ASYMMETRIC INFORMATION AND THE CROSS-SECTION OF
Alexander Mende, University of Hannover, Germany
Lukas Menkhoff, University of Hannover, Germany
Carol L. Osler, Brandeis University, USA *
This paper shows that the spreads charged by currency dealers vary inversely with deal size and
that they are wider for importers and exporters than for asset managers and other dealers. This
pattern is the opposite of that predicted by standard models of market making under asymmetric
information, given the information structure of currency markets. The paper suggests that private
information gives certain customers market power relative to their dealers. Symmetrically, it
suggests that dealers strategically quote narrower spreads to privately informed customers to
increase their access to information. Finally, the paper suggests that dealers primarily seek
information about transitory market developments rather than fundamentals.
Corresponding author: Lukas Menkhoff, Department of Economics, University of Hannover, D-30167 Hannover,
Germany, email@example.com; Alexander Mende, Department of Economics, University of Hannover,
Königsworther Platz 1, D-30167 Hannover, Germany, firstname.lastname@example.org; Carol Osler, Brandeis Interna-
tional Business School, Brandeis University, 415 South Street, Waltham, MA 02454, USA, email@example.com.
We are grateful for helpful comments from Alain Chaboud, William Clyde, Peter Eggleston, Thomas Gehrig, Val-
erie Kraus, Peter Nielsen, Dagfinn Rime, Erik Sirri, Erik Theissen, and Peter Tordo, along with participants at sev-
eral seminars and the Stockholm Workshop on FOREX Microstructure and International Macroeconomics. We are
deeply indebted to the bankers who provided and discussed the data with us.
ASYMMETRIC INFORMATION AND THE CROSS-SECTION OF
This paper examines the cross-section of currency spreads. From the seminal papers on
dealing under asymmetric information we learn that spreads compensate market makers for their
losses to privately informed counterparties (Copeland and Galai, 1983; Glosten and Milgrom,
1985; Easley and O'Hara, 1987). Thus, spreads should rise with the likelihood that a given
counterparty has private information, other things equal, a likelihood that depends on deal size
and counterparty type. Spreads should vary positively with deal size because, as shown by
Easley and O'Hara (1987), informed traders have an incentive to undertake larger trades. Spreads
should be narrower for counterparty types considered less informed if dealers can discriminate
among them. Since currency dealers generally do know their counterparty's type the narrowest
currency spreads should therefore be enjoyed by importers and exporters ("commercial
customers"), since they are considered less informed than asset managers and hedge funds
(jointly "financial customers") or other dealers.
This paper shows that spreads charged by currency dealers conform to the opposite of
these two predictions: They are inversely related to deal size and are wider for commercial
customers than for financial customers and other market makers. The resulting variation in
spreads is substantial. Average spreads vary from about 2 pips for large interbank deals to about
20 pips for small deals with commercial customers.1 The paper suggests two mutually consistent
explanations for this pattern, both based on asymmetric information. According to the first,
1 A pip is the smallest unit in an exchange-rate quote. For USD/EUR one pip equals USD 0.0001.
private information gives certain customers market power relative to their dealers. According to
the second, dealers strategically quote narrower spreads to privately informed customers to
increase the chances of transacting with them, thereby gathering some of their information. The
paper also notes that, even though the familiar models cited above do not explain how the
absolute size of currency spreads varies cross-sectionally, they do explain why the share of the
asymmetric information component of spreads is positively related to deal size and is lowest for
Our data comprise the entire USD/EUR transaction record of a bank in Germany over four
months in 2001. These data have two advantages relative to other currency transaction data
analyzed in the literature: they distinguish between financial and commercial customers, and
they cover a longer time period. The earliest study of currency spreads using transaction data,
Lyons (1995), concludes that spreads on interdealer transactions vary positively with deal size,
as predicted by Copeland and Galai (1983), Glosten and Milgrom (1985), and Easley and O'Hara
(1987) ("the standard models"). More recent studies generally find little relation between
currency spreads and deal size (Yao, 1998; Bjønnes and Rime, 2003).
