Int. Fin. Markets, Inst.

and Money 19 (2009) 387–401

Contents lists available at ScienceDirect

Journal of International Financial

Markets, Institutions & Money
j o ur na l ho me pa ge : w w w . e l s e v i e r . c o m / l o c a t e / i n t f i n

The role of private information in return volatility, bid–ask

spreads and price levels in the foreign exchange market
Frank McGroarty a,∗, Owain ap Gwilym b, Stephen Thomas c
a
School of Management, University of Southampton, Highfield, Southampton, Hampshire SO17 1BJ, UK
b
School of Management & Business, University of Wales, Aberystwyth, Ceredigion SY23 3DD, UK
c
Faculty of Finance, Cass Business School, 106 Bunhill Row, London EC1Y 8TZ, UK

a r t i c l e i n f o a b s t r a c t

Article history: Trading volume and order flow have both been closely associ-
Received 17 January 2007 ated with informed trader activity in the market microstructure
Accepted 5 April 2008
literature. Using theory that explains regular intraday patterns in
Available online 14 April 2008
trading data, we transform these two variables into proxies for
private information and examine their relationships with bid–ask
JEL classification:
spreads and return volatility. We use a unique and unusually rich
F31
high-frequency intraday dataset from the world’s largest financial
G12
G15 market, namely, the electronic inter-dealer spot foreign exchange
D4 market. Our analysis takes account of institutional features peculiar
to this order-driven market. Our empirical results strongly affirm
Keywords: our theoretical understanding of how these markets work. They
High-frequency data
also reveal how the structure of the inter-dealer spot FX market
Foreign exchange
affects exchange rate volatility. Finally, we also explore how private
Market microstructure
Asymmetric information information contributes to the evolution of prices.
Order-driven © 2008 Elsevier B.V. All rights reserved.

1. Introduction

Private information is a common theme across various theories in the context of the widely docu-
mented patterns that exist in intraday trading data. These patterns emerge in volume, volatility and
bid–ask spread data from a wide variety of financial markets. There is a distinct literature within market
microstructure that aims to explain why these intraday patterns occur (see ap Gwilym and Sutcliffe,
1999). Our research uses that literature to explore how private information influences the formation of
price changes and bid–ask spreads. We also introduce order flow as an additional variable for intraday
pattern analysis.

∗ Corresponding author. Tel.: +44 23 8059 2540; fax: +44 23 8059 3844.
E-mail address: F.J.McGroarty@soton.ac.uk (F. McGroarty).

1042-4431/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.intfin.2008.04.001
388 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401

Our data are sourced from the world’s largest financial market, the inter-dealer spot foreign
exchange market. This market is particularly important because, as McGroarty et al. (2006) explain, the
inter-dealer market effectively sets the spot exchange rates for the entire FX market.1 In spite of its size
and importance, the spot FX market has been underrepresented in the literature on intraday empir-
ical regularities, due to difficulties in obtaining data. Our analysis addresses this imbalance. Intraday
trading data exhibit M- and U-shaped patterns, as markets go through the daily ritual of opening,
trading normally and closing. Patterns in bid–ask spreads, trading volume and return volatility have
been observed in all major financial markets. We believe that we are the first to reveal the intraday
pattern for order flow, for any market.2
Evans and Lyons (2002) established a strong relationship between cumulative order flow and
exchange rates. They define order flow as “the net of buyer-initiated and seller-initiated orders” and
interpret it as “a measure of net buying pressure”. They argue that order flow is driven by (private)
information and they provide strong evidence that order flow is the proximate driver of price in the
spot FX market. Specifically, they show that cumulative order flow is highly correlated with cumulative
price change. From market microstructure theory, Easley and O’Hara (1992) suggest that another vari-
able is closely linked to private information, namely, unanticipated deviations from normal intraday
trading volume levels. We explore both ideas.
There is an abundance of theory which models bilateral relationships between the variables that
we observe in intraday data, e.g. linking volatility and volume, or volume and bid–ask spreads. As
ap Gwilym and Sutcliffe (1999) observe, the most common theme across these bilateral relationship
models is private information. We use a correlation matrix to test multiple contemporaneous hypothe-
ses, and to examine how the relationships under investigation changed following the introduction of
the euro. We reveal the role of private information in explaining bid–ask spreads and in determining
exchange rate volatility, whose determinants Flood and Rose (1995) tell us “are not macroeconomic”.
Finally, we explore how private information contributes to incremental returns which compound to
determine the overall price level.
We take account of structures and practices in the inter-dealer spot FX market that are different
from the assumed market structure underlying most of the existing market microstructure theory.
The market we study is electronic, order-driven and conspicuously lacks market makers at its core.
Rather, any eligible agent who wishes to trade in this market has two choices. He can submit a market
order or a limit order. A market order executes immediately by selecting a trader on the other side
of the market who has previously advertised on the system that he is willing to trade. A limit order
is where a potential trader submits an advertisement that he is willing to trade. This limit order will
sit alongside orders already in the system, awaiting execution. Limit orders can be either ask-side or
bid-side. The same is true for market orders. There is no pressure on any market participant to submit
two-way limit orders and no evidence that anyone routinely submits such orders.
The remainder of the paper is organized as follows. Section 2 reviews the empirical literature doc-
umenting patterns in intraday trading variables and explores the theory relating these data variables
to each other in order to deduce a comprehensive set of mutually consistent bilateral hypothesized
relationships. Section 3 describes the data and methodologies we use. Section 4 presents the empirical
results. Section 5 concludes.

