AN ANALYSIS OF MACROECONOMIC ANNOUNCEMENTS ACROSS LOCATIONS IN THE DM/DOLLAR MARKET* ANUSHA CHARI University of Michigan AUGUST 2002 *I am grateful to Michael Brennan, Avanidhar Subrahmanyam and seminar participants at the Univerisity of Chicago for providing insightful and valuable feedback on my ideas. Any errors and omissions are my sole responsibility. Correspondence to: Anusha Chari, University of Michigan Business School, 701 Tappan Street, MI 48104 U.S.A. Internet: achari@umich.edu

AN ANALYSIS OF MACROECONOMIC ANNOUNCEMENTS ACROSS LOCATIONS IN THE DM/DOLLAR MARKET This paper tests cross market linkages and geographic segmentation in the foreign exchange market in response to the arrival of macro-economic news. I characterize the behavior of DM/$ quotes (spreads, volatility, quote frequency) from banks physically located in Europe versus the United States in response to the arrival of news about the German or American economies. In particular, the focus is on the overlap period in the interbank spot market for foreign exchange located in Frankfurt and New York. The data show strong intradaily seasonality in activity patterns across regional segments of the spot market. Vector auto-regression estimates are then used to disentangle own and cross market linkages following the arrival of public information. The results support the hypothesis that geographic segmentation plays a role in currency markets despite trading an apparently identical asset suggesting differences between local and foreign trader perceptions about the domestic currency. KEYWORDS: exchange rates, microstructure, macroeconomic announcements, high frequency data, vector autoregression. 1 Introduction This paper examines the implications of geographic segmentation in the spot foreign exchange market. We investigate whether traders have imperfect and segmented knowledge of exchange rate fundamentals and competing trader perceptions of these fundamentals. We hypothesize that traders may be more capable of receiving and evaluating exchange rate signals which emanate from their country than from overseas. We argue that operating out of a particular country gives a trader an advantage in interpreting both the nuances in public signals generated in that country and the reactions of other traders in the country to the signal. This asymmetric information view of exchange rate determination parallels that found in the international home bias literature [Brennan and Cao (1998)] and

in recent work investigating the role of geography in domestic equity markets [Coval and Moskowitz (1999)]. To examine our hypothesis, we focus on country-specific macroeconomic announcements. Public announcements of macroeconomic fundamentals offer an attractive opportunity to investigate our hypothesis, as their timing is largely predictable, and they convey information which is highly related to exchange rate fundamentals. Recent evidence suggests foreign exchange markets are highly geographically segmented. In perfectly integrated markets, the arrival of new information which results in increased volatility in one financial center should immediately lead to increased volatility in other financial centers that are simultaneously open. Contrary to this, Hsieh and Kliedon (1996) find that the high volatility witnessed when the foreign exchange market opens in New York or closes in London is not related to the concurrent volatility in the other market. This is despite the fact that both sets of quotes appear on exactly the same trading screens at exactly the same time. This implies that although specific regional segments of the interbank market for foreign exchange obey the usual u-shaped patterns that have been rationalized by the clustering of informed trading1, other trading sessions do not display any trace of a u-shape at the corresponding point in time. They assume that these differences in regional volatility patterns cannot be due to new information reaching one market but not the other, within standard information frameworks. Since there is a high degree of electronic integration in the foreign exchange market, it cannot be plausible that markets that are "ostensibly closely linked are segmented in important ways not recognized in standard models." Instead, it must be the case that "some phenomenon other than the incorporation of private information is responsible for the behavior of quotes." An explanation that has received very little attention in the foreign exchange literature is that the regional differences in activity patterns observed in markets that are open at the same time could arise out of geographic segmentation. This implies that agents in different regional markets may have heterogeneous information sets. As a result, traders in different physical locations may have a different understanding or "feel" for the implications of news that reveals information about the fundamental value or future payoffs associated with a particular currency. Since the value of any given currency reflects the relative values of the fundamentals associated with two economies and in turn two monetary policies, the arrival of public information specific to any one economy may be interpreted differently by 1Admati and Pfliederer (1988) 2

