Do Top Trading Banks in FOREX Business Know More?

Do Top Trading Banks in FOREX Business

Know More?

Kate Phylaktis* and Long Chen

Cass Business School London

JLE: F31

Keywords: Market Microstructure; Foreign Exchange Rates; Foreign Exchange Markets; Macroeconomic Announcements; Price Discovery

July 2006

Abstract

This study investigates information asymmetry in the Foreign Exchange Market by examining the information share of the top 10 trading banks and their rivalling counterparts in the Reuters EFX system. Using 5 years of indicative GBP-$US data, we find that the top 10 banks, among around 100 quoting banks in the market, have a monthly average share of over 70% of total market information, and that increases to around 80% during some U.S. macro announcements. These results suggest the possibility of private information on public news in the Foreign Exchange Markets.

* Corresponding author. Cass Business School, 106 Bunhill Row, London EC1Y 8TZ, United Kingdom, tel.: + 44 20 7040 8735, fax: + 44 20 7040 8881, email: K.Phylaktis@city.ac.ukK.Phylaktis@city.ac.uk

1. Introduction

Though Foreign Exchange Market is the largest financial market, with a daily turnover of USD 1.9 trillion1, it is not necessarily a market of perfect competition. According to Euromoney (May, 1995) in 1994, the largest 10 foreign exchange banks accounted for 45% of global foreign exchange business. Consolidation in the banking sector since then brought further concentration into foreign exchange market. BIS triennial survey (2005) reports a substantial decline in the number of banks accounting for 75% of local turnover since 1995. In the U.S., there were only 11 banks conducting 75% of FX market transactions in 2004, compared to 13 banks in 2001, and to 20 banks in both 1998 and 1995. The same trend is observed in UK foreign exchange market. Such concentration suggests that top banks in FX business may exert greater impact on the price formation process than the relatively smaller banks. Cheung and Chinn’s (2001) survey suggests that 50% of the participants agree that there are dominant players in GBP-$US market.

Such findings motivate our investigation in the information asymmetry in the spot foreign exchange market by examining the information share of the top trading banks2 relative to that of their counterparts. Using indicative GBP-$US data over the period of January 1994 to December 1998, we find that the top 10 banks, among around one hundred quoting banks, have a dominant share (monthly average of over 70%) of the price information in the Reuters EFX system. We also look at the impact of U.S. macro announcements in top banks’ information share. After testing 21 categories and 1035 items of U.S. announcements, we find that during some categories of news announcements, the top 10 banks’ information share is further expanded to around 80%, which suggests the possible existence of private information on public news in the Foreign Exchange Markets.

The dominant banks’ information advantage over their rivalries could be interpreted under a microstructure framework. In this framework, transactions play a

1

BIS triennial survey 2005.

2

central, causal role in price determination3. In FX market prices are determined collectively by macro economic factors such as, economic growth, consumption, unemployment, interest rates and financial markets performance etc. Information on most of them is not directly observable on a real-time base, but dispersed widely among heterogeneous market players. To form a better picture of the macro economic information, one needs to jigsaw together the dispersed partial information. In FX markets, customer order flow help trading banks to aggregate this dispersed information and ‘feel’ the general movements of the economy, as suggested by numerous studies4. For instance, Lyons (1995) finds that international trades and services generated FX order flows help dealers to ‘know’ the trade balance figures long before the statistics are published. One hence may expect that the more order flow a trader processes, the more information she could garner from it. Therefore, the dominant role of major banks in the global FX transactions may well cause a dominant share of market information.

Though a more recent literature addresses the currency market and its response to news as a joint quantity and price response5, in contrast to previous investigations which focused solely on price, most of them suffer from a few drawbacks. Firstly they could not reveal the market information and news impact on a heterogeneous agent’s base. Due to commercial confidentiality reasons, the data on prices and order flows are aggregated at certain lower frequency with no information on the identity of the trading banks. However, some studies do find that at least different types of dealers exert a different impact on prices due to distinctly separate trading motivations (e.g. Fan and Lyons (2000) and Evans (2002)). Another drawback is that some studies could not examine high frequency price discovery due to daily or even monthly frequency of their data sets. Other problems include having a short window of no more than a few months or partial data sets that only cover a few dealers. These issues present difficult obstacles for the investigation of the FX market microstructures.

3

See, e.g., Glosten and Milgrom (1985); Kyle (1985).