2. Literature review and hypotheses

The vast majority of studies which look at the intraday patterns of bid–ask spreads, trading volume
and return volatility in financial markets find that they are U-shaped (see ap Gwilym and Sutcliffe,
1999 for a detailed survey). In some cases, the value at the open is larger than at the close, giving an L
shape. Sometimes, the close is larger than the open, giving a J shape. In some instances, there is a spike

1
All dealers in FX banks have EBS and Reuters 2000-2 terminals on their desks. These electronic inter-dealer markets are the
most liquid and their prices are the most recent. Dealers reference these prices when giving quotes to customers.
2
A working paper by Gencay et al. (2007) confirms our intraday order flow pattern for the EUR/USD exchange rate. Their
2003 and 2004 pattern closely resembles the pattern that we find for EUR/USD in 1998 and 1999 data.
 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401 389

of activity at lunchtime, giving a W shape. However, these are merely variations on the predominant
U shape which shows large volume, high volatility and wide bid–ask spreads at the open and close,
and more moderate levels of all three in between. The only persistent exception is the spot FX market,
which, as the only truly global market, does not experience daily open and close events. This market
produces intraday patterns for volume and volatility in the shape of the letter M and intraday bid–ask
spreads in a U-shaped pattern.
Danielsson and Payne (2001) find an M-shaped volume pattern and a W-shaped bid–ask spread
pattern for the USD/DEM spot exchange rate on the Reuters global inter-dealer FX trading platform.
Baillie and Bollerslev (1990) find that volatility for the major currency pairs peaks twice during the
day, when London and New York open, yielding an M-shaped plot. Low and Muthuswamy (1996) find
similar peaks in price change volatility for three major currency pairs when London and New York
open and close. Docking et al. (1999) find the same pattern for the USD/DEM. To date, the literature on
intraday empirical patterns neglects intraday order flow.
ap Gwilym and Sutcliffe (1999) divide the explanations for these observed patterns in intraday
trading data into two categories. The first, typified by Brock and Kleidon (1992), attributes these pat-
terns to differing trader behavior at the open and close. The other approach, typified by Admati and
Pfleiderer (1988), attributes the patterns to strategic behavior of informed traders. Since there is no
daily open and close in the global foreign exchange market, it is the latter approach that is the focus
of our attention.
Admati and Pfleiderer (1988) argue that informed traders trade when many uninformed traders
are present in the market, in order to minimize their transactions costs, i.e. bid–ask spreads. The
bid–ask spread will be narrower when there are many uninformed traders because market makers’
inventory management costs are lower in such circumstances. If informed traders can hide among
the uninformed traders, they can avail of narrower bid–ask spreads. This perspective fits well with
an analysis of the global foreign exchange market which has trading volume naturally concentrated
in London and New York business hours. However, it is not clear that we can use this argument to
treat volume as a direct proxy for private information because we also have more uninformed trading
activity at high-volume times. Any relationship that we might discover between high volume and
bid–ask spreads or return volatility may just as easily be caused by the high-uninformed trading
volume as by trading volume generated by informed traders. Easley and O’Hara (1992) provided an
important insight into the link between trading volume and private information when they suggested
that private information signals cause trading volume to deviate from its normal level. Empiricists,
including Bessembinder (1994), Jorion (1996), Hartmann (1998) and Danielsson and Payne (2001)
have used the Easley and O’Hara (1992) idea to split trading volume into expected and unexpected
components. Unexpected trading volume is linked to private information. Expected trading volume
is not. This leads us to conclude that expected and unexpected volume should be orthogonal to each
other.
H1. Expected correlation (expected volume, unexpected volume) = 0.
Our other proxy for private information is order flow. Evans and Lyons (2002) provide compelling
empirical evidence of a strong relationship between cumulative order flow and exchange rates. They
attribute this to unobservable shifts in expectations which affect not only future returns but also
risk appetites and discount rates. From microstructure theory, Kyle (1985), upon which Admati and
Pfleiderer (1988) base their model, uses order flow as the conduit by which private information affects
the bid–ask spread.
This contrasts starkly with Garman’s (1976) attribution of order flow to random fluctuations in
inventory levels. Garman’s (1976) model considers supply and demand volumes as being determined
separately from independent distributions, giving rise to temporal imbalances between supply and
demand that are not motivated by news, i.e. (uninformed) order flow. The daily ebb and flow of trading
volume is reflected on both the bid and ask sides of the market, suggesting that both supply and demand
peak at London and New York open times. Consequently, the variance of potential imbalances between
supply and demand volumes varies systematically over the trading day, as a function of trading volume.
It seems reasonable to separate information-driven order flow from non-information-driven
order flow by dividing order flow into expected and unexpected components, as we did with trading
390 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401

volume. To achieve this separation, we must use absolute order flow. As with volume, the unexpected
component would be associated with private information but the expected component would not.
Expected and unexpected order flow should be orthogonal to each other. The two proxies for private
information should be positively correlated. Similarly, the expected components linked to uninformed
trading should also be positively correlated. In addition, each proxy of private information should
also be orthogonal to the alterative proxy for uninformed trading.

H2. Expected correlation (expected order flow, unexpected order flow) = 0.

H3. Expected correlation (expected volume, expected order flow) = +.

H4. Expected correlation (unexpected volume, unexpected order flow) = +.

H5. Expected correlation (expected volume, unexpected order flow) = 0.

H6. Expected correlation (expected order flow, unexpected volume) = 0.

Much market microstructure research is devoted to exploring and explaining two price related
variables: the bid–ask spread and return volatility. We contribute some insight into the determinants
of these two variables, using our proxies for private information, as well as their expected counter-
parts which proxy for uninformed trading activity. Market microstructure theory tells us that bid–ask
spreads and return volatility are related to each other. Roll (1984) used the empirical phenomenon of
bid–ask bounce3 to derive the effective bid–ask spread directly from return volatility. The relationship
between bid–ask spread and return volatility is positive in Roll’s (1984) model, because wider bid–ask
spreads give rise to larger bounces between successive bid-side and ask-side trades, and so to higher
return volatility. Hartmann (1998) found strong empirical support for a more elaborate bid–ask spread
model which is positively dependent on return volatility, but negatively dependent on volume.
The above discussion may lead us to expect that bid–ask spreads and return volatility are positively
correlated. However, as McGroarty et al. (2006) point out, bid–ask spreads in the electronic inter-dealer
spot FX market are intrinsically different from those in established market microstructure theory4
which underlie the above models. Mainstream market microstructure bid–ask models incorporate
Amihud and Mendelson’s (1980) notion that market makers shade their prices down (up) to sell (buy)
excess inventory. This process dampens price volatility because it deliberately targets and entices
sequentially opposite sides of the market to trade. A market maker-less electronic order-driven market
lacks this price dampening mechanism. If a market order absorbs all the limit order volume at a specific
price level, the price of the next best limit order will be revealed. Subsequent traders are faced with a
higher prevailing mid-price on which to base their trading decisions. In a market maker-less market,
an imbalance between buyers and sellers will result in a price disturbance, even if that imbalance does
not reflect informed trading. Such price disturbances may be short-lived, but they will contribute to
higher volatility at the high-frequency level than would be present if market makers participated.
Bid-ask spreads are a feature of the electronic inter-dealer spot FX market. In this market, traders
have to weigh the benefit of immediate execution via market order, against the potential benefit of
waiting for a price improvement via limit order. When aggressive limit orders are submitted, the
bid–ask spread narrows. In this market, transaction prices should still bounce between bid and ask
quote prices. While this contributes to the overall observed return volatility, the effect should be much
weaker than that observed in settings with market makers at their core. For this reason, we expect to
find a positive but weak correlation between bid–ask spreads and return volatility.