domestic versus foreign traders. Ederington and Lee (1993) and Bollerslev and Anderson (1998) note that the arrival of public information induces abrupt price changes and that the average price move is typically attained within minutes. Yet, volatility and trading volume tend to remain elevated for several hours. If agents have identical information sets and interpret news similarly, the protracted response pattern is hard to explain and provides an argument in favor of models with heterogeneously informed agents. (see Kim and Verecchia (1991)) A potential source of information heterogeneity could stem from geographic segmentation. There are several reasons to expect systematic differences in the quote behavior and activity patterns of interbank traders located in Germany/Europe as opposed to the United States surrounding the release of macroeconomic news in the foreign exchange market. First, there may be differences in the timing of the arrival of information. Traders located in Frankfurt or London might have a geographic advantage in terms of receiving private information about German macroeconomic announcements before traders situated in New York and vice versa. This implies that there might be a sequential component to the receipt of public information that could include a local leakage of information. Second, traders may have an advantage in processing public signals generated in their own country. For example, as members of central bank committees jockey for support for their own monetary policy stances in the press, local traders may have superior ability to evaluate the consequences of such information for future domestic monetary policy and hence the domestic currency's value. Hence, although the signals themselves are public and simultaneously available to domestic and overseas traders, the "black box" which traders use to interpret public signals (i.e. an econometric model or gut intuition) may be more accurate in interpreting domestic signals than those overseas. Third, behavioral explanations could account for systematic differences in responses to public information. To the extent that traders located in Germany tend to trade on behalf of people in Germany while traders in the United States act on behalf of investors in the US, a geographic analysis may allow identification of any behavioral differences between German and US investors. For instance, Germans may respond differently to improved fundamentals for Mark, perhaps reacting with greater confidence than investors in the US. Similarly, news about third parties may have varying effects on the behavior of traders 3

located in different markets. News about an intervention by the Federal Reserve in the yen/dollar market may impact the strategy of German traders with respect to the dollar differently from US traders. In the same vein, news about German intervention in support of other EMS currencies could affect German and US traders differentially. Finally, volatility may change differently in response to German/US announcements in the two markets so that we witness spillover or lead-lag relationships in the volatility transmission depending on where the announcement is made. There could also be leadership or bandwagon effects following news announcements across markets, i.e., US traders may follow the strategy of German traders following a German announcement and vice versa. This paper examines the nature of return (volatility) clustering and spillovers across regional foreign exchange markets during trading hours in which two (or more) markets are simultaneously open. The analysis is conducted in an event study framework around the release of specific macroeconomic news announcements in both domestic and foreign markets. The paper proceeds as follows. Section 3.3 provides a description of the data sources and the construction of events around the release of U.S. and German macroeconomic announcements. Section 3.2 reviews the literature and Section 3.4 provides a preliminary data analysis of the activity patterns in the different regional segments of the DM dollar market. Section 3.5 presents the results from the event study analysis. Section 3.6 explores avenues for future research and concludes. 2 Related Literature There is a vast body of literature that tests the implications of geographic segmentation and cross border linkages in international equity markets. The transmission mechanism of stock returns and volatility has been the focus of numerous studies: Bennett and Kelleher (1988); von Furstenberg and Jeon (1989); Hamao, Musulis, and Ng (1990); King and Wadhwani (1990); Neumark, Tinsley, and Tosisni (1991); Becker, Finnerty, and Tucker (1992); and Dravid, Richardson, and Craig (1993) are some examples. Lin, Engle, and Ito (1994) summarize several empirical regularities reported in these studies: (i) the volatility of stock prices is time varying; (ii) when volatility is high, the price changes in major markets tend to become highly correlated; (iii) correlations in volatility and prices appear to be causal from the United States to other countries; and (iv) lagged spillovers of price changes and price 4

