberg chat, and another 3% used a voice broker. Voice execution can be particularly advantageous for a client looking to execute a large trade.

(12) EBS Market has become the primary venue for rates such as EURUSD (and its predecessors), USDJPY, EURCHF and, more recently, USDCNH, while Reuters Matching has become the primary venue for "commonwealth" rates such as GBPUSD, USDCAD and AUDUSD, Scandinavian currencies and most EME currencies.

(13) In addition, LC can request a continuous stream of quotes for a specific size and a specific time (e.g., during the trading day), which is called a request for stream (RFS), effectively a continuous form of RFQ. While the number (and names) of platforms have changed over time, early examples of MDPs such as Currenex, Hostpot FX and FXall are still important, and they have often diversified their offerings by execution method. Banks have also responded to the success of third-party MDPs by creating FXSpotStream, which operates as a bank-owned consortium and provides multi-bank FX streaming and RFS services to clients. As for SDPs, at the time of writing this chapter, their examples include J.P. Morgan eXecute, Deutsche Bank AutobahnFX, UBS Neo and Citi Velocity.

(14) At the time of writing this chapter, examples of secondary ECNs include CboeFX, EuronextFX, LMAX, 360T and others. For a complete list of the various trading platforms, we refer the reader to https://www.marketfactory.com/venues/. For convenience, we will continue to refer to platforms with RFQ as the primary MDP execution protocol.

(15) According to Euromoney FX research, examples of client-facing PTF liquidity providers include XTX Markets, Jump Trading, HC Technologies and Citadel Securities.

Market fragmentation and the declining role of the primary market

As shown in Figure 4, trading volume at prime electronic brokers has declined significantly since the global financial crisis, even as overall volume in the FX spot market has grown. The emergence of a large number of alternative trading platforms and the aforementioned decline in inter-dealer FX imbalance trading likely contributed to this decline. In addition, some cite the opening of primary CLOBs to PTFs (via PB agreements) as another factor. Some participants do prefer to trade on venues with less PTF participation, perhaps fearing that their order flow may be quickly detected by PTFs and that they may be subject to more unfavorable selection.

Notes: Primary CLOBs refer to EBS Market and Refinitiv Matching wholesale electronic brokers. In the latter part of the sample, total volume may include other EBS platforms such as EBS Direct. Exchange-traded futures volume primarily reflects volume from the Chicago Mercantile Exchange, as well as two smaller exchanges.

Despite the decline in trading volume, the major CLOBs are still viewed by most market participants as a key price-setting center, and market data generated by the EBS Market and Refinitv Matching platforms remains important to the market as a whole. Moreover, when volatility increases sharply or market liquidity on other platforms deteriorates, FX trading volume often returns to EBS and Refinitiv CLOBs (Moore et al., 2016). This was the case, for example, in March 2020, when financial markets suffered a shock at the start of the COVID pandemic.

Some LPs are increasingly turning to exchange traded currency futures as reference prices and to hedge their spot activities. This seems to be especially true for PTFs, many of which have entered the FX market with an established business model in the futures markets. However, a growing number of market participants of all types seem to view currency futures traded on the Chicago Mercantile Exchange (CME) as at least a close relative of primary CLOBs. Unlike primary CLOBs, the trading volume of currency futures has not diminished. It now often exceeds the volume of OTC spot currency trading in the primary market (Figure 4).

Figure 5 shows the total FX spot electronic trading volume over the last four three-year studies. Multilateral platforms (CLOBs, ECNs, MDPs) are highlighted in dark blue, and direct electronic trading via SDPs is highlighted in light blue.

Notes: Primary CLOBs refer to EBS Market and Refinitiv Matching wholesale electronic brokers (in the latter part of the sample, other EBS platforms such as EBS Direct may be included in the total). CLOB = central limit order book; ECN = electronic communication network; MDP = multi-dealer platform; SDP = single-dealer platform.