Though not previously identified in the literature, the negative relationship between
currency spreads and deal size found here is not just widely known, it is deeply woven into the
fabric of thought in the practitioner community. A similar relationship holds in the U.S.
municipal bond market (Green, Hollifield, and Schurhoff, 2004; Harris and Piwowar, 2004). By
contrast, the relationship between spreads and deal size is positive in U.S. equity markets
(Peterson and Sirri, 2003).
The pattern of spreads in stock markets can presumably be explained by the adverse
selection costs highlighted in the standard models. Indeed, real-world stock markets provided the
inspiration for these models. To explain the pattern of spreads in municipal bond markets, Green,
Hollifield and Schurhoff (2004) highlights that market's dealership structure and the associated
dispersion of information. In opaque markets, agents with knowledge of current market
conditions, like dealers, have market power. Green, Hollifield and Schurhoff (2004)
hypothesizes that agents making small municipal deals are tend to be poorly informed about
market conditions, a tendency that gives dealers the market power to extract wider spreads. This
logic can be apply directly to explain the inverse relationship between spreads and deal size in
currency markets, since small currency transactions also tend to be undertaken by uninformed
Green, Hollifield and Schurhoff's (2004) analysis can also explain our finding that financial
customers in currency markets pay lower currency spreads than commercial customers, since
financial customers are generally well informed about market conditions while commercial
customers are not. In addition, financial customers have private information about their own
large trades and large stop-loss and take-profit orders (Osler, 2003).2 Since deal flow is a key
determinant of exchange-rate dynamics (Evans and Lyons, 2002), and financial deal flow in
particular appears to be a key factor in extreme high-frequency moves (Fan and Lyons, 2003),
information about large trades and large orders can be critical to market makers.3 This private
information enhances financial customers' market power relative to their dealers and allows them
to demand narrower spreads.
2 Stop-loss and take-profit orders are conditional market orders, where the conditioning variable is market price. A
stop-loss order instructs a dealer to buy (sell) a specific amount at market prices if and only if the market price rises
(falls) to a certain pre-specified level. A take-profit order instructs a dealer to sell (buy) a specific amount at market
prices if and only if the market price rises (falls) to a certain pre-specified level. Orders are distinct from regular
deals, in which a market maker provides a two-way quote and the counterparty chooses whether to deal at those
3 Deal flow is defined as the net of buy-initiated and sell-initiated deals over a given interval.
The benefits from learning financial customers' private information provide a strategic
incentive for dealers to quote narrow spreads to such customers. When market makers compete
with each other as in dealerships markets, the more attractive a dealer's quoted spreads the more
likely customers are to transaction with that dealer, both now and in the future, and the more
chances that dealer will have to learn about their trading activity. That is, currency dealers may
not passively accept the information content of deal flow, as assumed in standard models, but
may instead set prices strategically to increase their access to information. This strategic dealing
hypothesis complements, and is fully consistent with, the market power hypothesis.
The possibility of strategic dealing was originally explored in theoretical papers by
Gammill (1989) and Leach and Madhavan (1992, 1993), which show that dealers may rationally
adjust prices in early transactions with the goal of becoming more informed, and thus more
profitable, in later transactions. In the Leach and Madhavan models, specialists rationally quote
wider spreads at the beginning of a trading session, driving informed trades out. In Gammill's
model, dealers choose to take a loss on early trades, driving informed trades in. Our hypothesis is
closest to Gammill's.
The hypothesis that currency dealers strategically subsidize deal flow with informed
customers raises the question: Are currency dealers seeking to gain fundamental information or
information about transient market developments? Consistent with the hypothesis that dealers
seek fundamental information (Evans and Lyons, 2002, 2004), we show that interbank and
financial-customer deal flow are both positively cointegrated with exchange rates. Nonetheless,
we propose that dealers primarily seek information about transient market developments ("non-
fundamental information"), and provide three justifications for this view. First, fundamentals are
primarily relevant in the long run, but currency dealers generally close their positions by the end
of the day. Second, dealers themselves characterize the information they seek as relating to
transient developments. Third, there may be relatively little fundamental information to be
gleaned from currency trades. Exchange-rate fundamentals are generally considered to be broad
macroeconomic aggregates such as money supplies and price levels, information about which is
in the public domain. By contrast, much market-relevant information about equity and municipal
bond issuers is unearthed by equity analysts but never becomes public.