3
In financial markets, some prices execute at the ask side of the market and some prices execute at the bid side of the market.
The oscillation of transaction prices between bid and ask quotes is known as bid-ask bounce.
4
McGroarty et al. (2006) argue that the existing theory assumes that the bid-ask spread is compensation to a market maker
for risking his own capital in order to facilitate trading by other market participants. They find this concept of a market at
odds with the electronic inter-dealer spot foreign exchange market, where they say there is no evidence of anybody is acting
like a market maker, i.e. providing simultaneous bid and ask limit prices (quotes). Instead, this market consists of FX dealers,
who may act as market makers when dealing with their customers, but who submit either one-sided limit or market orders
in the electronic inter-dealer market. In the electronic inter-dealer market, the bid-ask spread is the difference between the
best available bid limit order price and the best available ask limit order price, where each best limit order comes from an
independent party.
 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401 391

H7. Expected correlation (bid–ask spread, return volatility) = ∼+.

McGroarty et al. (2006) show that electronic inter-dealer spot FX bid–ask spreads fall as trading
volumes rise for a technical reason. They argue that if prices follow a random walk with drift, then bid
limit orders and ask limit orders, which arrive independently, will have a shorter time gap between
their arrival times when volume is high. The shorter time gap means that the drift component of mid-
price will be smaller, therefore the gap between the bid price and ask price will be lower. In the context
of our study, this means that bid–ask spreads and expected volume should be negatively correlated.
Similarly, bid–ask spreads should be negatively correlated with expected order flow, insofar as the
latter is also a proxy for uninformed trading activity.

H8. Expected correlation (bid–ask spread, expected volume) = −.

H9. Expected correlation (bid–ask spread, expected order flow) = −.

Clark’s (1973) mixture of distributions hypothesis (MDH) argues that volatility and volume move
together in response to common but not directly observable external stimuli, which is associated
with public information. Subsequently, Epps and Epps (1976) linked intraday volatility to disparate
opinions among traders following a public price signal. Tauchen and Pitts (1983) propose a bivari-
ate mixture model in which volume and price change are jointly distributed due to the presence of
a latent variable. This model shows the covariance between trading volume and (signed) return as
zero, whereas the covariance between volume and return volatility is distinctly positive, as has been
empirically observed. Order flow is not an explicit focus of these MDH models. However, it is hard to
conceive of how price could move more sharply up or down as volume rises, without a corresponding
imbalance between buy-side and sell-side trading volume. As such, order flow should prove to be the
conduit by which trading volume drives price (volatility). This idea is supported by the sequential
information arrival models of Copeland (1976) and Jennings et al. (1981). In these models, an individ-
ual trader receives a signal ahead of the market and trades on it, thereby creating volume and moving
price (volatility). These ideas lead us to expect that return volatility is positively correlated with both
expected volume and expected order flow.

H10. Expected correlation (return volatility, expected volume) = +.

H11. Expected correlation (return volatility, expected order flow) = +.

McGroarty et al.’s (2007) characterization of the bid–ask spread in the market maker-less electronic
inter-dealer spot FX market includes no adverse selection component of the bid–ask spread. They
argue that the absence of market makers means there is nobody to be compensated for accepting
inventory that they do not want, or cannot sell on before price changes adversely. Rather, both parties
to every executed trade want the outcome of that trade. This means that trades motivated by private
information should not affect the bid–ask spread. Instead, order flow (volume) which is prompted by
private information should impact directly on price, driving it up or down. Consequently, our private
information proxies should be unrelated to the bid–ask spread but positively correlated with return
volatility.

H12. Expected correlation (bid–ask spread, unexpected volume) = 0.

H13. Expected correlation (bid–ask spread, unexpected order flow) = 0.

H14. Expected correlation (return volatility, unexpected volume) = +.

H15. Expected correlation (return volatility, unexpected order flow) = +.

By laying out our 15 hypotheses in a matrix of predicted correlation signs, a succinct picture of our
predictions emerges:
392 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401

BA 1
RV ∼+ 1
EV – + 1
UV 0 + 0* 1
EO – + + 0 1
UO 0 + 0 + 0* 1
BA RV EV UV EO UO

BA, bid–ask spread; RV, return volatility; EV, expected volume; UV, unexpected volume; EO, expected order flow; UO, unexpected
order flow. * Should hold by construction.