Also shown is the declining trading volume on primary CLOBs (which are categorized as CLOBs/ECNs/MDPs). Recent growth in the dealer-to-customer segment has been driven by multi-dealer platforms, mainly at the expense of SDPs, and this also reflects the growth of ECNs offering different trading protocols as market participants can "store around" in search of the best execution.

Illustrating the extent of market fragmentation, Drehmann and Sushko (2022) show that clients can access liquidity through more than 30 MDPs and other ECNs, more than 20 SDPs, and through price feeds from more than a dozen PTFs.

Faced with a growing number of trading platforms and a wide variety of execution methods (such as RFQs, streaming and CLOBs), some LPs are also turning to liquidity aggregators. Liquidity aggregators are technology companies that help their clients access and compare different ECNs, MDPs and SDPs in real time to get the best prices and maximize execution quality.

Comparing the complex execution space

Figure 6 presents a stylized taxonomy of the various electronic trading platforms currently used in the spot FX market. The classification of these venues and their associated trading mechanisms is based on two important parameters: the level of pre-trade anonymity of counterparties and the level of "hardness" of liquidity offered on the trading venues. This taxonomy oversimplifies an increasingly complex market structure and probably misses some important exceptions, but it is a useful way of looking at the trading landscape.

Notes: A stylized taxonomy of e-commerce mechanisms in terms of pre-trade anonymity and last look capability. Note that the last look capability may or may not be implemented by LPs. The primary CLOBs refer to the wholesale electronic brokers EBS Market and Refinitiv Matching. CLOB = central limit order book; ECN = electronic communication network; MDP = multi-dealer platform; SDP = single-dealer platform; RFQ = request for quote; RFS = request for flow.

As shown by the horizontal arrow, the level of pre-trade anonymity can range from fully disclosed to fully anonymous. Trading on an SDP is, almost by definition, fully disclosed, since a single LP knows the identities of all potential LCs active on its platform. Similarly, on an MDP, where an LC submits a request for proposal to multiple LPs, the identities of the potential counterparties are known before the actual trade takes place.

At the other end of the spectrum, primary CLOBs are pre-trade completely anonymous, in the sense that the displayed orders do not reveal the source, and all participants can be matched based on typical CLOB price and time priority rules**(16).**(16).** Note that the futures exchange on the side offers the highest possible level of pre-trade (and post-trade) anonymity, almost by definition, since the exchange is the counterparty to every trade.

Secondary ECNs, which occupy a middle ground, offer their participants neither full pre-trade anonymity nor full pre-trade disclosure. This may be done, for example, by assigning an alphanumeric "tag" to each counterparty on the platform, which is a common practice on some ECNs. This may allow a participant, by observing the quality of trade execution against a counterparty with a particular label, to decide whether to trade with that unnamed counterparty again. In addition, some other ECNs have created multiple separate CLOBs, each limited to a specific pool or tier of participants with a common trading style or set of characteristics. In this case, although traders on each CLOB do not know the identities of their counterparties in advance and cannot follow them over time, they are guaranteed to be matched only with a particular type of counterparty.

The second dimension in Figure 6 is the "firmness" of the liquidity offered on the electronic platform, that is, the level of certainty of execution that follows a request to trade on a posted quote. This depends, importantly, on whether LPs use the practice known as "last look" or not.

Last look is the process by which an LP, or more specifically, an LP's computer, upon receiving a request to trade on one of the posted quotes, takes a "last look" at the trade request before deciding whether to proceed with the trade. The last look process occurs over a period of time measured in milliseconds (Oomen, 2016; Cartea, Jaimungal, and Walton, 2019).

The last look method was developed to address a number of potential problems arising from delays in communication networks, multiple electronic platforms, and counterparties with different levels of technological sophistication. The use of last look has become widely acceptable for conducting quasi-permanent "validity and price" checks, including verifying that there is sufficient bilateral credit and that the quoted price has not become obsolete. Other uses of last look, including setting an "additional waiting time" before accepting or rejecting a transaction, are much more controversial (see discussion in Section 5.1).