volatility are found between major markets. Lagged spillovers are defined as correlations between the foreign daytime return (volatility) and subsequent domestic daytime return (volatility), without any overlapping trading hours. Correlations in price changes can be associated with the dispersion of beliefs (see Shalen (1993) for a two-period noisy rational expectations model of a futures market). When new information arrives, different prior beliefs about the news create incentives to trade and lead to price changes. As traders observe the new price, they may revise their prior beliefs in response to new information, which leads to continued trading and future price changes. If it takes time for the market to resolve these heterogeneous beliefs when traders revise their prior beliefs in response to new information, this process of searching for the information price may lead to volatility clustering around the arrival of new information. Analyzing volatility correlations across markets also requires an examination of the speed of the market adjustment to new information. Lin, Engle and Ito (1994) devise tests for lagged returns and volatility spillovers to examine how promptly domestic stock prices react to overnight foreign news as the domestic market reopens using global and country specific shocks in return innovations. Engle et al. (1990,1992) document that news which is revealed when one foreign exchange market is open contributes to the return volatility when the next segment of the market opens. These volatility spillovers are dubbed 'meteor showers' and appear to be present for various time periods for the yen dollar exchange rate. Similar results were found for other currencies by Lin (1989). None of these studies, however, found any evidence that news in one market could predict the mean return in subsequent markets. Susmel and Engle (1994) presume that such effects are arbitraged away by the market. Information asymmetries around earnings announcements have been examined extensively in equity markets. Morse and Ushman (1983), Venkatesh and Chiang (1986) and Skinner (1991) use daily quoted spreads while Daley, Hughes and Rayburn (1991), Barclay and Dunbar (1991) and Seppi (1992) use block trades to conduct their analysis. Lee, Mucklow and Ready (1993) show that spreads increase dramatically in the half half hour containing the announcement, and remain wider than during non announcement periods for up to one day.2 They find that spreads widen and depths fall in anticipation of earnings 2This is consistent with Patel (1991) who also reports an increase in spreads following earnings announcements. 5

announcements and that these effects are more pronounced for announcements with larger subsequent price changes. Spreads are also wider following earnings announcements, but this effect dissipates rapidly after controlling for volume. Collectively, their results suggest that liquidity providers are sensitive to information asymmetry risk and use both spreads and depths to actively manage this risk. They suggest that although most extant models would predict an increase in information asymmetry before an earnings announcement, the predictions for the post announcement period are less clear. One hypothesis is that earnings news reduces the information advantage of the informed trader, so spreads should decrease during this time. Alternatively, Kim and Verrecchia (1991) outline a theoretical argument to suggest that information asymmetry will be higher after earnings announcements because the announcements are noisy signals and certain traders have a superior ability to process the earnings information. However, the post announcement liquidity effects should be interpreted with caution, because extremely heavy trading volumes characterize this period. In the Kim and Verrecchia specification, the asymmetric information risk arises from the public disclosure of the earnings and not the accompanying volume. Their model predicts a drop in post announcement liquidity that is independent of the general relationship between volume and liquidity. Lee, Mucklow and Ready (1993) show that after controlling for the volume increase, the drop in post announcement liquidity is significant for approximately half an hour following the release of the earnings information. This suggests that the information advantage from a superior ability to process earnings news, as formalized by Kim and Verrecchia may be a short-lived phenomenon. Tanner (1997) documents evidence using intraday data that unanticipated information about the trade deficit and consumer price index had an impact on the DM dollar exchange rate while there was no significant response to news about money supply, industrial production, the producer price index or unemployment. Ederington and Lee (1993) find that scheduled macroeconomic announcements are responsible for most of the observed timeof-day and day-of-the-week volatility patterns in foreign exchange futures markets. While the bulk of the price adjustment to a major announcement occurs within the first minute, volatility remains substantially higher than normal for roughly fifteen minutes and slightly elevated for several hours. Nonetheless, these subsequent price changes are basically independent of the first minute's return. 6