3. Data and methodology

The data used in this study are 5-min observations sampled from a large spot FX database of tick
data provided by EBS. This data has not previously been made available to academic researchers. EBS
is the larger of the two electronic venues that make up the inter-dealer spot FX market. The other
is Reuters 2000-2. EBS is dominant in the large currencies involving the USD, EUR and JPY, whereas
Reuters 2000-2 dominates the Commonwealth and Scandinavian currencies.
The foreign exchange market generally, and the electronic inter-dealer spot FX market in particular,
are very important financial markets. In 1998, the Bank of International Settlements estimated that
the total volume of spot foreign exchange trading was worth almost US$1.5 trillion per day. After
EMU, the total value of FX transactions fell. However, the FX market remains the largest financial
market in the world, by many orders of magnitude. BIS (1998) and BIS (2001) both show that spot FX
transactions account for a little under half of all trading activity in the FX market. Over the past decade,
the importance of electronic trading venues in the inter-dealer market has risen sharply. BIS (2001)
estimates that between 85% and 95% of inter-bank trading occurred over electronic trading systems
in 2000. In the UK, which is the largest FX trading centre, Bank of England (2004) ascribes two-thirds
of all inter-dealer spot FX trading volume jointly to EBS and Reuters. Furthermore, McGroarty et al.
(2006) argue that the electronic inter-dealer market sets the exchange rates for the rest of the FX
market because it is by far the most transparent and up-to-the-second segment of the FX market.
The EBS data contains spot FX limit order price data and trade price data for two sample periods
with five important exchange rates5 in each. The limit order price data comprises the best bid and ask
limit order prices per second (Greenwich mean time, GMT). Trade data is also time-stamped to the
nearest second (GMT). No information about the size of each transaction is provided. Also, there are
no identifiers of the parties to each trade. The first sample period covers 1 August 1998 to 4 September
1998 and consists of USD/DEM, USD/JPY, USD/CHF, DEM/JPY and DEM/CHF. The second sample covers
1 August 1999 to 3 September 1999 and contains data on EUR/USD, USD/JPY, USD/CHF, EUR/JPY and
EUR/CHF. Each sample contains 20 days’ data. Here, the EUR is taken to be the linear successor to the
DEM on the grounds that, pre-EMU, the DEM was a pan-European vehicle currency (see Hartmann,
1998). This enables us to pair exchange rates involving the DEM with those involving the EUR for the
purpose of analysis, giving us in effect five exchange rates to evaluate and two samples (pre-EMU and
post-EMU) for each exchange rate.
Bid-ask spread and return volatility variables are both calculated using the difference in log prices.
Bid-ask spreads use the last bid and ask prices from series of best quote prices in each 5-min interval.
Absolute return, which is used as a proxy for return volatility, is computed from the mid-quote price.
The other four variables, expected and unexpected order flow, and expected and unexpected volume
are derived from our total order flow and total volume variables respectively. Our data does not provide
actual volume data, so the number of trades is used as a proxy. Fortunately, our data does provide the
side of each trade. Our measure of total volume comprises the total number of buyer-initiated trades
over the interval plus the total number of seller-initiated trades. Following Hartmann (1998), our total

5
Trading volume is highly concentrated in the spot foreign exchange market. In the period studied here, EUR/USD, USD/DEM
and USD/JPY all have trading volume on EBS which is more three times the next most heavily traded exchange rate. The next
most important exchange rates include EUR(DEM)/JPY and rates involving CHF and GBP. We focus on CHF rates rather than GBP
rates because EBS is stronger in those markets.
 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401 393

order flow measure is arrived at by taking the absolute difference between the number of buy (ask
side) trades and the number of sell (bid side) trades.
To calculate expected volume, we follow Danielsson and Payne (2001) in measuring the across-
day average volume for each time-of-day segment. Unexpected volume is calculated as the difference
between the actual volume in each observation, per day, and expected volume. Similarly, expected
order flow is computed as the across-day average of absolute order flow per time-of-day segment.
Unexpected order flow is the difference between the absolute value of actual order flow and the
expected measure. The variables are formally defined as follows:

BAd,t = log(ad,t ) − log(bd,t ) (1)

RVd,t = |log(pd,t ) − log(pd,t−1 )| (2)


D
Vd,t
EVt = (3)
D
d=1

UVd,t = Vd,t − EVt (4)


D
|Od,t |
EOt = (5)
D
d=1

UOd,t = |Od,t | − EOt (6)

In these equations, the measures are shown with two subscripts, d which represents the day and
t which represents the time of day (at 5-min intervals). D represents the total number of days. These
variables are computed separately for each of the five currency pairs within each pre-EMU and post-
EMU data sample.
Our empirical analysis consists of two steps. The first step is to compute correlation matrices of
these six variables for each currency pair and each sample period. Close matches to our predicted
correlation matrix would lend credibility to our characterization of the institutional features of the
market and to our hypothesized bilateral relationships between the variables. Any major differences
would need explanation and would require us to re-evaluate the theory underlying our predictions.
We also need to establish the significance of the correlation coefficients we compute. The significance
test for a correlation coefficient is
√
r N−2
t=  (7)
1 − r2
where r is the correlation coefficient, N the number of observations and t is a t-distributed test statistic.
Critical values for t are 1.96 (5%) and 2.58 (1%). We analyze our data using a full day sample, where N is
usually above 7000. For each N, there is a strict monotonic relationship between r and t. All correlation
coefficients larger than 3% in magnitude will be statistically significant at the 1% level.
The second strand of our empirical analysis focuses on the relationship between order flow and
price evolution. Evans and Lyons (2002) find that cumulative daily order flow is highly correlated
with movements in exchange rates and their analysis attributes order flow to informed trading. This
finding raises two questions of our data. First, to what extents do (uninformed) expected order flow
and (informed) unexpected order flow both contribute to price formation? Second, is there evidence
of any lead or lag relationship between our measures of private information, order flow and price
movement? A priori, one might expect a positive relationship between lagged (informed) unexpected
order flow and subsequent returns, based on two complementary ideas. First, other traders could
extract information from earlier informed order flow and follow suit. Second, private information may
eventually become public, triggering a price move in the same direction as the pre-emptive informed
trading. On the other hand, in markets where market-maker inventory is important (uninformed)
expected order flow should be followed by price shading in the opposite direction, showing a negative
lagged relationship. However, since electronic inter-dealer markets do not have market makers, we
expect to find that returns are independent of lagged expected order flow in our data. On the grounds
394 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401

that the lead–lag relationship might be other than what we expect, we test lead order flow as well as
lagged. Up to now, we had been using absolute returns. However, to explore price evolution, we need
signed returns:

Rd,t = log(pd,t ) − log(pd,t−1 ) (8)

Earlier, absolute order flow was appropriate for measuring relationships with volatility, volume and
bid–ask spreads which are non-negative variables. Now, negative return requires negative order flow
for any meaningful connection to be established between these two variables. We derive expected
and unexpected components from absolute order flow. Then we multiply expected order flow and
unexpected order flow by the sign of the raw order flow (S) for each time interval. Combining the
preceding elements gives us the following regression equation, which we estimate using OLS:


4
Rd,t = ˛0 + ˛1 (Sd,t−i EOt−i ) + ˛2 (Sd,t−i UOd,t−i ) (9)
i=−4

A cornerstone of our analysis is that we investigate two proxies for private information. The pre-
ceding regression will explore how our first proxy, based on order flow, affects price evolution. We also
wish to investigate whether our other private information proxy, unexpected volume, can be shown to
play any similar role in how prices are formed. In addition, we will examine our other volume variable,
expected volume, to see if this measure of uninformed trading has any effect on returns. As with order
flow, volume needs to be able to change sign in order explain return. However, volume is a strictly
positive variable. For this reason, we impose the sign from order flow on expected and unexpected
volume to specify the second OLS regression equation:


4
Rd,t = ˛0 + ˛1 (Sd,t−i EVt−i ) + ˛2 (Sd,t−i UVd,t−i ) (10)
i=−4

4. Empirical results

Graphs of intraday variables exhibit the familiar shapes of previous FX intraday pattern studies.
These are not shown here due to space constraints. Bid-ask spreads have a prolonged U shape and
are narrowest when London and New York are open. Volume and volatility are both M-shaped with
peaks at London open and the New York open. These descriptions apply to all 10 exchange rate samples
that we use, with the exception of exchange rates involving the JPY, where there is an additional but
smaller peak for volume and volatility, as well as an additional dip in the bid–ask spread, when the
Tokyo market is active. We also plotted the intraday absolute6 order flow, which is broadly M-shaped
and tracks intraday volume particularly closely. This closeness is confirmed by the very high-positive
correlation values between volume and order flow variables in the tables below.
Tables 1–5 present the correlation matrices for our five exchange rates. They are based on 5-min
frequency data sampled from the 24-h global FX market. Part (a) of each table is based on the pre-EMU
data from 1998, while part (b) represents post-EMU data from 1999. All correlation matrices match
our predicted matrix remarkably well, which provides clear support for our characterization of the
relationships between bid–ask spreads, return volatility, volume and order flow. We also produced cor-
relation matrices with data taken only from the busiest part of the trading day (0700GMT–1700GMT),
i.e. when London and New York are operating and trading volumes are at their highest. These exhibited
very similar correlation coefficients overall, but are not presented here due to space constraints.
In almost all cases in Tables 1–5, a positive but small correlation coefficient is evident between
bid–ask spreads and return volatility. Almost all are statistically significant at the 95% confidence
level. Using the full data sample, bid–ask spreads are 5% correlated with return volatility on average.

6
We used the absolute value of order flow because negative values cancel positive values to give overall intraday averages of
zero for signed order flow.
 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401 395

Table 1
USD/JPY

(Part a) 1998
BA 1
RV 0.05 1
EV −0.23 0.21 1
UV −0.01 0.53 0 1
EO −0.22 0.19 0.89 0 1
UO −0.01 0.35 0 0.38 0 1
BA RV EV UV EO UO

(Part b) 1999
BA 1
RV 0.04 1
EV −0.2 0.19 1
UV −0.03 0.52 0 1
EO −0.2 0.18 0.9 0 1
UO −0.03 0.39 0 0.41 0 1
BA RV EV UV EO UO

The table is based on the full 24-h global spot USD/JPY data from EBS. Part (a) uses pre-EMU data from 1998, whereas part (b)
uses post-EMU data from 1999. Each correlation coefficient is computed using approximately 7000 observations per variable,
sampled at the 5-min frequency. BA, bid–ask spread; RV, return volatility; EV, expected volume; UV, unexpected volume; EO,
expected order flow; UO, unexpected order flow.

This rises to an average of 7% when we focus on the high-volume data alone. This evidence upholds
the McGroarty et al. (2006) contention that electronic inter-dealer spot FX bid–ask spreads are very
different from the mainstream bid–ask spread model. McGroarty et al. (2007) argue that, in this market
structure, there is no inventory to be managed and no traders who can be adversely selected. Instead,
both informed trades and uninformed trades disturb the price.
As predicted, the bid–ask spread is consistently negatively correlated with our uninformed trading
activity proxies: expected volume and expected order flow. The correlation coefficients range between
−20% and −40%. These support McGroarty et al.’s (2006) assertion that electronic inter-dealer spot FX
market bid–ask spreads are affected by the independent arrival time of bid and ask limit orders.
The correlation between the bid–ask spreads and our private information proxies, unexpected
volume and unexpected order flow, is 0%, on average, across all 10 correlation matrices, and in every
instance, the individual value is insignificant at the 95% confidence level. This supports McGroarty

Table 2
USD/CHF

Part (a) 1998

BA 1
RV 0.08 1
EV −0.36 0.08 1
UV −0.02 0.42 0 1
EO −0.4 0.07 0.93 0 1
UO −0.02 0.33 0 0.54 0 1
BA RV EV UV EO UO

Part (b) 1999

BA 1
RV 0.01 1
EV −0.3 0.19 1
UV −0.02 0.43 0 1
EO −0.33 0.19 0.95 0 1
UO −0.03 0.37 0 0.45 0 1
BA RV EV UV EO UO

The table is based on the full 24-h global spot USD/JPY data from EBS. Part (a) uses pre-EMU data from 1998, whereas part (b)
uses post-EMU data from 1999. Each correlation coefficient is computed using approximately 7000 observations per variable,
sampled at the 5-min frequency. BA, bid–ask spread; RV, return volatility; EV, expected volume; UV, unexpected volume; EO,
expected order flow; UO, unexpected order flow.
396 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401

Table 3
EUR(DEM)/USD

Part (a) 1998

BA 1
RV 0.05 1
EV −0.21 0.2 1
UV −0.01 0.52 0 1
EO −0.2 0.19 0.91 0 1
UO −0.01 0.35 0 0.35 0 1
BA RV EV UV EO UO

Part (b) 1999

BA 1
RV −0.06 1
EV −0.24 0.29 1
UV −0.04 0.48 0 1
EO −0.25 0.29 0.93 0 1
UO −0.03 0.38 0 0.33 0 1
BA RV EV UV EO UO

The table is based on the full 24-h global spot EUR(DEM)/USD data from EBS. Part (a) uses pre-EMU data from 1998, whereas part
(b) uses post-EMU data from 1999. Each correlation coefficient is computed using approximately 7000 observations per variable,
sampled at the 5-min frequency. BA, bid–ask spread; RV, return volatility; EV, expected volume; UV, unexpected volume; EO,
expected order flow; UO, unexpected order flow.

et al.’s (2007) claim that there is no adverse selection risk component in electronic inter-dealer spot
FX market bid–ask spreads. They argue that informed trades should only affect price levels, driving
them up or down, while leaving bid–ask spreads unaffected. This assertion finds strong support in the
consistently high-positive correlations between return volatility and both of our private information
proxies, which range from 33% to 53%.
Clark’s (1973) mixture of distributions hypothesis also finds support. Correlations between return
volatility and our uninformed trading proxies, expected volume and expected order flow, are all con-
sistently positive. However, with a maximum correlation coefficient of 29%, these variables are less
strongly correlated with return volatility than our private information proxies. Nevertheless, the evi-
dence clearly shows that private information and uninformed temporary supply/demand imbalances
contribute independently to spot FX return volatility. We find that informed trading is a more impor-
tant driver of volatility than uninformed trading (a result consistent with Gencay et al., 2007). Jointly,