As shown in Figure 6, last look is not allowed on primary CLOBs, so liquidity there is considered "firm". In contrast, last look is widely used on secondary ECNs, with some exceptions, and it is ubiquitous on MDPs and SDPs, at least in the sense that LPs always reserve the option to use it. Liquidity on these platforms is therefore seen as less stable: there is less certainty that a "hit" or request to trade on a quote will result in a trade.

Putting the two dimensions in Figure 6 together, it is clear that higher levels of pre-trade anonymity are associated with higher liquidity. The anonymity and certainty of execution in the auction structure make the pre-trade environment on primary CLOBs the closest to a full-fledged exchange. This is probably one of the reasons why, despite lower trading volumes, prices on primary CLOBs are still considered benchmark prices for the entire FX market.

(16) Nevertheless, reflecting the OTC nature of the FX spot market, participants in primary CLOBs can only trade with each other if they have established bilateral credit relationships, either directly or through their PBs. In practice, most participants in primary SLOBs can trade with each other.

Algorithmic trading in the foreign exchange market

The spread of electronic execution in the spot foreign exchange market, which currently accounts for about 75% of even dealer-to-customer trading (BIS, 2022), has been accompanied by the rise of algorithmic (computer-driven) trading.

A significant proportion of both LPs and LCs now use algorithms to manage at least part of their trading activity with limited "human" intervention. The expansion of algorithmic trading in the FX market occurred first on the two main CLOBs, but algorithmic trading has now expanded well beyond the main market (17).

Banks and non-banking organizations use different algorithms in the foreign exchange market. Execution algorithms attempt to minimize the price impact of large trades, market-making algorithms automatically publish executable quotes, and opportunistic algorithms monitor the market for profit or hedging opportunities. An important distinction between these algorithms is whether they only automate trade execution or whether the decision to trade is also left to the algorithm.

Figure 7 shows the trend in the proportion of "manual trading" and "API trading" on the EBS Market platform since 2004, illustrating the growth of algorithmic trading on one of the major CLOBs. At the beginning of the sample, only banks were allowed on the platform and all trading was done manually by entering trade instructions on the specialized keyboard of the EBS terminal. After EBS allowed direct interface of computers with its platform, banks started trading algorithmically in 2004 (bank API). Non-banks started in 2005 (non-banking API) when they were first admitted to the EBS platform through PB arrangements.

Notes: The figure shows the share of trading volume in the EBS market conducted manually and algorithmically by banks and non-banks. Sources: EBS.

Note that overall, the vast majority of non-bank API trades reported currently represent PTF activity, primarily using algorithms that automate both trade decision and execution. In contrast, much of the activity of bank APIs likely reflects trades where execution is automated but the initial decision to trade is made by a human (18).

As shown in Figure 7, algorithmic trading will dominate by the end of the sample, in 2022. Bank and non-bank APIs account for just over 40% of the total trading volume, and these shares have remained fairly even since 2015. In contrast, the share of manual trading has continued to decline and now accounts for only about 15% of trading volume**(19).**

As discussed above, banks and non-banks regularly use multiple types of algorithms to provide and consume liquidity. Among these are "delay arbitrage" algorithms, which are a somewhat infamous subtype of opportunistic algorithms that exploit a firm's speed advantage to generate profits**(20)**. Delay arbitrage algorithms are thought to be used almost exclusively by a subset of PTFs active in the market.

In an attempt to protect their clients from such strategies, especially on platforms where manual and algorithmic traders coexist, several trading platforms in the FX market have introduced various types of "speed limiters". ParFX, a platform launched in April 2013 by a group of large bank-dealers, added random pauses (20 to 80 milliseconds at the beginning) to all trade instructions before processing them. Later in 2013, EBS Market introduced "floor delays" of a few milliseconds - short periods during which all trade messages are sorted and randomized before hitting the order book. In 2014, Reuters/Refinitiv Matching, another major CLOB, introduced its own version of a speed limit (21).