To illustrate the type of non-fundamental information currency dealers may seek we note
that intraday currency deal flow is correlated, even though daily returns are not. As shown by
Goodhart, Ito, and Payne (1996), there tend to be runs of buy orders and of sell orders.
Presumably these runs often reflect large trades, which are typically divided into many smaller
transactions. By subsidizing transactions with the largest customers, a market maker may raise
the likelihood of participating in large trades and being informed about the associated runs.
The difference between spread determination in currency and equity markets can be
summarized with a cost-benefit analysis of market making. Standard adverse selection costs may
be lower for currency dealers because private fundamental information is less common in foreign
exchange markets. Meanwhile, currency dealers may benefit from trading with customers who
have private information about transient aspects of deal flow like large trades.
The bank from which this paper's data are derived is relatively small. Nonetheless, there
are three reasons why our conclusions should generalize to the overall currency market. First, the
intense competition in major currency markets means that any bank's pricing practices should
accurately represent practices at all banks. Second, traders from large banks tell us that their
pricing policies conform to those described here: supporting quotes from two market participants
are provided below. Third, our small bank behaves similarly to large banks in many other
dimensions. Indeed, a secondary contribution of the paper is to show this consistency between
the behavior of small and large currency dealers.
The paper proceeds as follows: Section 2 describes our data and shows how our small
bank's pricing and inventory management practices parallel those at large banks. Section 3
provides our core results that currency spreads vary inversely with deal size and tend to be larger
for commercial customers than financial customers and other dealers. This section also shows
that these results can be explained in terms of the market power and strategic dealing hypotheses.
Section 4 discusses whether currency dealers strategically seek fundamental or non-fundamental
information. Section 5 shows that the share of the asymmetric information component of spreads
is largest when counterparties are most likely to be informed, consistent with standard models.
Section 6 concludes.
2. Small banks and large banks
This section describes our transactions data and provides a preliminary comparison of our
bank's pricing and inventory management practices and those at large banks. It also presents the
model on which we base our core results, which are presented in the next section. We find that
our bank is indeed small relative to others examined in the literature, but that its behavior is
nonetheless consistent with that of large banks in recent years. Readers familiar with the model
and willing to trust that small and large banks behave similarly can safely skip to Section 3,
which presents our central results.
Our data comprise the complete USD/EUR transaction record of a bank in Germany over
the 87 trading days from 11 July, 2001 to 9 November, 2001. Though our data technically refer
to the overall bank, they are an accurate reflection of a single dealer's behavior because only one
dealer was responsible for the bank's USD/EUR trading. For each transaction we have the
following information: (1) the date and time;4 (2) the direction (customer buys or sells); (3) the
quantity; (4) the transaction price; (5) the type of counterparty: bank, financial customer,
commercial customer, preferred customer; (6) the initiator; (7) the forward points if applicable.
Table 1 provides basic descriptive statistics.5
We include outright forward trades, adjusted to a spot-comparable basis by the forward
points, as recommended by Lyons (2001). Since forward transactions account for 20 percent of
all trades, their inclusion could impede direct comparisons between our results and those of most
earlier papers, which focus exclusively on spot trades. Reassuringly, our main qualitative
conclusions are sustained when forward transactions are excluded.
The bank's inventory position is inferred by cumulating successive transactions. Following
Lyons (1995), we set the daily starting position at zero. This should not introduce significant
distortions since our dealer keeps his inventory quite close to zero. As shown Figure 1, which
charts the dealer's inventory over the sample period, the average inventory position is EUR 3.4
million during the trading day and only EUR 1 million at the end of the day.
Our ability to distinguish among customer types is almost unique in currency transaction
data. Lyons (1995) only uses data on interbank trading; Yao (1998) uses customer trade data but
does not generally distinguish among customer types; Bjønnes and Rime (2004) have insufficient
customer transactions to perform a detailed analysis; Carpenter and Wang (2003) have such
information but lack inventory data to test the relevant models; finally, Lyons (2001) and Fan
and Lyons (2003) can distinguish among customer types but only in daily data.