Table 4
EUR(DEM)/JPY

Part (a) 1998

BA 1
RV 0.08 1
EV −0.3 0.13 1
UV 0 0.38 0 1
EO −0.31 0.14 0.94 0 1
UO 0.02 0.36 0 0.46 0 1
BA RV EV UV EO UO

Part (b) 1999

BA 1
RV 0.05 1
EV −0.32 0.1 1
UV −0.02 0.39 0 1
EO −0.33 0.12 0.94 0 1
UO −0.02 0.41 0 0.62 0 1
BA RV EV UV EO UO

The table is based on the full 24-h global spot EUR(DEM)/JPY data from EBS. Part (a) uses pre-EMU data from 1998, whereas part
(b) uses post-EMU data from 1999. Each correlation coefficient is computed using approximately 7000 observations per variable,
sampled at the 5-min frequency. BA, bid–ask spread; RV, return volatility; EV, expected volume; UV, unexpected volume; EO,
expected order flow; UO, unexpected order flow.
 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401 397

Table 5
EUR(DEM)/CHF

Part (a) 1998

BA 1
RV 0.09 1
EV −0.27 0.12 1
UV 0.03 0.46 0 1
EO −0.28 0.12 0.94 0 1
UO 0.01 0.38 0 0.51 0 1
BA RV EV UV EO UO

Part (b) 1999

BA 1
RV 0.08 1
EV −0.35 0.04 1
UV −0.02 0.41 0 1
EO −0.36 0.04 0.94 0 1
UO −0.02 0.34 0 0.52 0 1
BA RV EV UV EO UO

The table is based on the full 24-h global spot EUR(DEM)/CHF data from EBS. Part (a) uses pre-EMU data from 1998, whereas part
(b) uses post-EMU data from 1999. Each correlation coefficient is computed using approximately 7000 observations per variable,
sampled at the 5-min frequency. BA, bid–ask spread; RV, return volatility; EV, expected volume; UV, unexpected volume; EO,
expected order flow; UO, unexpected order flow.

the two independent effects have correlations of over 50%, on average. Indeed, EUR/USD expected and
unexpected volume has a combined correlation of 77% with return volatility, in the 1999 sample.
Very large correlation coefficients are observed between expected volume and expected order
flow. Strong positive correlations are also evident between unexpected volume and unexpected order
flow. The close relationship between these two variables is consistent with the assertion that both
are proxies for private information. The independence of the expected and unexpected variables from
each other, illustrated by correlation coefficients of 0, is as predicted.
Results from the regression analysis are presented in Tables 6–10. All leading and lagged variables
proved to be uncorrelated with returns, with individual R-squared values of zero in every case. Conse-
quently, all lead and lagged variables are excluded from these tables, i.e. all displayed results relate to
contemporaneous variables only. While we had predicted that return would be independent of lagged
expected order flow, its apparent independence of lagged unexpected order flow is puzzling. It may
be that 5-min sampling frequency does not capture this relationship and that a different time interval
would fare better.

Table 6
USD/JPY

Part (a) 1998

S EO and S UO 0.36 0.21 S EV and S UV
O 0.35 0.2 S V
S EO 0.2 0.18 S EV
S UO 0.16 0.03 S UV

Part (b) 1999

S EO and S UO 0.38 0.31 S EV and S UV
O 0.38 0.3 S V
S EO 0.25 0.23 S EV
S UO 0.16 0.08 S UV

R-squareds: The table is based on the full 24-h global spot USD/JPY data from EBS. Part (a) uses pre-EMU data from 1998,
whereas part (b) uses post-EMU data from 1999. Each R-squared is computed using approximately 7000 observations per
variable, sampled at the 5-min frequency. In each panel, the first row shows the R-squared for regressions of return on expected
and unexpected components. The second row shows a regression of return on original order flow (O) and signed volume (S V).
The third and fourth rows show the R-squareds for individual regressions of return on expected and unexpected variables,
respectively. All variables are contemporaneous. Variables derived from order flow are displayed on the left of each panel, while
those on the right are derived from volume. In all cases, the correlation between S EO and S UO is zero. O, order flow; S EO,
signed expected order flow; S UO, signed unexpected order flow; S V, signed volume; S EV, signed expected volume; S UV,
signed unexpected volume.
398 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401

Table 7
USD/CHF

Part (a) 1998

S EO and S UO 0.29 0.23 S EV and S UV
O 0.28 0.22 S V
S EO 0.19 0.17 S EV
S UO 0.14 0.07 S UV

Part (b) 1999

S EO and S UO 0.35 0.23 S EV and S UV
O 0.33 0.21 S V
S EO 0.22 0.19 S EV
S UO 0.17 0.04 S UV

R-squareds: The table is based on the full 24-h global spot USD/CHF data from EBS. Part (a) uses pre-EMU data from 1998,
whereas part (b) uses post-EMU data from 1999. Each R-squared is computed using approximately 7000 observations per
variable, sampled at the 5-min frequency. In each panel, the first row shows the R-squared for regressions of return on expected
and unexpected components. The second row shows a regression of return on original order flow (O) and signed volume (S V).
The third and fourth rows show the R-squareds for individual regressions of return on expected and unexpected variables,
respectively. All variables are contemporaneous. Variables derived from order flow are displayed on the left of each panel, while
those on the right are derived from volume. In all cases, the correlation between S EO and S UO is zero. O, order flow; S EO,
signed expected order flow; S UO, signed unexpected order flow; S V, signed volume; S EV, signed expected volume; S UV,
signed unexpected volume.