Finally, while execution algorithms have long been used by banks to optimize the execution of large trades, such algorithms are now also being used directly by some of the most sophisticated clients in the FX market. According to a recent report on the use of execution algorithms in the FX market (Bank for International Settlements, 2020), sophisticated clients are increasingly relying on intelligent order routing and execution algorithms to distribute large orders over time and across multiple electronic venues. FX clients are also increasingly using transaction cost analysis (TCA) to monitor the execution quality of their trades.

(17) Due to the role that primary CLOBs play in the market, algorithmic trading has had a significant impact on price discovery in the FX market (Chaboud, Chiquoine, Hjalmarsson, and Vega, 2014).

(18) For example, a bank may decide to cover part of the order flow on its own SDP by trading on the primary CLOB using an execution algorithm.

(19) The evolution of these shares over time has some notable features. For example, manual trading falls in early 2020 when the COVID pandemic begins, likely reflecting the rise in the number of people working from home, bank algorithmic activity rises from late 2014, likely due to a number of banks switching to algorithmic execution around fixings following the fixing scandal (see Section 6.2), and non-bank algorithmic activity peaks in 2011 when EBS reduces tick size (Chaboud, Dao, and Vega, 2021).

(20) For example, a trader with faster technology than a particular LP can take advantage of the small time lag between when a trade occurs in the market and when the LP updates its quotes in response by trading quickly on one of these now outdated quotes.

(21) The "speed limiter" introduced by Refinitiv was seen as an evolution of the EBS "speed limiter." Specifically, Refinitiv's mechanism processed order cancel messages ahead of other types of messages when each batch was released to the order book, providing additional protection against aggressive delay arbitrage strategies. This feature was later adopted in EBS as well.

Role of the formal sector

Regulation, best practices, Global FX Code

The level of regulation of the foreign exchange market varies widely across countries, depending primarily on the exchange rate regime and restrictions on cross-border capital flows (22). The regulatory authority may be under the central bank, ministry of finance, or other government agency.

In the case of major floating exchange rates (often referred to as G-10 exchange rates), regulatory constraints tend to be low compared to equity and bond markets. This is probably largely due to the fact that the foreign exchange transaction does not involve the exchange of securities and that, by its nature, it involves the currencies of two different countries.

Even if central banks do not have formal authority to regulate the foreign exchange market, many of them sponsor Foreign Exchange Committees, which provide an opportunity for various market participants to meet regularly to discuss market functioning and promote "best practices." In addition, central banks of countries with major trading centers meet regularly under the auspices of the BIS to discuss the state of the global foreign exchange market.

It is within this framework that a broad group of central banks and market participants jointly developed the Global FX Code, first published in 2017 and updated in 2021. The Global FX Code, which is supported by the Global FX Committee (GFXC), lists and discusses principles of good practice for the global FX market (23). Specifically, the Code contains 55 principles of good practice in areas such as ethics, governance, execution, information sharing, risk and compliance, and confirmation and settlement. Market participants are invited to sign a "statement of commitment" to the Code to publicly declare their intention to adhere to its principles. The Code has gained wide international acceptance, particularly among large dealer banks, and is increasingly seen as having an important influence on the behavior of foreign exchange market participants.

For example, the 2021 update of the Code was accompanied by guidance on the appropriate use of last look. The guidance confirmed that last look should only be used for price and validity checks and recommended that these checks should be applied "without delay". Market participants viewed this recommendation as a statement against the widespread but controversial practice of some liquidity providers waiting to observe additional price movements before accepting or rejecting a trade. While LPs argue that this "extra holding time" protects them from potential adverse selection, opponents point out that it allows LPs to accept only the most profitable trades. As a result, the GFXC's recommendations had a swift and significant impact on market participant behavior, with many major bank-dealers announcing a waiver of the additional hold time and some trading venues announcing a reduction in the maximum duration of the "last look window."