Consistent with these findings, Bollerslev and Anderson (1998) find that the largest returns in the DM dollar market appear to be linked to the release of public information, and, in particular certain macroeconomic announcements. They conclude that major announcements dominate the picture immediately following the release, but their explanatory power is less than that of the intraday patterns at high frequencies, and much less than that of standard volatility forecasts at the daily level. They present evidence to show that the most significant U.S. announcements, namely the employment report, gross domestic product, trade balance figures, and durable goods orders are all related to the real economy while the important German announcements, the Bundesbank meetings and M3 supply figures, are monetary. This may reflect differences in the perceived central bank reaction functions. Or it could be the case that during their sample period, monetary policy in the United States was relatively stable, while in Germany it was highly controversial. 3 Data Sources and Construction The main data set consists of tick by tick indicative quotes from the interbank market for foreign exchange. The data set contains 1.6 million quotes for the DM-dollar spot exchange rate from October 1, 1992 through September 30, 1993. Each quote consists of time stamped bid and ask prices along with an identification of the bank advertising the quote and its location. Returns, effective spreads and average quote frequencies are constructed for five minute intervals from the exchange rate indicative quotes that appear on the Reuters's FXFX network over the sample period desegregated by three locations: Frankfurt, London and New York. For each five minute interval we calculate the number of quotes that appeared on the screen from each market, and use the quote frequency numbers to construct a measure of the effective spread that would have been paid had each indicative quote translated into a transaction for each five minute interval. Since the foreign exchange market is a decentralized dealership market transaction prices and amounts traded are not recorded by a centralized exchange. We assume unit size for all the quotes transaction size while constructing the effective spreads. This is consistent with models of market-maker pricing under asymmetric information that do not incorporate depth information by assuming equally weighted 7

transactions of unit size.3 To examine intraday volatility, log returns were calculated over the entire trading day in five minute intervals. Note that these returns only measure price changes and do not represent returns in an investment sense in that no money is actually invested up front. Average returns are constructed by using absolute values of the first difference of the log of the mid price of the spot rate.4 Standard deviations of these log returns are calculated across 255 trading days (weekends and holidays were excluded from the sample). However, all returns from Friday 21:00 GMT through Sunday 21:00 GMT were excluded from the analysis. See Bollerslev and Domowitz for an explanation for the slow interbank quote activity that justifies this definition of the "weekend." Our total sample consists of T = 255 weekdays for a total of N = 73440 five-minute return, spread and quote observations (where n=1, 2,...N and t=l, 2,...T). Since intra-daily data exhibits strong seasonal patterns5 we calculated average returns, spreads and quote frequencies for each five minute interval, j, across 255 trading days. The nth return, spread and quote frequency within day t are St, and Qt,n, respectively. All N = 288 intervals during the twenty-four hour trading cycle are used. These averages (N = 288) were used to deseasonalize the average returns, spreads and quote frequencies for every five minute interval, i, j (N=288*255). The deseasonalized values allow us to capture a measure of excess returns, excess spreads and excess quote frequencies while analyzing the impact of macroeconomic news on trading patterns. We also constructed a sub sample of the data-set to study the impact of news announcements that appeared during the overlap period when all three markets were simultaneously open. An analysis of the quote activity reveals that the overlap period consists of three hours between 13:00 GMT and 16:00 GMT. 3.1 News Releases The data set also includes all of the news headlines that appeared on the Reuters money news-alert screens. During the sample period from October 1, 1992 through September 3See Lee, Mucklow and Ready for a discussion. Copeland and Galai(1983), Glosten and Milgrom (1985),and Easley and O'Hara (1992) provide examples of this modelling strategy. 4Mid prices are constructed using the average of the bid and ask prices for any given quote. 5See Bollerslev and Anderson (1998, 1997) and Bollerslev and Domowitz (1993) 8

30, 1993, the total number of headlines that appeared in the screen was 105,065. These headlines are time stamped to the nearest second and constitute the basis for our analysis of announcement effects. For the DM dollar spot rate, Bollerslev and Anderson (1998) list the employment report, gross domestic product, trade balance figures, and durable goods orders to be the most significant U.S. announcements while the most important German announcements are Bundesbank meetings and M3 supply figures. Ederington and Lee (1993) highlight the employment figures, PPI, gross national product, the trade deficit, durable goods and retail sales figures to have the greatest impact in their analysis of the deutsche mark futures market. We examine a set of announcements that consist of weekly, monthly and quarterly scheduled announcements as well as unexpected news about interventions by the Bundesbank and Federal Reserve. We also include unanticipated news about the ERM as well as stock market developments in the United States. We selected this list of announcements as they could signal a change in the demand for foreign exchange and traders believe that these are important variables that central banks consider while formulating changes to monetary policy. We exclude announcements that are made when overlapping segments of the market have stopped trading-we are left with 105 announcements from the US and 86 from Germany. The Reuters' FXNB page is used to separate out the announcements by country of origin and two databases are constructed for German and U.S. macroeconomic news. These two databases pin point time stamped news events which are then used to evaluate segmentation effects in trading patterns across locations in response to changes in the public information environment. For example, do banks in New York respond differently from banks in London or Frankfurt to a German rate cut? We use the announcement events to analyze whether volatility levels, spread patterns and quote frequencies evolve differently across regional segments of the market. A second stage of the empirical exercise consists of testing for cross-market linkages in the form of volatility spillovers, lead-lag relationships and possible bandwagon effects in quote behavior. 9

