New Perspective on FX Markets:

Order-Flow Analysis

Richard K. Lyons*
Haas School of Business, UC Berkeley
Berkeley, CA 94720-1900
Tel: 510-642-1059, Fax: 510-643-1420
lyons@[Link]

29 June 2001

Abstract

Though fundamental and technical analysis are still widely used in

foreign exchange markets, a new type of analysis has emerged: or-
der-flow analysis. Order-flow analysis uses the flow of buy and sell
orders to both explain exchange rates contemporaneously and fore-
cast future movements. This article contrasts order-flow analysis
with the traditional approaches and reviews lessons learned. Most
important among those lessons is order flows ability to account for
the lions share of movements in the major floating rates. On the
policy front, widespread availability of electronic order-flow data
brings many policy questions within our reach for the first time. Af-
ter reviewing these policy questions, the article closes with a dis-
cussion of how FX market institutions are evolving and how this
evolution will affect application of order-flow analysis in the future.

A new approach has emerged for understanding the behavior of exchange

rates: order-flow analysis.1 Though fundamental analysis and technical analysis
are still widely used, practitioners, policy-makers, and academics are increas-
ingly using order-flow analysis to complement these other approaches. My thesis
is that this shift in perspective is well founded: there is now considerable
evidence that order flow accounts for the lions share of floating exchange-rate
movements.
* Forthcoming, International Finance. I thank two anonymous referees for their insightful comments and
the National Science Foundation for financial support.
1 Order flow is not synonymous with trading volume. Order flowa concept from microstructure

financerefers to signed volume. Trades can be signed in microstructure models depending on whether
the aggressor is buying or selling. (The dealer posting the quote is the passive side of the trade.) For
example, a sale of 10 units by a trader acting on a dealers quotes is order flow of 10, though volume is
10.
 To put order-flow analysis in perspective, the next section provides back-
ground on the more traditional approaches, fundamental and technical. The
following section introduces order-flow analysis, contrasts it with the traditional
approaches, and reviews lessons learned from this new type of analysis.2 I then
turn to the policy implications of order-flow analysis; the widespread availability
of electronic order-flow data allows us to address certain policy questions for the
first time. The paper closes with a discussion of the FX markets future. FX
market institutions have changed considerably over the last 10 years (e.g., the
major role now played by electronic brokers), and the pace has not slowed. To
what extent might these near-term changes render order-flow analysis obsolete?
In fact, the approach can be successfully applied even if the structure does
change. This final section explains why.

1. A Little Background

Since the 1970s, fundamental analysis has viewed (nominal) exchange rates
as determined by a set of macroeconomic variables that includes interest rates,
money supplies, inflation rates, GDPs, government budget deficits, and current
account balances (defined for both the home and foreign countries). Though
conceptually sound, the approach has not fared well empirically. In particular,
the macro variables that underlie the approach do not move exchange rates as
predicted. The classic reference is Meese and Rogoff (1983); they show that
macro-fundamental models fail to account for major-currency exchange rates
better than a simple no change model. Thus, the macro-fundamental models are
not even consistently getting the direction right. In his survey, Meese (1990)
summarizes by writing, The proportion of (monthly or quarterly) exchange rate
changes that current models can explain is essentially zero. 3 In a later survey in
the Handbook of International Economics, Frankel and Rose (1995) comment on
fundamental analysis this way:

To repeat a central fact of life, there is remarkably little evidence

that macroeconomic variables have consistent strong effects on
floating exchange rates, except during extraordinary circumstances
such as hyperinflations. Such negative findings have led the profes-
sion to a certain degree of pessimism vis--vis exchange-rate re-
search.

From the fundamentals perspective, there is clearly room for some fresh think-
ing.
2 For practitioner-oriented research using order-flow analysis, see, e.g., Citibanks Citiflows Global Flow
and Volume Analysis (various issues), Deutschebanks Flowmetrics Monthly (various issues), and
Lehman Brothers Global Economic Research Series, particularly the issue on FX Impact of Cross-
Border M&A. For evidence from practitioner surveys, see Gehrig and Menkhoff (2000). It is noteworthy
thatunlike fundamental and technical analysisorder-flow analysis is not available to everyone: one
needs sufficient order-flow data.
3 The literature documenting this poor empirical performance is vast; for surveys see Frankel and Rose

(1995), Isard (1995), and Taylor (1995). Macro-fundamental models do perform better over longer
horizons, e.g., horizons of 3-5 years (see, e.g., Mark 1995).

2
 The other traditional approach, technical analysis, attempts to identify fu-
ture exchange-rate movements using only the pattern of past prices. Unlike
fundamental analysis, which is widely used in all three circles noted above
(practitioners, policy-makers, and academics), technical analysis is used mostly by
practitioners. This is largely because technical analysis is targeted at forecasting
rather than explanation (where by explanation I mean identification of the
concurrent variables that drive exchange rates). Empirically, there is some
evidence that technical rules have predictive power for exchange rates (see, e.g.,
Levich and Thomas 1993 and Chang and Osler 1998). Though statistically signifi-
cant, however, no one has ever argued that this predictive power accounts for
more than a small percentage (<10%) of exchange-rate variation.