4 The time stamp indicates the time of data entry and not the moment of trade execution, which will differ slightly.
Nevertheless, there is no allocation problem because all trades are entered in a strict chronological order.
5 We exclude trades with "preferred customers", typically commercial customers with multi-dimensional
relationships with the bank, because these customers' spreads may reflect cross-selling arrangements and because
their trades are typically very small (average size EUR 0.18 million). We also exclude a few trades with tiny
volumes (less than EUR 1,000) or with apparent typographical errors.
A preliminary comparison of our dealer with the other large dealers described in the
literature is provided in Table 2. The dimensions in which our dealer is small include total daily
trading value, average transactions per day, and average inventory position. Our dealer is
comparable in size to a NOK/DEM dealer employed by a large dealing bank examined in
Bjønnes and Rime (2004). Small dealing banks are far more common than large ones (B.I.S.,
2002), so our bank is probably a reasonably good representative of the average currency-dealing
bank. Nonetheless, big banks are thought to dominate such dealing.
The small size of our bank is also reflected in the prominence of customer deals,
especially those with commercial customers (Table 1). Our bank’s customer business is 23
percent of its spot trading value. Though this does not differ much from the 33 percent share of
customer business at all foreign exchange banks (B.I.S., 2002), it greatly exceeds the customer
shares reported for bigger dealers, which range from zero percent (Lyons, 1995) to 14 percent
(Yao, 1998). Commercial customers generate roughly twice the business of financial customers
at our bank, by value. By contrast, commercial customers do roughly half the business of
financial customers (B.I.S., 2002) in the foreign exchange market overall.
The small size of our bank is also reflected in the large mean absolute change in
transaction price between successive deals, 10.7 pips. This presumably reflects the relative
infrequency of transactions at our small bank as well as the high proportion of small commercial
customer deals, which tend to have wide spreads (as we show in Section 3). Table 3 provides
information on the size distribution of our dealer’s transactions.
2.2. Pricing and inventory management practices
Despite the small size of our bank, there are a number of reasons to believe our qualitative
conclusions generalize to the entire currency market. First, currency markets are extremely
competitive. Hundreds of banks deal in the major currency pairs and even the largest dealer's
market share is only on the order of 10 percent. In such a market, the behavior of any agent
should accurately represent the behavior of all agents. Second, market participants consistently
confirm that the patterns we identify are correct. Third, our small bank's pricing and inventory
management strategies are generally consistent with those documented in recent years for large
banks, as we show next.
Lyons' (1995) version of the Madhavan-Smidt model of market making (1991) has been
widely used in subsequent studies of currency dealers (e.g., Yao, 1998; Bjønnes and Rime,
2004). The model assumes a representative dealer in a competitive market whose counterparty
has private information about the asset's fundamental value. The model's agents are fully rational
and there is a detailed informational setting. Agent j calls dealer i requesting a quote on amount
Qjt; that amount is determined as follows:
Xjt represents agent j's nonspeculative need for currency, which constitutes agent j's private in-
formation. The term µjt represents agent j's expectation of the asset's true value, conditional on Xjt
and on public information. Pit, dealer i's regret-free price, is determined as follows:
t it it it it
Here, Iit is dealer i's inventory at the beginning of period t, I*it is his desired inventory, and Dt is
the direction of trade [Dt = 1 (-1) if agent j is a buyer (seller)].
After solving for conditional expectations and taking first differences, one arrives at the
following expression for the price change between incoming transactions, ∆Pit = Pit - Pit-1:
The model predicts that β1 > | β 2| > 0 > β 2 and that | β 2| equals the baseline half-spread,
meaning the half-spread that would apply before adjustment for deal size or existing inventories.
The model assumes that dealers shade prices to help manage existing inventory (e.g., dealers
lower prices in response to high inventory), implying γ2 > 0 > γ1.