4 Preliminary Data Analysis Figures 3.1 documents the trading hours in Frankfurt, London and New York, respectively. This figure depict the quote frequency on the Reuters's screens in five-minute intervals dis-aggregated by market. Hsieh and Kliedon (1996) document that each location shows activity from about 7:00 A.M. (local time) that lasts till about 6:00 P.M. (local time). The data are consistent with Bollerslev and Domowitz (1993) who note that trading activity as measured by the number of quote arrivals in Frankfurt and London begin high and decline until New York opens, then increases until the close of the trading day in London. Activity in New York follows that of London and continues to increase after the London close as New York becomes the main trading center. Tables 3.1-3.3 present t-tests of differences in the mean number of quotes during different intervals in the overlap period. 4.1 Frankfurt, London and New York: Integrated Market We replicate the analysis in Hsieh and Kliedon (1996) to document the U-shaped patterns in trading activity in each of the three individual markets across the trading day. Figure 3.2 plots the average standard deviations of quote midprices for the half hour intervals and the results confirm the U-shaped patterns documented by Bollerslev and Domowitz (1993) and Hsieh and Kliedon (1996). The average variances are much higher at the opens and closes. Figure 3.3 confirm that the patterns in the bid-ask spread mirror that of the variances for all three markets. Figure 3.2 plots the standard deviations of the mid price changes in all three markets in GMT and shows that there is no correspondence between the patterns in the European versus US markets. This is particularly striking given that the markets are virtually instantaneously linked in terms of quote information. In addition, there appears to be little coherence in the volatility patterns with the open or close of one market on the other markets. Tables 3.4-3.6 presents evidence of t-tests of the significance of the difference in spreads across the three market pairs, in fifteen-minute intervals from noon GMT to 5:00 P.M. GMT. Following Hseih and Kliedon (1996), the test assumes that the sub samples are uncorrelated. This indicates that the t-statistic is downward biased if there is a positive 10

correlation across the samples which is a reasonable assumption if there are inter-linkages between quote patterns across the regional segments. The results confirm the pattern from figure Figure 3.3 and show that the indicated spreads show that spreads in Frankfurt are consistently higher than spreads in New York. The results also show that spreads in New York are significantly higher that spreads in London during the overlap period with the exception of the closing of the London market when the reverse is observed. Studies of the effects of international dual listings in equity markets using intraday data show that spreads do not decrease following a dual listing while the depth of quotes increases as predicted. Noronha, Sarin and Saudagaran (1996) examine the impact on the liquidity of NYSE/AMEX listed stocks when they were subsequently listed on the London or Tokyo Stock Exchanges. They find that the level of informed trading increases, which increases the cost to the specialist of providing liquidity, and explains why spreads do not decline despite increased competition. Consistent with an increase in informed trading, they also document an increase in trading activity. Werner and Kliedon (1996) conduct an intraday analysis of market integration by analyzing the patterns for U.K. and U.S. trading of British cross listed stocks. They document evidence to show that cross border competition for order flows tends to reduce already declining spreads in London during the overlap period. By contrast, New York specialists maintain high spreads during the overlap period and overall, the evidence indicates that the order flow for cross-listed securities is segmented. (note: make a point about how differences in spreads indicate that the order flow for FX is also segmented-although not directly related to segmentation conditonal on differences in interpreting macro announcements it is prima facie evidence for unconditional segmentation-while Hsieh and Kliedon talk about regional U shapes in activity patterns here we find evidence for Frankfurt consistently setting spreads above the other two markets, not just at opens and closes. We are also conducting this exercise to argue for using deseasonalized data to study the impact of announcements). Tables 3.7-3.9 test for the difference between quote mid-prices in all three markets. Although the indicated spreads data shows a distinct ordering in spread patterns, the pattern does not translate to the mid-price data. T-tests of the difference between mean mid-prices do not indicate a clear pattern although there exist intervals when there are statistically significant differences across the three markets. 11