In summary, although still a fairly new experiment, the impact of the Global FX Code on the FX market is seen by some as evidence that a "soft" regulatory approach that encourages best practice rather than imposing rigid rules can have a significant impact on an important financial market (24).

(22) The IMF's Annual Report on Exchange Arrangements and Exchange Restrictions is a comprehensive source of information on this topic.

(23) The GFXC Code was established in 2017 and its members include various FXCs from around the world. The Code is available on its website: http://www.globalfxc.org.

(24) Skeptics may argue that a "soft" approach to regulation is only effective because deviation from common practice may result in civil legal liability for the financial firm.

Currency intervention and exchange rate management

The official sector, mainly central banks and ministries of finance, often exerts a decisive influence on floating exchange rates (25). Participants in foreign exchange markets closely follow the release of official macroeconomic data and central bank announcements, and major exchange rates typically react to the unexpected component of this news in milliseconds.

In addition, a more direct and deliberate way for the official sector to influence exchange rates is foreign exchange intervention - the buying or selling of currency in the foreign exchange market with the specific intention of affecting exchange rates.

Many large industrial countries with floating exchange rates, including the United States, resorted to foreign exchange intervention quite frequently until the mid-1990s (26). Since then, with a few notable exceptions, including Japan in the early 2000s and Switzerland in the years following the global financial crisis, currency intervention by large industrial countries has become rare (27). In contrast, currency interventions by developing and newly industrialized countries have remained fairly frequent (28).

A distinction is often made between "sterilized" and "unsterilized" currency interventions, where sterilization of an intervention operation means offsetting the associated impact on the domestic monetary base, usually by buying or selling bonds. Historically, most foreign exchange intervention operations have been sterilized. By their nature, unsterilized foreign exchange interventions are as much monetary policy as exchange rate policy (29).

A substantial academic literature has discussed the effects of currency interventions. Earlier work focused on interventions by large industrialized countries (Dominguez and Frankel, 1993), but has recently begun to examine the effects of currency interventions on a broader range of countries (Fratzscher, Gloede, Menkhoff, Sarno, and Stohr, 2019), including countries that employ "managed floating" for their currencies (Frankel, 2019; Cavallino, 2019).

Broadly speaking, studies have traditionally considered three channels through which foreign exchange interventions can affect exchange rates: the portfolio balance channel, where investors view domestic and foreign assets as imperfect substitutes; the signaling channel, where foreign exchange interventions are interpreted by the market as conveying information about future central bank policy; and the coordination channel, where intervention activities influence the trading behavior of currency traders in a way that amplifies central bank activity

It is widely believed that foreign exchange interventions, especially large interventions, can have an immediate impact on exchange rates, but there is no consensus on the medium- to long-term impact of foreign exchange interventions on exchange rates. In general, studies have shown that announcing a currency intervention, conducting the intervention in a direction consistent with monetary policy, and coordinating the intervention among central banks can enhance the short-run impact of currency intervention.

In addition, the impact of currency intervention on currencies other than major currencies is often larger and longer lasting. This suggests that currency intervention may have a greater impact in the presence of market imperfections, such as limited intermediation capacity (Gabaix and Maggiori, 2015) and several other financial constraints that make domestic and foreign assets imperfect substitutes (Popper, 2023).

(25) Clearly, the official sector also plays an important but different role when countries decide to peg or tightly regulate the exchange value of their currency.

(26) The intervention episodes that followed the 1985 Plaza Accord, an agreement among G5 countries to weaken the dollar, and the 1987 Louvre Accord, a G7 agreement to stabilize dollar exchange rates, are particularly well known in that era. Several European countries also intervened heavily in 1992 when the European Exchange Rate Mechanism (ERM) came under pressure, especially shortly before Britain's withdrawal from the ERM.