4

See, e.g., Lyons (1995), Yao (1998), Bjonnes and Rime (2000), Rime (2001) and Payne (2003).

5

Our work has four novel features compared to previous research in this area. First, our study is one of the first to test general macro news effects under the heterogeneity market assumption, in contrast to previous work that either implies the market homogeneity assumption, or only tests for the central bank intervention as public news. We use a unique GBP_USD database, which identifies the quoting banks and enables us to group the top banks according to the Euromoney survey. Second, we use a grouping method, i.e. we divide market participants into two groups, one, which comprises the top banks and the other the rest of the banks, which catches the major factors that contribute to market heterogeneity. Thirdly, we apply both the Hasbrouck (1995) information share technique and the Gonzalo and Granger (1995) common component method, as complementary approaches to calculate the information shares, instead of relying on one of them as has been the case in previous related studies. Finally, our database is 5 years long compared to much shorter and relatively dated data of most previous work.

The structure of the remainder of the paper is as follows. Section 2, describes our methodology. Section 3, looks at the exchange rate data and macro announcements used in our tests. Section 4, reports the empirical results and discusses the implications. Section 5, concludes the paper.

2. Methodology

2.1 Group Test

We use Euromoney’s biennial market foreign exchange polls as a guide for our selection of top banks. We subsequently divide the quoting banks into two groups, the top bank group and the non-top bank group. We include the voted top 10 banks into the top group, and leave the rest of the banks as what we call the non-top group. All the quotes from the top group will be treated as one entity, and the same applies to the non-top group. Using such a method, we explore whether the top group takes up more information share during normal intraday trading time and expands its information share during the announcements.

The first 5 criteria used by Euromoney to rank the banks are reported in Table 16. Those are price and quote speed, which are directly related to information (see as Melvin and Yin (2000)); customer relationship, i.e. better relationship suggests more efficient information flow between banks and their customer; and higher credit rating and liquidity, which also suggests greater market trading capability, which enables banks to deduce more private information from customer order flow. The survey voted top ten banks from 1994 to 1998 are displayed in Table 2.

Previous papers on information sharing concentrate on individual dealer’s quotes thereafter referred to as individual tests. Compared to individual tests, the group test has several merits. First, the previous individual tests mainly relate to German central bank intervention (see e.g. Peiers (1997) and Sapp (2002)). For German central bank interventions, certain banks have advantage compared to other banks in detecting and interpreting central bank’s interventions. But such an information advantage of a certain bank does not necessarily exist in general macro announcements forecasting and interpretation. By putting more superior banks into one group, the combined private information on the interpretation of the macro news would be significantly enhanced.

Secondly, the group tests are more robust than the individual tests. The individual tests relate to certain banks and at certain periods. Due to the natural market evolution, some banks would not possess informational advantage after a year or two, and in some cases even discontinue their market presence. In addition, the competition among electronic trading systems also changes the availability of some banks’ quotes. These factors impose a challenge to long horizon tests of 5 to 10 years. In contrast, a group is a portfolio and hence is capable of constant update of its components and limits the negative impact of the above mentioned factors, giving us reliable results in the case of long horizons.

Finally, a group test is fundamentally different from the individual test. If we call the individual test as a ‘micro experiment’, then group test would be called ‘macro

6

experiment’, as it includes almost every quote of the entire market (or quoting system) into our investigation. Therefore, if the individual test tells a story about several specific banks, a group test gives us results about a class of banks and allows us to draw more meaningful lessons.

2.2 Measures of Information Share

We first need to calculate the share of price discovery of each of the two groups. If we deem each bank’s quotes as one price, all the prices of the same currency are

cointegrated I(1) variables, sharing common stochastic factors. Since we only deal

with two prices from top and non-top groups, they share one common factor.

IS (information share) and PT (permanent-transitory) models are the two most prevalent common factor models. They are directly related and the results of both models are primarily derived from the vector error correction model (VECM). They provide similar results if the VECM residuals are uncorrelated. However, if substantial correlations exist the two models usually provide different results. Hasbrouck (1995) handles this correlation by using Cholesky factorization. Therefore, the IS results are variable order dependent. Hasbrouck (1995) suggests that different orders may be used and upper and lower information share bounds can be calculated to obtain the average figure as the final information share result. However, the bounds are often very much apart since high frequency exchange rate data have a high residual correlation. Therefore in our paper, we rely more on the PT model, while treating IS model as a parallel check due to their similarity. The following estimation approaches for both models are mainly adapted from Baillie et al. (2002).