2. Order-Flow Analysis

Within the academic literature, order-flow analysis is typically referred to

as microstructure analysis. The term microstructure comes from the field of
microstructure finance, which is concerned with, among other things, the role
that order flow plays in impounding information in price. Unlike macro models of
exchange rates, which assume that all traders share the same information and
beliefs, microstructure models recognize that individuals use different informa-
tion in forming their beliefs. In a context where individuals use different informa-
tion to form their beliefs, the market needs a means of measuring those beliefs.
Order flow plays this role. In these models, it is precisely when a traders beliefs
differ from the market enough to put money on the tablea tradethat those
beliefs warrant counting. Think of the orders like votes. The market performs the
service of tallying the order-flow votes, and setting the market price on the basis
of the tally.
Figure 1 provides an illustration. In fundamental analysis, the mapping
from fundamentals to price is assumed to be direct and immediate. In order-flow
analysis, information flows through orders on its way to price. More recent,
hybrid analysis, allows for both channels to affect price.4
What type of information is the order flow conveying? (Or, if you prefer:
What is driving the order flow?) This is the right questionorder flow is a
proximate cause, not an underlying cause. There are many examples of dispersed
information that needs to be aggregated for pricing assets: differential interpre-
tation of news, shocks to hedging demands, shocks to liquidity demands, time-
varying risk tolerances (of financial institutions, for example), and so on. Empiri-
cally, we have not determined conclusively which of these information types are
the most important. What we do now know is that order flow is important for

4 Though figure 1 focuses on the complementarity between fundamental and order-flow analysis, there
is also an emerging complementarity between technical and order-flow analysis. Osler (2001) in
particular makes this case. Using data on stop-loss and take-profit orders in FX, she shows that
clustering of these orders at particular prices helps to explain two familiar predictions from technical
analysis, namely that (1) trends tend to be reversed at support and resistance levels and (2) trends tend
to gain momentum if support and resistance levels are breached.

3
how exchange rates are determinedit is the transmission mechanism. Now
that we know how, we are in a better position to learn why. 5
Implicit in the last paragraph is the point that fundamental analysis and
order-flow analysis differ in terms of research strategy. Order-flow analysis
starts from the meeting of demand and supply and proceeds to identify the (more)
exogenous variables behind that order flow. Going to the micro source provides
exchange-rate theory with much needed empirical guidance.6 A simple strategy
that has already made progress along these lines is based on our ability to break
order flow into parts. (That it can be decomposed is one of its nice properties.)
One can test whether all parts of the aggregate order flow have the same price
impact. They do not: the price impact of FX orders from financial institutions
(e.g., mutual funds and hedge funds) is significantly higher than the price impact
of orders from non-financial corporations (Lyons 2001). This suggests that order
flow is not just undifferentiated demand. Rather, the orders of some participants
are more informative than the orders of others. Analyzing order flows parts
illuminates the information structure underlying this market.
Let me provide a brief review of what we have learned from the applica-
tion of order-flow analysis. The lessons learned thus far can be divided into two
broad groups: those that are more macro-oriented and those that are more micro-
oriented. Because the approach is still young, more work is required before these
lessons can be considered stylized facts. As data sets covering longer time periods
become available, these lessons surely will be refined.
First consider the more macro-oriented lessons. Six lessons in particular
seem especially important:

Order Flow Drives a Large Share of Longer-Horizon Price Movements.

Even when based on data sets that include only a fraction of market-wide
flow, the concurrent impact of these flows accounts for 40-70 percent of the
persistent movements in exchange rates (i.e., at monthly horizons and
longer, see e.g. Payne 1999, Evans and Lyons 1999, Evans 2001, Rime 2000).
As richer order-flow data sets become available (e.g., as they span a larger
share of the market and sign the flows more precisely), that percentage
may rise still higher.

Even Macroeconomic Announcements Affect Price via Order Flow. The

flipside of the first lesson is that concurrent macro announcements and
other readily identifiable macro changes do not directly explain a large
share of longer-horizon price movements (i.e., after controlling for order
flow; see Evans 2001, Evans and Lyons 1999). Rather, order flow appears to

5 Some skeptics argue that the order flow driving exchange rates is irrational, and therefore does not
represent information. Well, irrationality is always a possibility. But even if part of the order flow is
determined irrationally, if this part is affecting prices, then it is because rational players view offsetting
these orders fully as being too risky. In this case, the resulting effect on prices is a portfolio-balance
effectthe adjustment in price is exactly that required to induce the rational players to step in. Even in
this case, then, order flow is conveying information about equilibrium risk premiums.
6 As suggested below, models of market incompleteness of various types are a natural fit with current

empirical results. See, e.g., Duarte and Stockman (2001) for a contact point within the theoretical
literature on dynamic open-economy models.