Tables 3.10-3.12 examine the difference between the first difference of the log of the mid-prices across the three markets to illustrate differences in variance patterns across the three markets. The results once again confirm the observations from Figure 3.2. From noon to 3:00 PM (GMT) the variance in New York consistently exceeds that of its European counterparts. With the largest average variances in New York at the beginning of the trading day. However, first the variance in Frankfurt followed by the variance in London start edging upwards with the drawing of a close of trading the trading days in the two locales sequentially. The t-statistics are strongly significant at both the 1 and 5 percent levels for all the consecutive intervals between 3:00 PM (GMT) and 4:30 PM (GMT) for Frankfurt and between 4:00 PM (GMT) and 5:00 PM (GMT) for London. These results corroborate the evidence that changes in the volatility in one regional segment are not reflected simultaneously in other segments that are open concurrently. The cross market variance results from Figure 3.2 and the cross market spread results from Figure 3.3 and Tables 3.4-3.6 suggest that there are strong intra daily seasonal patterns that are distinct across regional segments of the spot foreign exchange market. This constitutes the rationale for using deseasonalized data in our estimations of the responses of different regional segments to the arrival of macroeconomic news. 5 Results 5.1 Announcement Effects In this section we present evidence from regression analyses designed to capture the impact of individual announcements on the volatility, spreads and market activity on different regional segments on the spot foreign exchange market. The analysis is conducted for US announcements and German announcements separately. In particular, we construct a series of dummy variables Dk,t where Dk,t = 1 if announcement k is made on day t and Dk,t = 0 otherwise. The dependent variable in our regression is the absolute value of the difference between the actual quote frequency, bid ask spread value or return for the five minute interval j on day t and the mean quote frequency, bid ask spread or return for interval j over all 260 trading days in our sample period. Thus the dependent variables are deseasonalized (for time of day effects) five minute quote frequencies, spreads, first difference of the log of the spot midprice. The independent variables are dummies for the 12

announcements and lagged values of the dependent variable to correct for persistence. For the US there are six categories of announcements: CPI, durable goods orders, employment numbers, GDP, retail sales and the merchandise trade deficit and FED interventions. The German announcements are M3 figures, Bundesbank meetings, and interest rate changes and interventions by the Bundesbank. For example, following Ederington and Lee our sample format for the return regressions is: abs(Rj,t - Rj) = ao + K=lakDk,t + ej,t (1) Further, note that if log returns are normally distributed with constant mean and time varying variance, E Rjt - Rjl=(2/7r)05aj,t where aj, t is the standard deviation of returns in interval j on day t.6 This means that (r-/2)05ao=1.2533ao provides and estimate of the standard deviation of returns on non announcement days. Furthermore, since we are considering absolute values of deseasonalized returns, regardless of whether or not an announcement provides good or bad news about the economy to the market, the estimated coeffiecient ak should have a positive value if the announcement has an impact on the market. As a consequence, the figure 1.2533(ao + ak) provides an estimate of the standard deviation of returns when announcement k occurs. This also implies that we should expect the coefficient ak to be approximately zero if it has little news value for the market. Note also that the set of dummy variables varies across intervals. While an interval earlier in the day only contains dummy variables for announcements that have taken place up until that particular interval, intervals later in the day contain dummy variables for all announcements prior to that interval. This is done to capture the impact of earlier announcements on the persistence of volatility throughout the day. 5.2 Pooled Announcement Effects In this section we present evidence from regression estimates for pooled announcements across regional markets. The regression format was designed to estimate the impact of news originating from one regional segment on its own market and across markets. Again, the analysis was conducted for market activity as proxied by quote frequency, spreads and log returns. For example, in order to estimate the impact of German news on market 6See Ederignton and Lee (1993) for more details. 13