We consider the two price quotes from the two groups to be I(1) processes,

)' , ( _{1t 2t}

t p p

P = with the differential being the error correction term t

t t

t P p p

d =β' = ₁ − ₂ , where β is the cointegration vector. Both models start from the estimation of the following VECM:

t k j j t j t t P A P e P = + ∆ + ∆

∑

where α is error correction vector and et is a zero-mean vector of serially uncorrelated innovations with covariance matrix Ω

⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ = Ω ₂ 2 2 1 2 1 2 1 σ σ ρσ σ ρσ σ .

The VECM has two parts: the first part, αβ'P_t−1, represents the long-run or equilibrium dynamics between the price series, and the second part,

∑

= − ∆ k j j t j P A 1 ,

shows the short-term deviation induced by market imperfections.

Hasbrouck (1995) transforms Eq. 3.1 into a vector moving average (VMA) in an integrated form t t s s t e L e P (1) *( ) 1 ψ ψ + =

∑

where ψ(L) and ψ *(L) are matrix polynomials in the lag operator, L. If we denote ψ =(ψ₁,ψ₂) as the common row vector in ψ(1), Eq. 3.2 becomes

t t s s t e L e P ( ) *( ) 1 ψ ιψ + =

∑

where )'ι=(1,1 is a column vector of ones.

Hasbrouck (1995) states that the increment ψe_t in Eq. 3.3 is the component of the price change that is permanently impounded into the price and is presumably due to new information. If price innovations are significantly correlated across prices, Hasbrouck (1995) uses Cholesky factorization Ω=MM' to eliminate the contemporaneous correlation, where:

⎥ ⎦ ⎤ ⎢ ⎣ ⎡ = 22 12 11 0 m m m M .

Granger (1995)’s PT model, then the information shares of the two prices are: 2 22 2 2 12 2 11 1 2 12 2 11 1 1 ) ( ) ( ) ( m m m m m S γ γ γ γ γ + + + = , and 2 22 2 2 12 2 11 1 2 22 2 2 ) ( ) ( ) ( m m m m S γ γ γ γ + + = .

To finally get the information share of each group, the order of them is changed and the calculation process is repeated. The average of the two results is the final figure for information share.

Gonzalo and Granger (1995) define the common factor to be a combination of

the variables P_t, such that h_t =ΓP_t, where Γ is the common factor coefficient vector. To calculate information share from PT model, we simply calculate:

2 1 1 1 _{γ γ} γ + = S , and 2 1 2 2 _{γ γ} γ + = S .

Before we calculate the information share for each group, the usual procedures of unit root and cointegration tests are conducted. Unless otherwise stated, the price series in our empirical tests satisfy both conditions.

3. Exchange Rate and Macro News Data

3.1 Exchange Rate Data

Our data are EFX tick-by-tick GBP_USD spot quotes, as posted on the Reuters ‘FXFX’ screen, which have later been collected and filtered by the Olsen and Associates (O&A) over the period January of 1994 to December of 1998. It should be noted that our paper is the first one to use GBP_USD exchange rates which identifies the banks that made the quotes.

The EFX data are indicative quotes, which means even though the dealers may intend to trade at their quoted prices they have no commitment to do so. Goodhart, Ito

and Payne (1996) and Danielsson and Payne (1999) find that the basic characteristics of 5-minute FX returns constructed from quotes closely match those calculated from transactions prices. Phylaktis and Chen (2006) compare four months (inside this paper’s time window) of EFX data to D2000-1 transaction data, and find that EFX data are in fact superior to the latter data set by measuring the embedded information. Since around this sample period Reuters’ trading system takes more than 90% of the world’s direct inter-dealer transaction7, this finding should be very supportive of the applied data’ quality.

There are also other reasons why indicative data are more suitable than transaction data in conducting our empirical tests. First, Goodhart and O’Hara (1997) argue that indicative quotes are better than transaction prices in demonstrating traders’ heterogeneous price interpretation, as transaction price needs agreement between two parties, while indicative quote is not so restrictive. Then Hasbrouck (1995) indicates that an analysis of a stock, if based on last sale prices, would have problems of autocorrelation induced by infrequent trading. Though this issue is less serious in foreign exchange markets, the last sale prices would be less informative. Indicative quotes, on the other hand, could be updated in the absence of trades. Finally, empirical investigation using transaction data may turn out to be biased because it ignores the informational content of non-trading intervals. This sampling bias is reduced when using bid-ask quote series, which are continuously updated by the market makers.