4
 mediate most price movements. Though in some sense this result is a re-
discovery of the well-known empirical failure of macro-fundamental mod-
els, work using order-flow analysis is clarifying the factors that supplant
concurrent macro variables as the driver of prices. An additional point,
distinct from the first, is that effects from major announcementsdirect or
notdo not account for a large share of exchange-rate volatility (see
Anderson and Bollerslev 1998).

The Elasticity of Price with Respect to Order Flow is High. The elasticity
of the exchange rate with respect to order flow by non-dealer customers is
roughly 0.8 percent per $1 billion (in the largest markets; see Lyons 2001,
Table 9.2). With world financial wealth measured in trillions of dollars,
this is puzzlingly high. The result is consistent with a common view that
Milton Friedmans stabilizing speculators are not bold enough. Why this
boldness might be lacking remains an open question. On the other hand,
from an information-theoretic perspective high elasticity may not be so
puzzling: small net flows may be conveying significant amounts of informa-
tion.

Order Flow is a Factor in Floating-Rate Volatility. We now have substan-

tial evidence that order-flow is an important proximate factor driving vola-
tility, and may account for apparently excessive volatility under floating
regimes (see, e.g., Evans and Lyons 1999, Killeen et al. 2000). Though work
on order flow as a driver of price is focused on the sign of the relationship,
there are also implications for volatility: a good model of return first mo-
ments is a good model of return second moments (but not vice versa). Also
relevant for this lesson is the work of Osler (2001). She finds that stop-loss
orders on the buy (sell) side tend to cluster at prices just above (below)
round numbers, which can cause trends to gain momentum once support
and resistance levels are crossed.

The Accelerationist View is Not Supported. Empirical results in FX mar-

kets are not consistent with what is sometimes called the accelerationist
viewthat order flow simply impounds information in price a few min-
utes faster than would otherwise have been the case. Under the accelera-
tionist view, the asset-pricing consequences of order flow are not that in-
teresting: price paths that differ only by a time-shift of a few minutes are
equivalent to a macroeconomist. If the accelerated-by-a-few-minutes view
were true, then one would expect public information flow to account for as
much exchange-rate variation as order flow. As noted above, however, pub-
lic information flow is virtually uncorrelated with the direction of ex-
change-rate movements (at horizons of one year or less).
At the same time, a fascinating possibility is that an acceleration-
ist story is indeed operating, but over much longer horizons. Suppose the
information conveyed by order flow reflects changing market expectations
about macro-fundamentals that are more distant (i.e., beyond the next
month, quarter, or even year). In that case, order flow serves to telescope
these forward-looking fundamentals into todays spot rate. Note too that

5
 this possibility is consistent with findings that over longer horizons (e.g.,
three to five years), macro variables do begin to account for a substantial
share of exchange-rate variation (despite concurrent macro-fundamentals
being virtually uncorrelated with exchange rates; see, e.g., Mark 1995).7

Order Flow Does Not Have to Sum to Zero. The Evans and Lyons (1999)
model shows why order flow between dealers does not have to sum to zero.
Conceptually this is important, because many people are under the mis-
taken impression that order flow must sum to zero (and therefore that any
flow measure that correlates positively with price must be unrepresenta-
tive in some way). This is not always the case.

Let me turn now to more micro-oriented lessons. By micro-oriented, I

mean that they are based on intraday analysis of individual dealers. As such,
they parallel more closely bread-and-butter work within microstructure fi-
nance.

Order Flow is Private Information. The behavior of individual dealers

shows that they consider FX order flow to be a source of private informa-
tion, and that they set prices accordingly (see, e.g., Lyons 1995). This em-
pirical result at the micro level accords well with the above-noted impor-
tance of order flow at lower frequencies. Moreover, all orders are not alike
in terms of their information content. Identifying which orders are the
most informative, and who is behind them, is helping to decipher the mar-
kets underlying information structure.

Dealer Inventories Affect Price. Inventory control among spot FX dealers

is strong relative to that found for other markets. Most spot FX dealers
prefer to end their trading day flatthat is, with no net position. Accord-
ingly, the half-life of the typical dealers inventory is significantly less than
one day, and has been estimated to be as low as ten minutes (e.g., Lyons
1998). These half-lives are much shorter than those found in equity and fu-
tures markets, where half-lives longer than one week are common.8 Not
only do FX dealers control their inventories intensively, some also adjust
their prices to induce inventory-decumulating order flow. (Lyons 1995
finds these inventory effects on price, but Yao 1998 does not find them for

7 This conjecture circumvents a misleadingly compelling argument why order flow cannot be conveying
macro fundamentals. That misleading argument starts by supposing that the R-squared statistic from
a regression of exchange-rate returns on order flow is one. Because the R-squared statistic from
regressing exchange-rate returns on macro variables is nearly zero, the argument goes, order flow cannot
be picking up macro information. The shortcoming in the argument is that regressions using current and
past macro variables very likely provide poor measures of expected future macro paths.
8 Resolving these differences may lie in the fact that non-FX marketmakers hedge inventory risk with

instruments other than those in which they make the market (e.g., with related derivatives), whereas
spot FX dealers find that inventory control using spot currencies alone is least expensive. See Naik and
Yadav (2000) for evidence that non-FX marketmakers do indeed use derivatives for inventory control.
Another possible resolution of these differences lies in the obligation of the NYSE specialist to smooth
prices, a task which existing inventory may facilitate.