activity in New York, we estimated the following regression: j+it aO + al * *DGE + a2Qjt * (1-DGt ) + a3Qj,t * Dt + a4jt *l — GE (2) where: * QNY is the quote frequency in New York in period j + 1. * QjtR is the quote frequency in Frankfurt in period j. * D1t is a dummy variable for German news announcements. * QNY is the quote frequency in Frankfurt in period j. The coefficients, al and a2 are designed to measure cross market effects. al measures the impact of the quote frequency in Frankfurt lagged by one period and interacted with a dummy variable DjGtE, that takes the value of 1 when a German announcement takes place and 0 otherwise. a2 measures cross market linkages in market activity between Frankfurt and New York during trading interval when no announcement takes place. a3 and a4 capture own market effects on the quote frequency in New York. While a3 measures the impact of the lagged quote frequency in New York interacted with the German announcement dummy, a4 quantifies the significance of the lagged quote frequency in New York when there is no announcement. A similar exercise was conducted for the impact of German news on the quote frequency in Frankfurt. The regressions were also conducted to measure the impact of US news announcements on market activity in both New York and Frankfurt, respectively. Results from these regressions are presented in Tables 3.13 and 3.14. We also conducted two additional sets of regressions to estimate the own and cross market effects of German and US news on both bid ask spreads as well as volatility levels to study the impact of these announcements on market uncertainty as well as to assess any regional differences in these responses. The results are presented in Tables 3.15 to 3.18. The reaction in New York to US news (105 announcements) suggest that the quote frequency, spreads and volatility increase and the coefficient estimates are statistically significant. The reaction in Frankfurt to US news on the other hand shows that while the quote frequency and bid-ask spreads-decrease (negative and significant), the volatility increases 14

but the effect is not statistically significant. The estimates were corrected for autocrorrelation by using lagged quote frequencies, spreads and volatility since without the correction the Durbin Watson statistic was considerably below the benchmark value of 2.0. One lag sufficed in taking care of this problem. The errors were estimated using Newey West heteroskedasticity consistent covariance matrices. A potential rationale for narrowing German spreads could involve the market becoming more competitive following news releases (i.e. German spreads lowering following US announcements to attract US business and US spreads widening following German announcements to extract rents from German traders). Following German news in Frankfurt (68 announcements), the quote frequency falls (negative and significant) while spreads and volatility increase (positive and significant). The reaction in New York from Tables 3.13 to 3.18 suggest that market activity as measured by quote frequency, spreads and volatility all rise significantly in response to German news. 5.3 Vector Autoregression estimates Next we turn to regression estimates using a VAR framework. The vector autoregression (VAR) is commonly used for forecasting systems of interrelated time series and for analyzing the dynamic impact of random disturbances (here the arrival of macroeconomic news) on the system of variables. The approach sidesteps the need for structural modeling by modeling every endogenous variable in the system as a function of the lagged values of all of the endogenous variables in the system. Since only lagged values of the endogenous variables appear on the right-hand side of each equation, there is no issue of simultaneity, and OLS is the appropriate estimation technique. Note that the assumption that the disturbances are not serially correlated is not restrictive because any serial correlation could be absorbed by adding more lagged dependent variables.7 The mathematical form of a VAR is: Yt = alyt-1 +.......+ apyt-p + bxt + et (3) where yt is a k vector of endogenous variables, Xt is a d vector of exogenous variables, al,...,ap and b are matrices of coefficients to be estimated, and ct is a vector of innovations that may be contemporaneously correlated with each other but are uncorrelated with their own lagged values and uncorrelated with all of the right-hand side variables. 7See Hamilton (1994) for further details 15