(27) Japan has intervened in foreign exchange markets several times since the 2000s, including in 2011 after the earthquake and nuclear accident and, most recently, in 2022.

(28) A growing number of central banks in emerging markets are also intervening in currencies using derivatives, allowing them to save on foreign exchange reserves.

(29) A case in point is the SNB's recent intervention purchases of foreign exchange. In this case, the SNB specifically chose not to sterilize its intervention operations, which resulted in a significant increase in the size of its balance sheet, essentially quantitative easing through foreign exchange purchases.

Special topics

Outbreaks and other extreme events

Like some other financial markets with a large share of electronic trading, the foreign exchange market has experienced a number of "flash events" in recent years, when very sharp but often quickly recoverable movements in exchange rates were accompanied by a temporary but almost complete disappearance of market liquidity. Analysis of these events can provide insight into the behavior of market participants and the impact of certain aspects of the market structure.

Recent events include the following: On October 7, 2016, the British pound sterling depreciated rapidly by almost 9 percent against the U.S. dollar and then quickly recovered most of its losses within minutes (Bank for International Settlements, 2017); then on January 3, 2019, the Japanese yen appreciated sharply against the U.S. dollar and several other currencies and then quickly recovered most of its movement.

These two events have some interesting features. Both occurred in the late afternoon in the US (early morning in Asia), a time of day when the FX market typically sees the lowest trading volume and liquidity. In both cases, the linkages between the spot and futures market appear to have been at work. The liquidity of the British pound fell to its lowest level precisely when CME futures contracts halted trading in response to a strong but still fairly orderly price movement. The JPY event occurred at a time of day when there is no trading in CME futures (30). These two flash events also have a number of individual characteristics. In particular, the yen event was associated with retail currency investors dumping positions in higher-yielding currencies such as the Turkish lira and the Australian dollar (31).

Finally, what is considered to be the most extreme event in the foreign exchange market in recent years was not in itself a flash event, as it had a clearly recognized trigger. On January 15, 2015, the SNB removed the floor it had set for the EURCHF exchange rate since 2011, which effectively prevented the Swiss franc from appreciating (Breedon, Chen, Ranaldo, and Vause, 2022). The SNB maintained this floor through regular interventions in the foreign exchange market, buying euros in quantities sufficient to keep the exchange rate at or above CHF 1.20 per euro.

Market participants did not expect the floor to be lifted (Mirkov, Pozdeev, and Soderlind, 2106), and the Swiss franc quickly appreciated to 40 %¨ against the euro, forcing the SNB to resume intervention purchases of euros about 20 minutes later.

The Swiss franc ended the day about 15% stronger against the euro. The impact of the associated volatility on leveraged currency positions led to a widespread review of risk management practices in the FX industry.

(30) FX futures on the CME are traded 23 hours a day, five days a week. Trading halts for one hour between 5:00 p.m. and 6:00 p.m. Eastern Time.

(31) The retail FX sector in Japan is larger than in other developed countries and has a significant number of retail margin traders. When the yen suddenly appreciates in value, the currency movement can be amplified by triggering stop-loss orders (Tomohiro Niimi, 2016).

Base exchange rates and the "fixing scandal"

Benchmark exchange rates, also commonly referred to as "fixings", are calculated and published many times a day. They are widely used for a variety of purposes, including the valuation of portfolios containing securities denominated in foreign currencies, the construction of multi-country indices such as the MSCI stock indices, and even the settlement of some derivatives transactions.

The best known and most widely used of these benchmark rates is the rate calculated by WMR at 4:00 p.m. in London (32). Because some important FX market participants, such as asset managers, regularly request trades executed at the WMR 16:00 rate, the FX market experiences a significant daily spike in trading volume at this time of day. Particularly high volumes occur at the end of each month due to the rebalancing of multi-country portfolios. There are other significant daily fixings in the FX market, including those calculated by the ECB at 14:15 CET and the morning Tokyo fixings calculated by various Japanese banks at 9:55 am (Financial Stability Board, 2014; Ito and Yamada, 2017).