6
 the dealer he tracks.) This finding of inventory effects on price is impor-
tant: these effects are the linchpin of the whole inventory branch of micro-
structure theory, despite the fact that empiricists working on markets
other than FX have not found them.9

Hot Potato Trading Contributes to Trading Volume. Hot potato trading is

the passing of positions from dealer to dealer in the process of interdealer
risk management. Dealers describe this type of trading as an important
source of the FX markets enormous trading volumes. The large share of
trading between dealers that we find in FX relative to other markets is
consistent with a significant role for hot potato trading. On the theoretical
front, our models show that hot potato trading is consistent with optimiz-
ing behavior (e.g., Lyons 1997). On the empirical front, we also find direct
evidence that hot potato trading is present (e.g., Lyons 1996).

Having reviewed what we have learned, let me close this section with
some thoughts on what we still need to know. First, we need to determine why
the price impact of order flow from different customer types is so different, which
links to the nature of the underlying information. Second, we need to determine
which components have the most out-of-sample forecasting power and at what
horizons. (That they have forecasting power is not a violation of efficient markets
because these data are not publicly available.) An answer to the forecasting
question should help us answer the first question about the nature of the underly-
ing information. Third, we need to close the gap between order-flow analysis and
fundamental analysis. Are order flows conveying changing expectations about
future macro paths? Does order flowor parts of order flowlink to balance-of-
payments flows of various types? These are some of the larger open issues that
researchers in this area are poised to address.

3 Policy Implications

There are five broad areas where I envision order-flow analysis having im-
pact on policy. I introduce each of these with an eye toward future work that is
likely to be useful. Though in some areas there is already a basis for specific
recommendations, in other areas recommendations will have to wait for further
analysis of these policy questions.

Policy Area 1: The Price Impact of Order Flow

9 In an empirical analysis of NYSE specialist trading, Madhavan and Sofianos (1997) find that
specialists tend to manage inventory by strategically timing their trades, rather than through adjusting
their own prices.

7
 It would be useful for official institutions to begin collecting data on FX
order flows.10 This is likely to be particularly valuable for policy-making in
developing countries. Specifically, order-flow data allow one to quantify the price
impact of currency trades (both transitory and persistent), as demonstrated
repeatedly in existing empirical work. This is a direct measure of market liquid-
ity: price impact is what liquidity is all about. The lower the price impact, the
higher the liquidity, other things equal. It would be interesting to get a sense for
how price impact in developing-country markets changes as a function of the
state of the market (devaluation likelihood, etc.). Also, one could determine
whether customer forward trades have the same price impact as customer spot
trades of similar size. If not, one could quantify the difference. (Many developing
countries restrict or even forbid forward trading on the belief that such trading is
more speculative in nature than spot trading and is therefore more destabiliz-
ing.) One might also compare price impact across countries, in an effort to
determine which institutional structures are better at promoting liquidity.
The question of price impact is related to the issue of stability. Policymak-
ers in some developing countries appear to believe that additional liquidity is
destabilizing. In theory, it is less liquidity that is destabilizing, not more liquidity:
the less the liquidity, the larger the price impact, and the more prices move,
other things equal. To make the case that other things are not equal, in a way that
might reverse the relationship between liquidity and stability, one might use the
discipline of microstructure trading models to identify the countervailing forces
at work.
Another issue that is relevant the world over is stability of exchange-rate
pegs. Microstructure-style trading models help us to understand how and why
particular types of orders have price impact when exchange rates are pegged
(see, e.g., Calvo 1999, and Corsetti, Morris, and Shin 1999). As an empirical
matter, we have a lot to learn about which types of order flow cause pegs to
collapse (see Carrera 1999). A better understanding of these issues will aid in the
design of more resilient pegged regimes.

Policy Area 2: Central Bank Intervention

Central bank intervention is a classic topic within exchange-rate econom-

ics. It is also a natural topic for order-flow analysis. A key open issue in the
intervention literature is whether different-currency assets are imperfect
substitutes. Imperfect substitutability is important because it governs whether
portfolio-balance effects are being driven by order flow (irrespective of whether
those orders are from the central bank or the private sector). Portfolio-balance
effects on price arise under imperfect substitutabilityeven if order flow is
known to be liquidity motivated (i.e., known to be uncorrelated with traditional
macro fundamentals)because risk-averse participants need to be compensated

10Indeed, data availability is absolutely essential for keeping this empirically driven area of research
moving forward. Part of the challenge for official institutions will be to convince private-sector firms to
part with the employee time and money involved in providing such data. I hope that the progress I have
surveyed above will help make the case that the effort is a worthy one.