In Table 2 we can find that the voted top 10 banks are roughly the same during the years 1994-95, 1996-97, and 1998 separately. Thus, we form three different top groups for those three time periods. The rest of the banks in each period are allocated to the non-top group.

After setting up the two groups, we count the quotes for each group during our sample period. Graph 1 shows that from October 1995 onwards, the total quotes appearing on the FXFX have more than doubled. The Olsen & Association which provides our data explains that it is due to the different data delivery method of

7

Reuters before and after October 1995. The data fed into the system since then have been significantly increased. However, the quotes from the top group are relatively stable at 25% of total quotes. As we show later, the information shares of both groups experience no fundamental change following the quote frequency jump8.

Graph 2 presents the intraday quotes distribution of both groups. As top group is composed of only European and U.S. banks, its quoting activity is heavily concentrated during London and New York trading hours. Non-top group’s quoting distribution reflects Asian trading banks’ presence during Tokyo trading hours. To avoid sparse trading, we investigate only trading hours between 8:00 to 16:00 GMT, when London and New York markets are active. Weekends and holidays are also excluded due to the same reason. There are total of 1,214 valid days during our sample period. Following general practice, we convert our data into 5-minute frequency.

3.2 Macro Announcement Data

We test 21 categories of US government macro announcements’ effect in our analysis9, as displayed in Table 3. Among the 29 categories of major US macro announcements, we exclude 8 categories of announcements either due to the high frequency of announcements (e.g. initial claim is announced weekly) or due to overlapping announcement days (e.g. industrial production and capacity utility are announced at the same hour and same day of each month). Except for GDP related announcements, which are quarterly, and Fed funds rate announcements, which are on a six-week base, all remaining items of news are announced monthly.

Though some announcements, e.g. the target federal funds rate announcements from FOMC are announced outside our daily GMT window, due to their importance, we still include them. In our sample period, we have total of 1035 items of valid announcements.

8

The increased data feeds into the Reuters system decrease the average quote duration from around 6 seconds to 3 seconds, which bring no big impact on the data we selected at a 5 minute frequency.

9

4. Empirical Results

4.1 Lead-lag Analysis

Lead-lag analysis provides a simple means to test the speed of the information embedded in the return of the two prices. For instance, many studies have investigated the lead and lag relationships between cash and futures markets10. The cross

correlations between two series p and q are given by,

) 0 ( . ) 0 ( ) ( ) ( , qq pp pq q p c c l c l r = , 4.1 where ...l=0,±1,±2, and ⎪ ⎪ ⎩ ⎪⎪ ⎨ ⎧ = − − = − − =

∑

∑

If the top group’s quotes are faster in incorporating trading information into the price, its return should lead non-top group’s return, i.e. top group’s return could predict the non-top group’s return.

Graph 3 shows the lead lag relationship between the returns of the two groups. The contemporaneous correlation between the two return series is understandably high at 0.27. When top group’s return is in the lag, the correlation is around 0.1 and statistically significant. However, when non-top group’s return is in the lag, there is no significant correlation, which suggests that the predictability runs only in one direction.

4.2 Monthly Information Share

We subsequently estimate the top group’s monthly information share during our 60 months of sample period. Since we only consider the London and New York trading hours, to avoid overnight price jumps generating excessive noise, we

10

eliminate overnight lag returns and use seemingly unrelated regression (SUR) to estimate our models. Throughout this paper, we only present the top group’s result, since the top and non-top groups’ information shares add up to 100%. The non-top group’s information share is 100% deducted by the top group’s information share.

In Graph 4 we present the information share of the top group. As it can be seen the top group’s information share is relatively stable. During the five years, the top group’s informational share as a percent of total information is 73% if PT model is used, and 71% if the IS model is used during the five years (see Table 5). Among the 60 months, only once does the top group’s information share drop below 50% according to the PT model, and twice if using IS model.

Yearly results (see in Table 5) suggest that top group starts with a low share in 1994, and then it experiences a rise in 1995 and then again a fall in 1996. For the rest two years its share jumps significantly and reaches around 80% in 1998. To explain this dynamic changing pattern of top group’s information advantage, we estimate the yearly volatility at daily frequency and find an interesting negative correlation between them (see Graph 5). When the market is relatively volatile during the year of 1994 and 1996, the top group’s information share is low and the furthest away from its five-year average. While in 1998 when the market is relatively calm, the top group also enjoys its highest information share. The only exception is in 1995, when although the market is relatively stable, the top group still underperforms compared to its yearly average results. The above negative relationship between market volatility and the Top group’s information advantage suggests that top banks perform better when the market is relatively predicable. Too much noise or shock would cause difficulty for top banks to extract information from their trading with customers and from markets.