According to the model, the coefficient on deal size should be positive, reflecting adverse
selection considerations: spreads should be wider for larger deals because they are more likely to
be undertaken by privately informed agents (Easley and O'Hara, 1987). However, a positive
coefficient on deal size could also capture a second type of inventory concerns: as shown in Ho
and Stoll (1981), larger deals leave market makers with higher inventory and thus greater
inventory risk, so they should carry wider spreads. We call this a “prospective” inventory effect,
since it concerns inventories that may arrive if the counterparty decides to deal at the current
quote and because we need to distinguish it from the effect of existing inventory captured by γ1
and γ2. The adverse selection and prospective inventory effects both predict δ > 0 and are
observationally equivalent in this setting.
The model is typically estimated using generalized method of moments with Newey-West
correction for heteroskedasticity (e.g., Yao 1998; Bjønnes and Rime, 2004). We first estimate
Equation (3) without discriminating among counterparties (Table 4A, column 1), and then re-
estimate it distinguishing interdealer transactions and customer transactions by interacting
dummy variables for each with the direction, inventory, and deal size variables (Table 4B,
column 1). Since existing inventories appear to have no influence we re-run both regressions
excluding inventories. We also re-run the regressions using only spot transactions. The results
are robust to these changes, as shown in columns 3 and 4 of each panel of the table.
We compare these results with the results of similar regressions using large-bank data
reported in earlier studies. This comparison indicates that dealer behavior is consistent across
dealers of all sizes in three dimensions: baseline spreads, the influence of existing inventories,
and the relationship between deal size and spreads. We discuss each dimension in turn.
Baseline spreads: Our bank's average baseline half-spread for interbank transactions is
about 1.5 pips (Table 4B), similar to estimates from other studies. For example, Goodhart et al.
(2002) finds that the average spread for USD/EUR transactions on Electronic Brokerage Service
(EBS, one of the two major electronic brokerage systems for interbank trading) was 2.8 pips
about one year after the euro was introduced. Our bank's average half-spread for customer deals,
9 pips, is much higher than its interdealer spread. Bjønnes and Rime's (2001) NOK/DEM dealer
also makes a sharp distinction between dealers and customers.
Influence of existing inventories: Our results indicate that existing inventories have no
influence on the prices our dealer quotes to other dealers, consistent with recent studies of large
banks (Yao, 1998; Bjønnes and Rime, 2004). By contrast, Lyons (1995) provides evidence that
his dealer did engage in inventory-based price shading towards other dealers in 1992. This may
reflect the unusual character of Lyons' dealer who, as a jobber, dealt exclusively with other
dealers at extremely high frequency. Yao (1998) claims that his dealer avoided such shading
because it would reveal information about his inventory position.
Bjønnes and Rime (2004) argue that the apparent shift away from inventory-based price
shading over the 1990s may reflect the way the interbank market shifted rapidly to a heavy
reliance on electronic brokerages after their introduction in the mid-1990s (Melvin and Wen,
2003).6 Indeed, our dealer reports that for interbank trades he generally uses EBS because it is
less expensive and faster than direct interbank dealing.7 Together, this these observations imply
that our dealer controls inventories via interbank trading instead of price shading, a conclusion
we support empirically later in this section.
The estimates seem to provide slight evidence of price shading with respect to customers,
but the shading seems to go the "wrong" way. Reassuringly, this can be traced to one trade
carried out in the first month of our sample period. When that month is excluded, both
coefficients are insignificant.
Trade size and spreads: The coefficient on deal size is statistically insignificant for
interbank trades, suggesting that neither information asymmetries nor prospective inventories
cause large interbank deals to be priced less attractively than small deals. This is consistent with
recent empirical depictions of interbank trading at large banks. Bjønnes and Rime (2004) finds
that spreads are independent of deal size for the brokered trades that now dominate such trading.
It also finds that spreads rise with deal size for direct interbank transactions, a distinction that
makes economic sense. Dealers have limited control over the relationship between deal size and
spread for brokered transactions, but they have full control for direct deals. Notably, the earliest
studies of currency dealers (Lyons, 1995; Yao, 1998), which do not control for the distinction
between direct and brokered trades, found that interbank spreads do rise with deal size,
6 In a direct interbank trade, one bank calls another and asks for a two-way quote for a specific amount. Electronic
brokers take limit orders from dealers and post the best bid and ask prices. Limit orders are then crossed with deal-
ers' market orders.