8
with additional expected return to take the other side of order flow. In contrast,
if at a given expected return investors were indifferent to holding dollar- or yen-
denominated assets (perfect substitutability), then exogenous innovations in
order flow would have no price effect.11
As an empirical matter, we have not yet reached consensus on the rele-
vance of FX portfolio-balance effects. Views about their presence have shifted
over the last twenty years from negative to moderately positive. The earlier
negative view was based on early empirical work that finds no evidence of
portfolio-balance effects (for an overview see Dominguez and Frankel 1993b, page
105). These early studies examine FX markets at a broad level, and address
whether different-currency returns are driven by changing asset supplies. In
general, these studies suffer from lack of statistical power, however, because
changing asset supplies are notoriously difficult to measure. Studies that focus
narrowly on the effects of central-bank interventiona kind of event study on
changing asset suppliesare more successful in finding effects from portfolio
balance (e.g., Loopesko 1984, Dominguez 1990, Dominguez and Frankel 1993a).
But even with this narrow focus on intervention events, results are not exclu-
sively positive (e.g., Rogoff 1984).
The order-flow analysis of Evans and Lyons (2000) is a new approach to
measuring portfolio-balance effects. They measure portfolio-balance effects
directly from the order flow of non-central-bank participants. Though this new
approach is not without drawbacks, it does avoid several of the drawbacks of the
earlier literature. For example, the Evans-Lyons approach is arguably more
powerful (statistically) than the approach of the early broad-level studies because
it does not rely on measuring changing assets supplies, making it less vulnerable
to measurement error. It may also be more powerful than the event-study
intervention approach because the number of intervention events one can
examine is not large, and the average size of interventions is small.
Let me try to make this point about statistical power more vivid. Envision
the FX market as a choke point. The choke point is where portfolios are actually
being balanced (which includes the portfolios of central banks). Measures of order
flow provide precise measures of this rebalancing, and the price effects that arise
as a result. As a choke point, the FX market is the venue where market partici-
pants (effectively) say to one another, Here, hold this, where this might be 10
billion euros. If we are to detect portfolio effects anywhere, this may be the right
place to look.

Policy Area 3: Emerging market design

A bread-and-butter policy issue in microstructure finance is how best to

design markets. This issue is relevant to FX as well. Its relevance in FX is
limited, however, by the fact that the major currencies are traded in a truly
worldwide market, making it difficult for any single regulatory authority to

11For a survey of the macro literature on portfolio-balance models, see Branson and Henderson (1985).
The macro literature does not address order flow per se. Nevertheless, the exogenous shifts in asset
demands and supplies the literature does address have a natural counterpart in order flow (as long as
that order flow is not motivated by traditional macro fundamentals).

9
impose structural changes. Any country attempting to alter or constrain the
structure of trading within its own borders would find order flow migrating
rapidly to other trading venues. As a practical matter, worldwide harmonization
of this type of policy change is infeasible at present.
The area where FX market design remains a hot topic is in emerging mar-
kets. Most of these currencies are not traded on a worldwide basis, due to lack of
convertibility of one form or another. Because trading in these currencies is
largely within-country, it is feasible to legislate market design in a way that is not
possible in major markets. Microstructure analysis is well-suited to address
whether fledgling FX markets should be organized as auction markets, or as
dealer markets, or both (for analysis along these lines see Kirilenko 1997), as well
as the level of transparency that should be required. Institutions like the Interna-
tional Monetary Fund confront this type of policy question regularly. The micro-
structure approach provides valuable guidance.

Policy Area 4: International currencies

What role should specific currencies play in the international monetary

and financial system? Recent introduction of the euro has brought this question to
a level of policy relevance not seen since the early 1970s (when the Bretton
Woods fixed-rate system collapsed). The issue of international currenciesthose
that act as a universal means of exchangecenters on three key aspects: (1) use
as a reserve currency by central banks, (2) use as an invoicing currency for
international transactions, and (3) use as a vehicle currency for currency transac-
tions. (Vehicle currencies are used when the transaction cost of trading two
currencies directly is higher than the cost of trading them indirectly, via two
transactions through the vehicle currency.) For aspects (2) and (3), a single
currencys success as an international currency is heavily dependent on the level
of transaction costs. So, to predict whether the euro will be successful as an
international currency, one needs to model the transaction costs that will arise
once it is fully adopted. Though not an easy task, it is one that microstructure
analysis is well suited to address (see, e.g., Hau, Killeen, and Moore 2000). This
type of analysis can also identify which institutional features of the new euro will
help to reduce those transaction costs. In this analysis, a central issue is the
degree to which a reduction in transaction costs stimulates trading, both directly,
and indirectly through adoption as a vehicle currency. These volume and liquidity
responses to various policy alternatives are the focus of recent work by Hartmann
(1998a, 1998b, 1999), and Portes and Rey (1998).