To test the robustness of the grouping method, we also estimate the information share of the top 5 instead of top 10 banks. The top 5 banks in the Euromoney survey are unchanged during our five year sample period, though a few of them went through merger and acquisition. In Graph 6 we find that the information shares of the top 5 banks are less stable than top 10 banks’ results. The average monthly information

share drops to 62% (PT) and 60% (IS) of total market information. The number of months during which top 5 banks information drops below 50% has increased to 8 (PT) and 10 (PT). These results suggest even though these top 5 banks still take dominant share of the market information, this advantage is less obvious due to decreased group components and strengthened rivalry group.

4.3 Information Share during Macro Announcement Effects

To test whether macro announcements have any impact on the information share of the top banks, we import 21 U.S. news announcements and estimate the information share of the top banks during these announcements days. Though UK macro announcements may cause impact on the information share of the banks, we assume these effects to be insignificant11.

The results are reported in Table 6. Among the 21 announcements, GDP preliminary and Fed funds rate produce the largest information shares, over 80% according to PT model, for top 10 banks. This is an interesting finding compared to Evans’ (2005) result. In his paper, these two announcements are the only announcements that have a significant impact on order flow for just one day, while other 16 announcements having relatively longer effects. This suggests that these two announcements cause more concentrated and intensive reaction from market players, which forces the top banks to release their information advantage in only a few hours instead of distributing the advantage among several days like in the case of other announcements. Another interesting finding is that trade balance, as an important determinant in traditional exchange rate theory, contributes very little to the information advantage of the top banks. As aforementioned, this may be explained by the easy predictability of trade balance figure by the market players. Therefore top banks are less likely to know more than their rivalries.

We are also interested in the news effect by allocating them into different categories. Following Andersen et al (2003), we put news into 8 different groups such

11

As reported in Andersen et al. (2003), the impact of most non-US announcements is insignificant on the level of major exchange rates.

as real activity, forward looking and net export etc. In Table 7 we report the result. As we can see, FOMC, i.e. Fed funds rate, cause the highest information share for top banks. Given the importance of the interest rate in foreign exchange market, this could be a reasonable result. Government expenditure and price are the next two most important news categories. As an important fiscal policy tool, government expenditure’s impact on foreign exchange market is significant. And price level has long run impact on the exchange rate according PPP theory.

6. Conclusion

This paper is one of the first to directly tackle the information asymmetry issue in foreign exchange market. Traditional exchange rate theory assumes that the agents in a given market are homogeneous and therefore price formation process is only determined by public information. However, when we cast our eyes in the real foreign exchange market, the assumption of market players’ homogeneity is unsound and misleading. To correctly assess and depict the picture of market participants’ heterogeneity in information sense may help us to solve or explain the exchange rate determination puzzle. There are many factors that may contribute to foreign exchange dealers’ heterogeneity. One is the business type of the dealers. For instance, Fan and Lyons (2001) and Evans (2005) find that non-financial corporations, mutual funds, and hedge funds present different trading behaviour due to separate trading motivations.

We investigate this information asymmetry by categorizing the banks into top banks and non-top banks. By relating the top banks to dealing with much larger order flow, especially from customer, we assume that they could garner more information on the macro economy. There are two features that distinguish our paper from others. Firstly, using two prevalent common component models, we pin down the exact proportion of total market information that belongs to top banks. Secondly, we also find that public news could attribute greatly to top banks’ information set. This information advantage could come from top banks information collected through

customer trading or inter-dealer trading, and it may also come from after-announcement news interpretation, or non-common-knowledge, as Evans (2002) claims.

Further research could focus on finding more contributing factors to the information asymmetry in the foreign exchange market. For instance, does geographic location contribute to traders’ private information? In equity markets, geography has generated long and persistent interest in academics. One general hypothesis raised, as in Malloy (2005), is that geographically proximate analysts possess an information advantage over other analysts. In FX markets, it is a natural extension of the hypothesis to test whether geographic proximity helps traders to interpret and predict public news and information. For instance, due to the importance of US economy, does US traders’ geographic advantage contribute to their information of correctly pricing US dollar?