All exchange rates reveal remarkably similar dynamics in both time periods. We carried out the
same analysis for the high-volume only trading period, for which the results were very similar and are
not presented due to space constraints. All independent variable coefficients were positive. Unexpected
order flow proves to be a significant explanatory factor for contemporaneous returns, confirming the
importance of private information in the price evolution process. However, in almost every case,
expected order flow appears even more influential than unexpected order flow. The signed vol-
ume regression results confirm this, with individual R-squared values for signed expected volume
regressions being very much larger than the corresponding values for the signed unexpected volume
regressions. Both order flow and volume measures tell the same story, i.e. that uninformed trading
moves price by more than trading arising from private information.
The greater importance of uninformed trading in the regression results contrasts with the correla-
tion results, where informed trading appeared to be the most important variable in explaining return
volatility. The introduction of the sign from order flow in the regression analysis seems to dissipate

Table 8
EUR(DEM)/USD

Part (a) 1998

S EO and S UO 0.33 0.2 S EV and S UV
O 0.33 0.2 S V
S EO 0.18 0.15 S EV
S UO 0.15 0.05 S UV

Part (b) 1999

S EO and S UO 0.39 0.26 S EV and S UV
O 0.38 0.26 S V
S EO 0.25 0.21 S EV
S UO 0.16 0.05 S UV

R-squareds: The table is based on the full 24-h global spot EUR(DEM)/USD data from EBS. Part (a) uses pre-EMU data from
1998, whereas part (b) uses post-EMU data from 1999. Each R-squared is computed using approximately 7000 observations
per variable, sampled at the 5-min frequency. In each panel, the first row shows the R-squared for regressions of return on
expected and unexpected components. The second row shows a regression of return on original order flow (O) and signed
volume (S V). The third and fourth rows show the R-squareds for individual regressions of return on expected and unexpected
variables, respectively. All variables are contemporaneous. Variables derived from order flow are displayed on the left of each
panel, while those on the right are derived from volume. In all cases, the correlation between S EO and S UO is zero. O, order
flow; S EO, signed expected order flow; S UO, signed unexpected order flow; S V, signed volume; S EV, signed expected volume;
S UV, signed unexpected volume.
 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401 399

Table 9
EUR(DEM)/JPY

Part (a) 1998

S EO and S UO 0.35 0.24 S EV and S UV
O 0.34 0.23 S V
S EO 0.21 0.2 S EV
S UO 0.16 0.04 S UV

Part (b) 1999

S EO and S UO 0.37 0.28 S EV and S UV
O 0.36 0.28 S V
S EO 0.22 0.2 S EV
S UO 0.21 0.1 S UV

R-squareds: The table is based on the full 24-h global spot EUR(DEM)/JPY data from EBS. Part (a) uses pre-EMU data from
1998, whereas part (b) uses post-EMU data from 1999. Each R-squared is computed using approximately 7000 observations
per variable, sampled at the 5-min frequency. In each panel, the first row shows the R-squared for regressions of return on
expected and unexpected components. The second row shows a regression of return on original order flow (O) and signed
volume (S V). The third and fourth rows show the R-squareds for individual regressions of return on expected and unexpected
variables, respectively. All variables are contemporaneous. Variables derived from order flow are displayed on the left of each
panel, while those on the right are derived from volume. In all cases, the correlation between S EO and S UO is zero. O, order
flow; S EO, signed expected order flow; S UO, signed unexpected order flow; S V, signed volume; S EV, signed expected volume;
S UV, signed unexpected volume.

some of the apparently strong relationship between private information and return suggested by the
correlation analysis. Close inspection of the data reveals that negative UO, i.e. below normal order flow,
does not have a consistent sign relationship with return. However, positive UO, i.e. above normal order
flow, is consistent in its relationship with return. In other words, high-positive (negative) order flow
is consistently linked to high-positive (negative) returns. Similar conclusions apply to the relationship
between UV and returns.
Bloomfield et al. (2005) provide a potential explanation for the puzzling low-order flow and low-
volume behavior. In a simulated trading room experiment, they found that informed traders preferred
to submit limit orders when deviations from fair value were small and market orders when price
deviated greatly from fair value (see McGroarty et al., 2007 for a discussion of the Bloomfield et al.,
2005 model in the electronic inter-dealer spot FX market context). This insight suggests that informed
traders may alternate between limit orders and market orders, especially when profit opportunities
are low. Such behavior would mask the direction of order flow and cause the regression analysis to

Table 10
EUR(DEM)/CHF

Part (a) 1998

S EO and S UO 0.29 0.19 S EV and S UV
O 0.28 0.19 S V
S EO 0.16 0.14 S EV
S UO 0.16 0.06 S UV

Part (b) 1999

S EO and S UO 0.28 0.21 S EV and S UV
O 0.26 0.21 S V
S EO 0.17 0.15 S EV
S UO 0.18 0.08 S UV

R-squareds: The table is based on the full 24-h global spot EUR(DEM)/CHF data from EBS. Part (a) uses pre-EMU data from
1998, whereas part (b) uses post-EMU data from 1999. Each R-squared is computed using approximately 7000 observations
per variable, sampled at the 5-min frequency. In each panel, the first row shows the R-squared for regressions of return on
expected and unexpected components. The second row shows a regression of return on original order flow (O) and signed
volume (S V). The third and fourth rows show the R-squareds for individual regressions of return on expected and unexpected
variables, respectively. All variables are contemporaneous. Variables derived from order flow are displayed on the left of each
panel, while those on the right are derived from volume. In all cases, the correlation between S EO and S UO is zero. O, order
flow; S EO, signed expected order flow; S UO, signed unexpected order flow; S V, signed volume; S EV, signed expected volume;
S UV, signed unexpected volume.
400 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401

understate the role played by private information in price formation, while the correlation analysis
would get closer to the true relationship because it ignores direction.