Policy Area 5: Transaction taxes

The issue of transaction taxes has attracted much attention among ex-
change-rate economists. Proponents of levying transaction taxes tend to associate
high volume with excessive speculation. As the literature has shown, however,
much FX volume reflects dealer risk management (hot-potato trading), rather
than speculation. Imposing a transaction tax would therefore impede risk
management. Though unintentional, this misunderstanding of the causes of high

10
volume could lead to bad policy. I emphasize the word could here because order-
flow analysis only adds a new dimension to this important policy question, it does
not invalidate the arguments of transaction-tax supporters. Looking forward, I
expect that order-flow analysis has a good deal more to contribute to this policy
issue. (For recent treatments using order-flow analysis, see Hau and Chevallier
2000 and Habermeier and Kirilenko 2000.)

4 Where FX is GoingImplications for Order-Flow Analysis

The FX market has undergone important changes over the last ten years.
Perhaps the most important change is the shift from voice-based interdealer
brokers to electronic interdealer brokers. This trend away from human-
intermediated transactions is evident in many securities markets throughout the
world.12 It shows no sign of abating.
The shift from voice-based interdealer brokers to electronic brokers is im-
portant in itself because electronic brokers provide a different (mostly higher)
level of order-flow transparency than was provided by the voice-based brokers.
This alters dealers information sets, which affects their trading strategies.
A larger implication of the shift to electronic brokers, however, is that it
suggests a future for spot FX trading that is more centralized, electronic, and
this is the crucial partopen to customers. Major markets will shift toward a
structure with open, electronic limit-order books that are accessible to a large
number of market participants.13 Under this scenario, institutions I have been
calling customers would be able to provide liquidity to one another, rather than
having to depend wholly on dealers. At that point, they would cease being
customersin the sense of always demanding liquiditytransformed instead into
both liquidity demander and supplier.
How can one be confident that the market is going in this direction? Three
pieces of evidence support this view. First, in June of 2000, three investment
banks (Goldman Sachs, Merrill Lynch, and Morgan Stanley Dean Witter) an-
nounced that they will be launching an electronic system of this kind for the US
bond markets (government and corporate bonds). The past structure of the US
bond market shares many characteristics with FX markets. Though several equity
and derivatives markets have already shifted to a centralized-electronic struc-
ture, those markets do not share the same FX-market characteristics that the
bond markets share, and are therefore not as appropriate as models. Second, in
the last year, new companies have introduced forms of centralized FX trading for
12 For an interesting article on the advent of electronic trading in FX, see Euromoney (2000). For equity
markets, Institutional Investor (2000) is a nice treatment of the electronic-trading threat to more
traditional trading methods.
13 Frankel (1996) was perhaps the first to write about such a scenario in FX, though he was considering

a Tobin tax as the possible trigger. He wrote: It is possible that the imposition of a Tobin tax would
alter the structure of the market in a fundamental way. It might become more like other major financial
markets, in which a sale or purchase by a customer generates only one or two transactions, rather than
five or eight. This would be the case particularly if such a tax triggered a transition to a new trading
structure equilibrium, with the decentralized dealer network replaced by a system in which foreign
currency was traded on a centralized exchange in the manner of the NYSE.

11
customers (e.g., [Link], FXchange, FXconnect, Atriax, [Link],
[Link], among others). These new companies typically promote
themselves as operating at the fringes of the dealer-market structure, coexisting
with it. But there is nothing obvious that prevents one of them from growing to a
scale that captures the network externalities inherent in concentrating liquidity
into a single pool.14
The third piece of evidence that, in my judgment, points to more central-
ized FX trading in the future is that existing interdealer systems like EBS can be
opened to customer-companies quite easily. (EBS is an electronic trading sys-
tema limit-order bookthat is available only to dealers.) The customer rela-
tionships are there: banks that own EBS are the same banks that have customer
relationships via their dealing services. The technology is not a major hurdle; the
switch could be flipped in much the same way as it promises to be flipped in the
US bond markets. What might the catalyst be? A natural catalyst would be
significant growth in market share by one of the new electronic entrants. If EBS
decided to open its system to customers, it would be difficult for any competitor
to beat it. From the EBS perspective, it is essential to maintain the threshold
effects of network externalities in its favorif the market is going in the direc-
tion of centralized customer trading, EBS cannot afford to wait.
Does immanent market-structure change threaten the relevance of the or-
der-flow analysis? My answer should not surprise anyone who has read this far.
Order-flow analysis is not concerned only with whether the market is organized
with a single dealer, multiple dealers, or a limit-order book. The role of order
flow in conveying information transcends market structure. And the types of
information that order flow conveysparticularly the types with persistent price
effectsare not likely to change radically when (if) the FX market structure
changes in the future. Put another way, the underlying information structure of
this market has more to do with the properties of the asset being tradedforeign
exchangethan it does with the market structure per se. Order flow will con-
tinue to tell us something about peoples view on how public information should
be mapped into price. It will continue to tell us something about current risk
preferences and endowments. In short, it will continue to convey dispersed
information that needs to be aggregated. And that is what the FX market is all
about.