Reference

Andersen, Bollerslev, Diebold, and Vega, 2003, “Micro Effects of Macro Announcements: Real-Time Price Discovery in Foreign exchange.” American Economic Review, 93, 38-62.

Baillie, R T., Booth, G. G., Tse Y., and Zabotina T., 2002, ‘Price Discovery and Common Factor Models’, Journal of Financial Markets, 5, 309-321.

BIS, 2005, Triennial Central Bank Survey – Foreign exchange and derivatives market activity in 2004.

Bjønnes, G.H. and D. Rime, 2001, Customer Trading and Information in Foreign Exchange Markets. Working paper # 4. Stockholm Institute for Financial Research.

Carlson, J., and M. Lo, 2004, One minute in the life of the DM/$: Public news in an electronic market, Journal of International Money and Finance, forthcoming.

Chan, K., 1992, A further analysis of the lead-lag relationship between the cash market and stock index futures market, Review of Financial Studies, (5)(1), pp. 123-152.

Cheung, Y. –W., and Chinn, M. D., 2001, “Currency Traders and Exchange Rate Dynamics: A Survey of the US Market,” Journal of International Money and Finance, 20, 439-471.

Danielsson, J., and R. Payne (2002): “Real trading patterns and prices in the spot foreign exchange markets,” Journal of International Money and Finance, 21, 203–222. de Jong, F. & Nijman, T. 1997, ‘High frequency analysis of lead-lag relationships between financial markets’, Journal of Empirical Finance, vol. 4, pp. 259–277.

Euromoney, May 1995, Treasurers put their views on banks, 65-76. _________, May 1997, Taken aback by a leap forward, 61-76. _________, May 1995, Life after execution, 89-104.

Evans, M., D., 2002, FX trading and exchange rate dynamics, The Journal of Finance, VOL. LVII, NO. 6, p2405-2447.

Evans, M., and R. Lyons, 2003, How is Macro News Transmitted to Exchange Rates? NBER Working Paper 9433, January.

____________________ , 2005, Exchange rate fundamentals and order flow, typescript, U.C. Berkeley, September.

_____________________, 2005, Do currency markets absorb news quickly? Journal of International Money and Finance, 24, 197-217.

Fan, M., and R. Lyons, 2001, Customer-dealer trading in the foreign exchange. market, typescript, UC Berkeley.

Glosten, L., and P. Milgrom, 1985, Bid, ask, and transaction prices in a specialist market with heterogeneously informed traders, Journal of Financial Economics 13, 71–100.

Gonzalo, J. and C.W.J. Granger, 1995, Estimation of common long-memory components in cointegrated systems, Journal of Business & Economic Statistics 13, 27-35.

Goodhart, C.; T. Ito; and R. Payne, 1996, One day in June 1993: A Study of the Working of Reuters 2000-2. In The Microstructure of Foreign Exchange markets, J. Frankel, G. Galli, and A. Giovanni, eds. Chicago, IL: Univ. of Chicago Press.

Goodhart, C., and O’Hara, M., 1997, High Frequency Data in Financial Markets: Issues and Applications. Journal of Empirical Finance, 4, 73-114.

Hasbrouck, J., 1995, One Security, Many Markets: Determining the contributions to Price Discovery. The Journal of Finance, Vol. L, No. 4, 1175-1199.

Kyle, albert, 1985, Continuous auction and insider trading. Econometrica, 53, 13-32.

Love, R., and R. Payne, 2004, Macroeconomic news, order flows, and exchange rates, typescript, London School of Economics.

Lyons, R., 1995, Tests of Microstructural Hypotheses in the Foreign Exchange Market, Journal of Financial Economics, 39, 321-351.

Malloy, Christopher J., 2005, The geography of equity analysis, The Journal of Finance, VOL. LX., NO. 2, 719-755.

Melvin, M., and X. Yin., 2000, “Public Information Arrival, Exchange Rate Volatility and Quote Frequency.” Economic Journal, 110, 644-661.

Payne, R., 2003, Informed trade in spot foreign exchange markets: an empirical investigation, Journal of International Economics, 61: 307-329.

Peiers, B., 1997, “Informed Traders, Intervention and Price Leadership: A Deeper View of the Microstructure of the Foreign Exchange Market.” Journal of Finance, 52, 1589-1614.