5. Conclusions

This paper presents compelling evidence that our theoretical understanding of how this market
works is generally correct. Strong support is shown for 15 theory-based bilateral relationships between
6 trading variables: bid–ask spread, return volatility, expected volume, unexpected volume, expected
order flow and unexpected order flow. These results are robust across 10 data samples, made up
of 5 exchange rates which were sampled from 2 separate time periods. McGroarty et al.’s (2006)
characterization of the bid–ask spreads as being very different from that normally assumed in the
market microstructure literature is strongly vindicated by the evidence.
Our analysis documents the M-shaped intraday pattern of order flow and identifies that this closely
tracks the intraday pattern for trading volume. This scenario also accords with Garman’s (1976) asser-
tion that order flow arises from random temporal imbalances between supply and demand, rather
than being a strict proxy for private information, as might be inferred from Evans and Lyons (2002).
This uninformed order flow disturbs price, contributing to return volatility, as is consistent with Clark’s
(1973) MDH model. This contrasts with financial markets which have market makers at their core, who
suppress this particular source of return volatility by ‘shading’ quote prices in order to buy back or sell
on any acquired inventory, as Amihud and Mendelson (1980) describe.
While it may indeed be less volatile, trading with market makers will be more expensive than
trading on electronic market maker-less venues because market makers must be compensated via
wider bid–ask spreads. This study shows that, while electronic inter-dealer spot FX bid–ask spreads
are always of a low-general level, they widen and narrow with the normal ebb and flow of trading
activity over the day, i.e. they are narrowest when expected volume is high and widen when expected
volume falls. McGroarty et al. (2006) show that this is not due to changes in adverse selection risk
facing market makers, but is attributable to volume-driven variations in the frequency at which the
drift component of the underlying price is sampled.
Utilizing Easley and O’Hara’s (1992) notion that price signals are associated with deviations from
normal trading levels, we use unexpected order flow and unexpected volume as proxies for private
information. Both proxies show that private information has no discernable effect on the bid–ask
spread in this market, as predicted. This is consistent with McGroarty et al.’s (2007) assertion that
bid–ask spreads in this type of market have no adverse selection risk component and that, instead,
informed trading passes straight through to return volatility. Furthermore, we show that this private
information based return volatility is a separate, additional, source of volatility and is larger than
the other (uninformed) return volatility which arises from normal trading activity alone. According
to McGroarty et al. (2006), both these sources of return volatility radiate outwards from the inter-
dealer market to permeate the entire foreign exchange market. This sheds some light on the excess
exchange rate volatility puzzle identified by Flood and Rose (1995), by suggesting that some of this non-
macroeconomic volatility is due to market microstructure effects and that a portion results directly
from the way the market is structured, i.e. this source of volatility would not arise under alterna-
tive market structures. However, this does not necessarily imply that market structure contributes to
exchange rate volatility measured at lower frequencies because much of this high-frequency volatility
may dissipate when returns are aggregated.

References

Admati, A.R., Pfleiderer, P., 1988. A theory of intraday patterns: volume and price variability. Review of Financial Studies 1, 3–40.
Amihud, Y., Mendelson, H., 1980. Dealership market: market making with inventory. Journal of Financial Economics 8, 31–53.
ap Gwilym, O., Sutcliffe, C., 1999. High-frequency financial market data: sources, applications and market microstructure. Risk
Books.
Baillie, R.T., Bollerslev, T., 1990. Intraday and inter-market volatility in foreign exchange rates. Review of Economic Studies 58,
565–585.
Bank of England, 2004. The foreign exchange and over-the-counter derivatives markets in the UK. Bank of England Quarterly
Bulletin 44, 470–484.
 F. McGroarty et al. / Int. Fin. Markets, Inst. and Money 19 (2009) 387–401 401

Bank of International Settlements (BIS), 1998. Central bank survey of foreign exchange and derivative market activity in April
1998. Bank of International Settlements.
Bank of International Settlements (BIS), 2001. Central bank survey of foreign exchange and derivative market activity in April
2001. Bank of International Settlements.
Bessembinder, H., 1994. Bid-ask spreads in the interbank foreign exchange market. Journal of Financial Economics 35, 317–348.
Bloomfield, R., O’Hara, M., Saar, G., 2005. The ‘make or take’ decision in an electronic market: evidence on the evolution of
liquidity. Journal of Financial Economics 75, 165–200.
Brock, W.A., Kleidon, A.W., 1992. Periodic market closure and trading volume: a model of intraday bids and asks. Journal of
Economic Dynamics and Control 16, 451–489.
Clark, P.K., 1973. A subordinated stochastic process model with finite variance for speculative prices. Econometrica 41, 135–155.
Copeland, T., 1976. A model of asset trading under the assumption of sequential information arrival. Journal of Finance 31,
1149–1168.
Danielsson, J., Payne, R., 2001. Measuring and explaining liquidity on an electronic limit order book: evidence from Reuters
D2000-2. Working Paper. Financial Markets Group, London School of Economics.
Docking, D.S., Kawaller, I.G., Koch, P.D., 1999. Mid-day volatility spikes in US futures markets. Journal of Futures Markets 19,
195–216.
Easley, D., O’Hara, M., 1992. Time and the process of security price adjustment. Journal of Finance 47, 577–605.
Epps, T., Epps, M., 1976. The stochastic dependence of security price changes and transaction volumes: implications for the
mixture of distributions hypothesis. Econometrica 44, 305–321.
Evans, M., Lyons, R., 2002. Order flow and exchange rate dynamics. Journal of Political Economy 110, 170–180.
Flood, M., Rose, A., 1995. Fixed exchange rates: a virtual quest for fundamentals. Journal of Monetary Economics 36, 3–37.
Garman, M., 1976. Market microstructure. Journal of Financial Economics 3, 257–275.
Gencay, R., Gradojevic, N., Selcuk, F., 2007. When do informed traders arrive in foreign exchange markets? Working Paper.
Available at SSRN.
Hartmann, P., 1998. Currency Competition and Foreign Exchange Markets. Cambridge University Press.
Jennings, R., Starks, L., Fellingham, J., 1981. An equilibrium model of asset trading with sequential information arrival. Journal
of Finance 36, 143–161.
Jorion, P., 1996. Risk and turnover in the foreign exchange market. In: Frankel, J.A., Galli, G., Giovannini, A. (Eds.), The Microstruc-
ture of Foreign Exchange Markets. University of Chicago Press, pp. 19–37.
Kyle, A.S., 1985. Continuous auctions and insider trading. Econometrica 53, 1315–1335.
Low, A.H.W., Muthuswamy, J., 1996. Information flows in high frequency exchange rates. In: Dunis, C. (Ed.), Forecasting Financial
Markets: Exchange Rates, Interest Rates and Asset Management. John Wiley, pp. 3–32.
McGroarty, F., ap Gwilym, O., Thomas, S., 2006. Microstructure effects, bid–ask spreads and volatility in the foreign exchange
market pre and post-EMU. Global Finance Journal 17, 23–49.
McGroarty, F., ap Gwilym, O., Thomas, S., 2007. The components of electronic inter-dealer spot FX bid–ask spreads. Journal of
Business Finance and Accounting 34, 1371–1650.
Roll, R., 1984. A simple implicit measure of the effective bid–ask spread in an efficient market. Journal of Finance 39, 1127–1139.
Tauchen, G., Pitts, M., 1983. The price-variability volume relationship on speculative markets. Econometrica 51, 485–505.