14 For a theoretical treatment of whether centralized limit-order structures are likely to capture
liquidity and thereby dominate trading, see Glosten (1994). An issue not addressed in that paper that
is important for FX is credit risk. Bank dealers may have a comparative advantage in managing the
credit risk arising from large transactions with customers. New entrants who want to centralize this
market around an electronic trading platform need to solve this problem because non-financial
corporations do not want to take the counterparty credit risk that banks are comfortable taking. The
standard approach is to establish a clearing-house system with margin accounts (akin to those used in
futures markets).

12
 References

Anderson, T., and T. Bollerslev, 1998, Deutsche mark-dollar volatility: Intraday

activity patterns, macroeconomic announcements, and longer-run dependen-
cies, Journal of Finance, 53: 219-266.
Branson, W., and D. Henderson, 1985, The specification and influence of asset
markets, in R. Jones and P. Kenen (eds.), Handbook of International Econom-
ics, Volume 2, North-Holland: Amsterdam.
Calvo, G., 1999, Contagion in Emerging Markets: When Wall Street Is a Carrier,
University of Maryland working paper.
Carrera, J., 1999, Speculative attacks to currency target zones: A market micro-
structure approach, Journal of Empirical Finance, 6, 555-582.
Chang, K., and C. Osler, 1998, Methodical madness: Technical analysis and the
irrationality of exchange-rate forecasts, Economic Journal, 109: 636-661.
Corsetti, G., S. Morris, and H. Shin, 1999, Does one Soros make a difference? The
role of a large trader in currency crises, typescript, Yale University,
March.
Dominguez, K., 1990, Market responses to coordinated central bank intervention,
Carnegie-Rochester Series on Public Policy, volume 32.
Dominguez, K., 1999, The market microstructure of central bank intervention,
NBER Working Paper 7337, September.
Dominguez, K., and J. Frankel, 1993, Does foreign exchange intervention matter?
The portfolio effect, American Economic Review, 83: 1356-1369.
Dominguez, K., and J. Frankel, 1993b, Does foreign-exchange intervention work?
Institute for International Economics, Washington, D.C.
Dooley, M., and P. Isard, 1982, A portfolio-balance rational-expectations model of
the dollar-mark rate, Journal of International Economics, 12: 257-276.
Duarte, M., and A. Stockman, 2001, Rational speculation and exchange rates,
typescript, University of Rochester, March.
Euromoney, 2000, A true exchange for forex, July.
Evans, Martin, 2001, FX Trading and Exchange Rate Dynamics, NBER Working
Paper 8116, February.
Evans, Martin, and Richard Lyons, 1999, Order flow and exchange rate dynamics,
NBER Working Paper 7317, August, forthcoming in the Journal of Political
Economy.
Evans, Martin, and Richard Lyons, 2000, The price impact of currency trades:
Implications for Secret Intervention, U.C. Berkeley typescript, June, pre-
sented at the NBER Summer Institute, July 2000.
Flood, R., and A. Rose, 1995, Fixing exchange rates: A virtual quest for fundamen-
tals, Journal of Monetary Economics, 36: 3-37.
Frankel, J., 1996, How well do markets work: Might a Tobin tax help?, in M. ul
Haq et al. (eds.), The Tobin Tax: Coping with Financial Volatility (Oxford
University Press: New York), 41-82.
Frankel, Jeffrey, and Andrew Rose, 1995, Empirical research on nominal ex-
change rates, in G. Grossman and K. Rogoff (eds.), Handbook of Interna-
tional Economics, Elsevier Science: Amsterdam, 1689-1729.