Phylaktis, K. and Chen., L., 2006, Price discovery in indicative and transaction FX prices system -- A comparison of D2000-1 and EFX data, working paper, Cass Business School.

Rime, D., 2000, U.S. Exchange Rates and Currency Flows, unpublished working paper.

Sapp, S., 2002, “Price Leadership in the Spot Foreign Exchange Market.” Journal of Financial and Quantitative Analysis, Vol. 37, No. 3.

Yao, J., 1998, “Market Making in the Interbank Foreign Exchange Market.” Salomon Center for the Study of Financial Institutions Working Paper s-98-3, New York, NY.

Table 1

Euromoney Survey’s Criteria of Top Banks

Corporations Institutions Banks Others Total

1. Relationship 100 10 48 7 165

2. Price alone 91 14 49 10 164

3. Quote speed 74 12 50 9 145

4. Credit rating 66 10 27 7 110

5. Liquidity 41 11 38 7 97

Source: Euromoney research, Euromoney May 1995. There were in total 16 criteria listed in the original table. The first column displays the most important 5 criteria judged by the total votes (given in the last column) from the customers of currency trading banks. The votes from each business type of customers are listed separately in the columns in the middle.

Table 2

Top Ten Banks in GBP/USD Currency Pair

99 98 97 96 95 94

1 1 HSBC 1 4 HSBC Midland 1 1= NatWest Markets

2 2 Citibank 2 1 Chase 2 4 HSBC Mkts/Midland

3 3 Chase Mahattan 3 3 BZW 3 3 Barclays

4 4 NatWest Global 4 2 NatWest 4 1= Citibank

5 5 Barclays Capital 5 5 Citibank 5 5 Chase Mahattan 6 6 Deutsche Bank 6 - Royal Bank of Canada 6 9 Chemical

7 7 Royal Bank of Canada 7 6 Standard Chartered 7 10 Bank of America 8 - Warburg Dillon Read 8 7 Bank of America 8 - Lloyds

9 9 Bank of America 9 - SBC Warburg 9 - Standard Chartered 10 12 ABN Amro 10 9 Deutsche Morgan Grenfell 10 6 Indosuez

Table 3 U.S. Announcements

Item Frequency EST #N

1 GDP Advance QTR 8:30 13

2 GDP Preliminary QTR 8:30 16

3 GDP Final QTR 8:30 13

4 Fed Funds Rate 6WK 14:20 38

5 Personal Income MTH 8:30 45 6 Factory Orders MTH 10:00 51 7 Confidence MTH 10:00 53 8 Leading Indicators MTH 8:30 55 9 Housing Start MTH 8:30 57 10 Durables MTH *8:30 57 11 Construction Spending MTH 10:00 57 12 Retail Sales MTH 8:30 58 13 Treasury Budget MTH 14:00 58 14 CPI MTH 8:30 58 15 Business Inventories MTH *10:00 58 16 Unemployment MTH 8:30 58 17 Non-farm Employment MTH 8:30 58 18 Consumer Credit MTH 15:30 58 19 Industrial Production MTH 9:15 58 20 NAPM MTH 10:00 58 21 Trade Balance MTH 8:30 58 Total 1035

* The announcement time were irregular or changed during our sample period.

The third column reports the scheduled announcements frequency, where QTR, MTH and 6WK stand for quarterly, monthly and 6 weeks respectively. EST is the U.S. Eastern Standard Time. Last column reports the total number of the corresponding announcements during our sample period

Table 4. Statistics of Top Group’s Monthly Information Share

Mean Std. D. Kurtosis Skewness Min Max Count

PT 73.1% 10.5% 0.24 0.09 45.7% 98.2% 60

IS 70.9% 8.9% 1.05 0.14 48.0% 94.6% 60

Table 5. Yearly Information Share of Top Group

1994 1995 1996 1997 1998

PT 67.5% 72.1% 68.4% 76.3% 81.0% IS 68.2% 69.8% 67.2% 72.3% 77.2%

Table 6. Top Group’s Information Share during U.S. Macro News

Item PT IS

1 GDP Preliminary 80.7% 76.8% 2 Fed Funds Rate 80.1% 78.6% 3 Consumer Confidence 79.3% 76.2% 4 Retail Sales 79.0% 75.8% 5 Treasury Budget 77.9% 75.8% 6 CPI 77.7% 75.9% 7 Personal Income 76.6% 73.7% 8 Business Inventories 76.0% 73.0% 9 Housing Start 75.6% 71.9% 10 Industrial Production 74.8% 72.8% 11 Unemployment 74.6% 72.2% 12 Non-farm Employment 74.6% 72.2% 13 Durables 74.0% 71.2% 14 Factory Orders 73.4% 72.7% 15 Consumer Credit 72.1% 69.5% 16 GDP Final 71.6% 71.1% 17 NAPM 71.0% 69.6% 18 Trade Balance 70.5% 69.3% 19 GDP Advanced 69.9% 68.5% 20 Leading Indicators 68.9% 67.2% 21 Construction Spending 68.8% 68.9%

The information shares are measured by only considering the days when news is announced. The announcements are reported in a descending order of the PT results.