7 This preference is supported by the transactions data. Our dealer's mean interbank transaction size was only EUR
1.42 million (Table 1), the maximum interbank trade size was only EUR 16 million, and the standard deviation of
these trade sizes was only 1.42. These small values are consistent with heavy use of EBS, where the mean
USD/EUR transaction size in August 1999 was EUR 1.94 million and the standard deviation of (absolute) transac-
tion sizes was 1.63 million. By contrast, interbank deals averaged closer to USD 4 million prior to the emergence of
electronic brokerages (Lyons, 1995).
consistent with standard models. This could reflect the fact that interbank trading was mostly
carried out through direct transactions until the late 1990s.
The coefficient on deal size is also insignificant for customers in our baseline regression.
We note in passing that this coefficient is negative and significant when inventories are
excluded. Section 3 shows that the overall relationship between spreads and transactions sizes is
indeed negative for customer transactions.
2.2.2. Inventory management
Our dealer's tendency to keep inventories close to zero (Figure 1) is itself similar to
inventory management practices at large banks. As Table 1 shows, currency dealers of all sizes
tend to keep minimal inventories. A more rigorous description of our dealer's approach to
inventory management comes from the following regression:
If the dealer instantly eliminates unwanted inventories, ρ ≈ -1. If the dealer allows his inventory
to change randomly, ρ = 0.
Results from estimating Equation (4), once again using GMM with Newey-West correction
for heteroskedasticity, are presented in Table 5. They confirm that our small bank does actively
keep inventories close to zero. The negative and statistically significant coefficient on lagged
inventory implies that our dealer typically brings inventories halfway back to zero within 19
minutes of an inventory shock. This is quite close to the 18-minute median inventory half-life for
Bjønnes and Rime's (2004) NOK/DEM dealer. By contrast, the median inventory half-lives of
that bank's DEM/USD dealers are only 0.7 to 3.7 minutes.
Since our dealer does not seem to use price shading to control inventory, it seems likely he
controls it through the interbank market, instead. This would be consistent with at least two of
14 Download full-text
the dealers at the large bank examined in Bjønnes and Rime (2004), including the NOK/DEM
dealer. To examine this possibility, we carry out a probit analysis of the probability that a given
trade is outgoing:
Prob(Tradet=IBout) = P(|Iit|, Iit2, |Qjt|, IB|Qjt-1|, FC|Qjt-1|, CC|Qjt-1|) . (5)
If dealers are more aggressive in eliminating large inventories than small ones the coefficient on
the absolute value of inventory, |Iit|, will be positive. If dealers automatically eliminate inventory,
this coefficient will be insignificant. We include squared inventory, Iit2, to capture nonlinearities
in this relationship. The absolute transaction size, |Qjt|, may capture technical aspects of dealing
described below. The variable IB|Qjt-1| is an interaction term between absolute transaction size
and a dummy that equals unity if the previous transaction was an incoming interbank (IB) deal
and zero otherwise; FC|Qjt-1|, and CC|Qjt-1| are defined accordingly for financial customer (FC)
and commercial customer (CC) transactions. Coefficients on these variables should be positive if
outgoing transactions are customarily used to eliminate unwanted inventory. We allow the
coefficients to vary according to counterparty type because one might expect dealers to be
relatively aggressive in eliminating inventory accumulated in deals with more informed
The results of estimating Equation (5), shown in Table 6, indicate that the likelihood of an
outgoing deal rises with the absolute amount of existing inventory, and that the relationship is
convex. The positive overall relationship implies that our dealer does rely on outgoing
transactions to manage his inventory, consistent with the large dealers of major currency pairs
analyzed by Bjønnes and Rime (2004).
We find a positive relationship between absolute deal size, |Qjt|, and the likelihood that the
deal itself is outgoing, which indicates that the transactions submitted to the brokers tend to be