13
Gehrig, T., and L. Menkhoff, 2000, The use of flow analysis in foreign exchange:
Exploratory evidence, typescript, University of Freiburg (Germany) De-
partment of Economics, August.
Glosten, L., 1994, Is the electronic open limit order book inevitable? Journal of
Finance, 49: 1127-1162.
Habermeier, K., and A. Kirilenko, 2000, Securities transaction taxes and financial
markets, typescript, International Monetary Fund, June.
Hartmann, P., 1998a, Currency competition and foreign exchange markets: The
dollar, the yen, and the euro, Cambridge University Press: Cambridge.
Hartmann, P., 1998b, Do Reuters spreads reflect currencies' differences in global
trading activity? Journal of International Money and Finance, 17: 757-784.
Hartmann, P., 1999, Trading volumes and transaction costs in the foreign ex-
change market: Evidence from daily dollar-yen spot data, Journal of Bank-
ing and Finance, 23: 801-824.
Hau, H., and A. Chevallier, 2000, Evidence on the volatility effect of a Tobin tax,
typescript, Insead, April.
Hau, H., W. Killeen, and M. Moore, 2000, The euro as an international currency:
Puzzling first evidence, CEPR working paper no. 2510.
Institutional Investor, 2000, Trading meets the millennium, January.
Isard, P., 1995, Exchange Rate Economics, Cambridge University Press: Cam-
bridge, England.
Killeen, W., R. Lyons, and M. Moore, 2000b, The puzzle of the euro, typescript,
U.C. Berkeley, July.
Kirilenko, A., 1997, Endogenous trading arrangements in emerging foreign
exchange markets, working paper, University of Pennsylvania (Wharton).
Levich, R., and L. Thomas, 1993, The significance of technical trading-rule profits
in the foreign exchange market: A bootstrap approach, Journal of Interna-
tional Money and Finance, 12: 451-474.
Lewis, K., 1988, Testing the portfolio balance model: A multilateral approach,
Journal of International Economics, 24: 109-127.
Loopesko, B., 1984, Relationships among exchange rates, intervention, and
interest rates: An empirical investigation, Journal of International Money
and Finance, 3: 257-277.
Lyons, R. 1995. Tests of Microstructural Hypotheses in the Foreign Exchange
Market, Journal of Financial Economics, 39: 321-351.
Lyons, R. 1996. Foreign Exchange Volume: Sound and Fury Signifying Nothing?
in J. Frankel et al. (eds.), The Microstructure of Foreign Exchange Markets,
University of Chicago Press: Chicago, 183-201.
Lyons, R. 1997. A simultaneous trade model of the foreign exchange hot potato,
Journal of International Economics 42, 275-298.
Lyons, R. 1998. Profits and position control: A week of FX dealing, Journal of
International Money and Finance, 17: 97-115.
Lyons, R., 2001, The Microstructure Approach to Exchange Rates, MIT Press:
Cambridge, MA. Fall (chapters available at [Link]/~lyons).
Madhavan, A., and G. Sofianos, 1997, An empirical analysis of NYSE specialist
trading, Journal of Financial Economics, 48: 189-210.
Mark, N., 1995, Exchange rates and fundamentals: Evidence on long-horizon
predictability, American Economic Review, 85: 201-218.

14
Meese, R., 1990, Currency fluctuations in the post-Bretton Woods era, Journal of
Economic Perspectives, 4: 117-134.
Meese, R., and K. Rogoff, 1983, Empirical exchange rate models of the seventies,
Journal of International Economics, 14: 3-24.
Naik, N., and P. Yadav, 2000, Do market intermediaries hedge their risk exposure
with derivatives? Typescript, London Business School, February.
Osler, C., 2000, Support for Resistance: Technical Analysis and Intraday Ex-
change Rates, Economic Policy Review, 6: 53-68.
Osler, C., 2001, Currency orders and exchange-rate dynamics: Microfoundations
of technical analysis, typescript, Federal Reserve Bank of New York, March.
Payne, R. 1999, Informed trade in spot foreign exchange markets: An empirical
investigation, typescript, London School of Economics, January.
Portes, R., and H. Rey, 1998, The emergence of the euro as an international
currency, Economic Policy, 26, 307-343.
Rey, H., 2001, International trade and currency exchange, Review of Economic
Studies, 68: 443-464.
Rime, D., 2000, Private or public information in foreign exchange markets? An
empirical analysis, typescript, University of Oslo, March
([Link]/~dagfinri).
Rogoff, K., 1984, On the effects of sterilized intervention: An analysis of weekly
data, Journal of Monetary Economics, 14: 133-150.
Taylor, M., 1995, The economics of exchange rates, Journal of Economic Litera-
ture, 83, 13-47.
Taylor, M., and H. Allen, 1992, The use of technical analysis in the foreign
exchange market, Journal of International Money and Finance, 11: 304-314.
Yao, J. 1998. Market making in the interbank foreign exchange market, New York
University Salomon Center Working Paper #S-98-3.

15
 Figure 1

Macro-Fundamental Analysis versus Order-Flow Analysis

Macro-Fundamental Analysis

Public information
about fundamentals Price

Order-Flow Analysis

Non-public information Order

about fundamentals Price
Flow

Hybrid Analysis

Information about
Order
fundamentals Price
Flow

* The top panel illustrates the connection between fundamentals and price in macro-fundamental
analysis: information about fundamentals is public, and so is the mapping to price, so price adjustment
is direct and immediate. The middle panel illustrates order-flow analysis. The focus in that case is
information that is not publicly known. This type of information is first transformed into order flow,

16
which becomes a signal to the price setter (e.g., dealer) that price needs to be adjusted. Actual markets
include both, which is illustrated in the bottom panelhybrid analysis.

17