Prior to 2013, the role of these fixings, with few exceptions (Melvin and Prins, 2015), attracted very little attention outside the industry. This changed after allegations emerged in 2013 that some traders at large banks may have manipulated underlying exchange rates (Vaughan, Finch, and Choudhury, June 12, 2013).

The scandal, dubbed "currency fixing," eventually led to significant fines being imposed on dealing banks in a number of countries, and reforms to the fixing process were introduced in 2015 following a Financial Stability Board study (Financial Stability Board, 2014, 2015). Among other reforms, the window over which the 4-hour fixing is calculated was increased from one to five minutes in order to reduce the likelihood of manipulation (33).

Despite the 2015 reforms, the issue of benchmark exchange rates continues to be a concern. In particular, traders point to persistent unusual volatility patterns around 16:00, especially at the end of the month, leading to some calls for further reforms. The issue of benchmark exchange rates has also sparked a number of academic studies that focus on the original question (Evans, 2018; Ito and Yamada, 2017; Osler and Turnbull, 2017) and the impact of fixing reform (Evans, O'Neill, Rime, and Saakvitne, 2018).

(32) WM/Refinitiv (formerly known as WM/Reuters) calculates reference rates for a large number of spot, forward, and NDF currencies every hour.

(33) At the end of 2015, the ECB, in an effort to contain the trading activity associated with its daily reference exchange rates and thereby reduce the likelihood of manipulation, delayed their publication from 14:30 to 16:00 CET.

Settlement risk

Transactions in the "spot" foreign exchange market are, by definition, concluded by exchanging balances in one currency for balances in another. For the vast majority of currency pairs, settlement occurs two business days after the trading date, i.e. on "T+2". Currency settlement risk is the risk of loss incurred by a counterparty when it pays out the currency sold but does not receive the currency purchased.

The bankruptcy of Bankhaus Herstatt in 1974 demonstrated how currency settlement risk can undermine financial stability. Herstatt was a medium-sized German bank active in foreign exchange markets. On June 26, 1974, German authorities closed the bank. Although Herstatt had already received German marks from its counterparties, it had not yet made the corresponding U.S. dollar payments in New York. Herstatt's non-payment forced a number of banks to stop outgoing payments until they received the payments due to them. The consequences were systemic: the international payment system was severely affected for a time, and the erosion of confidence led to a sharp rise in lending rates and a contraction of credit.

In 1996, the central banks of the G-10 countries launched a comprehensive strategy to reduce foreign exchange settlement risk, resulting in the launch of the Continuous Linked Settlement (CLS) system in 2002.

CLS is a specialized institution that settles foreign exchange transactions on a payment versus payment (PvP) basis, eliminating settlement risk by ensuring that a payment in one currency occurs if and only if a payment in another currency occurs.

The creation of the CLS and other measures have led to a significant reduction in currency settlement risk. Kos and Levich (2016) estimate that the share of foreign exchange turnover settled through traditional correspondent banking arrangements fell from 85% to 13% between 1997 and 2013.

In 2013, the share of global FX transactions settled through CLSs reached around 50%, but the adoption of PvP settlement appears to have slowed recently (Bech and Holden, 2019). Among the likely reasons are the increase in trading volume in emerging market currencies that cannot be used by CLS and the internalization of LP client flows, in which case the dealer is likely to settle on its own books.

In recent years, new efforts have been made to mitigate the remaining FX settlement risk in the system, including efforts to introduce PvP settlement in additional currencies. In addition, the updated Global FX Code 2021 encourages market participants to utilize PvP settlement wherever possible.

Source: https://www.bis.org/publ/work1094.pdf