Table 7. News Effect by Category

Category #N PT IS 1 FOMC 38 80.1% 78.6% 2 Government Purchase 58 77.9% 75.8% 3 Prices 58 77.7% 75.9% 4 Real Activity 277 75.4% 72.8% 5 GDP 42 74.1% 72.1% 6 Forward Looking 223 73.7% 71.2% 7 Investment 223 73.1% 71.4% 8 Net Exports 58 70.5% 69.3%

Graph 1. Monthly Quotes from Top Group and as Percentage of Total Quotes 0 20,000 40,000 60,000 80,000 100,000 120,000 Jan-94 Feb -94 Mar-9 4 Ap r-9 4 May -94 Jun-94 Jul -94 Aug -94 Se p -94 Oct -94 Nov -94 De c-94 Jan-95 Feb -95 Mar-9 5 Ap r-9 5 May -95 Jun-95 Jul -95 Aug -95 Se p -95 Oct -95 Nov -95 De c-95 Jan-96 Feb -96 Mar-9 6 Ap r-9 6 May -96 Jun-96 Jul -96 Aug -96 Se p -96 Oct -96 Nov -96 De c-96 Jan-97 Feb -97 Mar-9 7 Ap r-9 7 May -97 Jun-97 Jul -97 Aug -97 Se p -97 Oct -97 Nov -97 De c-97 Jan-98 Feb -98 Mar-9 8 Ap r-9 8 May -98 Jun-98 Jul -98 Aug -98 Se p -98 Oct -98 Nov -98 De c-98 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Tot Q Perc.

The column stands for the number of quotes from the top ten banks, as the top group, in each month. The line stands for the percent of the top group’s quotes compared to that of total quotes from all the banks in the same month.

Graph 2. Daily Quotes Distribution of Both Groups

0% 1% 2% 3% 4% 5% 6% 0: 00 1: 30 3: 00 4: 30 6: 00 7: 30 9: 00 10 :3 0 12 :0 0 13 :3 0 15 :0 0 16 :3 0 18 :0 0 19 :3 0 21 :0 0 22 :3 0 Top NonTop

Each half hour’s quotes are aggregated and presented as the percentage of total quotes of individual

Graph 3. Cross Correlations of the Top and Non-top Group’s Return -0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30 -5 -4 -3 -2 -1 0 1 2 3 4 5

Cross-Corr UpperBound LowerBound

The positive figure on the x-axis means that top group’s return is in the lag. Each lag stands for 5 minute. The upper and lower bounds of the cross correlogram are the approximate two standard error

bounds computed as ±2/ T , where T is the available number of observations.

Graph 4. Top Group’s Monthly Information Share

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Jan-94 Ap r-94

Jul-94 _{Oct-94 Jan-95 Ap}

r-95

Jul-95 _{Oct-95 Jan-96 Ap}

r-96

Jul-96 _{Oct-96 Jan-97 Ap}

r-97

Jul-97 _{Oct-97 Jan-98 Ap}

r-98

Jul-98 _Oct-98

Graph 5. Top Group’s Information Advantage and FX Market Volatility -8% -6% -4% -2% 0% 2% 4% 6% 8% 10% 94 95 96 97 98 0% 1% 2% 3% 4% 5% 6% PT IS YrVol.

The yearly information shares from the two models are all deducted by the five-year average. The

market yearly volatility is estimated from daily price change.

Graph 6. Top 5 Banks’ Monthly Information Share

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Jan-94 Ap r-94

Jul-94 _{Oct-94 Jan-95 Ap}r-95 Jul-95 _{Oct-95 Jan-96 Ap}r-96 Jul-96 _{Oct-96 Jan-97 Ap}r-97 Jul-97 _{Oct-97 Jan-98 Ap}r-98 Jul-98 _Oct-98