We assume that quote frequencies in Frankfurt and New York are jointly determined by a two variable VAR along with a dummy for news and a constant as exogenous variables. The system of equations we estimated uses two lags of the dependent variables and a lagged dummy variable for German or US news. We arrived at the two lags specification after constructing cross corrollelograms to study the pattern of autocorrelations in the two regional segments. The VAR specification is as follows: NY NY NY FR FR US Q+2,t - ao + alQj+l,t + a2Qj,t + a3Qj+l,t + a4Qjt + a5Djt (4) QFR ao+a NY NY FR FR US Qj+2,t - ao + a2QQj+ + a3Q+t + a4QR + a (5) where: * QNY2 t QNY t, QNtY are quote frequencies in New York for intervals j + 2, j + 1 and j, respectively. * QF t QFjl t, QjFR are quote frequencies in Frankfurt for intervals j + 2, j + 1 and j, and * D1t is a dummy variable fore US news announcements. The exercise for repeated using a dummy for German news announcements. In addition VARs were estimated to study the impact of these two categories of announcements on spreads and volatility as well. Results for these estimates are presented in Tables 3.19 to 3.22. The results for the quote regressions show that while lagged values of own market activity have a positive impact on quote activity, lagged values of cross market activity have an opposite impact. This is evidenced by the positive and significant coefficients for QNY and QNY and the negative, significant and much lower magnitudes for the coefficients for QFR and QFR. The results also show that while US news leads to a significant increase in quote activity in New York the result is reversed for the Frankfurt market. Surprisingly, the trend remains the same following German news announcements. These results suggest that traders in New York tend to become more active, while traders in Frankfurt shut down and perhaps adopt a wait and see strategy following the release of macroeconomic news. The results for the VAR estimates for quote revisions in the two markets bear out these results and once again provide evidence that the New York market responds by increasing 16

activity following the release of macroeconomic news while the Frankfurt market shows depressed activity. Interestingly, while quote revisions in New York are significantly impacted by own market quote revisions, the results lose significance for the coefficients on the cross market quote revisions from Frankfurt. However, the cross market impact of quote revisions from New York to Frankfurt continue to be significant. These results seem to indicate evidence of price leadership by the US following US announcements but there is little if no evidence of the converse result following German news. The VAR estimates for the impact of lagged own and cross market volatility following news announcements suggest that volatility in any one regional segment is affected by persitence in its own market volatility as well as spillovers from other regional markets (albeit with coefficients of lower magnitude) following the release of news. A somewhat puzzling result is evidenced by the statistical insignificance of the dummy for US news on the Frankfurt market. A potential explanation could be that as evidence from the quote frequency VARs suggest that the German market shows reduced activity following the release of US news, volatility is reduced as well. However the same result does not hold for the impact on volatility in Frankfurt following the release of German news. Estimates from VARs for bid ask spreads show that while lagged own market spread effects tend to be positive and significant, cross market lagged spread effects are significantly negative. Again, the magnitudes for cross market effects are lower. The coefficient for the impact of US news on bid-ask spreads follows the same trend as that of the volatility regression estimates with traders in New York widening spreads and traders in Frankfurt lowering them. An explanation for the lowered spreads by Frankfurt traders could be that since spreads in Frankfurt tend to be statistically significantly greater than spreads in New York during the overlap period (except for the Frankfurt close), traders in Frankfurt lower spreads in order to become more competitive following the release of US news. Interestingly, German news leads to a widening of spreads by both US and German traders. 6 Future Research Since the overlap period between the trading hours in London and Frankfurt is longer (8:30 GMT to 16:00 GMT), future research could study the impact of German macroeconomic news that appeared when the New York market was closed. This would allow us to analyze 17

the impact of news that was specific to Europe on the two markets and since this was a particularly turbulent period for the European Monetary System it would also allow us to study the inter-market linkages within European financial markets. Preliminary evidence indicates the presence of intra-daily seasonality in the price leadership pattern with certain segments leading the market at different points of time in the trading day. This evidence could be used to compare how price leadership patterns change around event windows surounding the release of macroeconomic news. According to the efficient market view asset prices incorporate rational assessments of the fundamental values underlying the assets and reflect future payoffs. This implies that the arrival and processing of new information must result in changes in asset prices. In turn, since financial markets display a high degree of integration, standard information models (Admati and Pfliederer (1988), Subrahmanyam (1991) suggest that if the arrival of new information results in increased volatility in one financial center, then the high volatility should be observed in other financial centers that are simultaneously open. Future research could describe the price leadership dynamics more thoroughly as well as examine differences in the efficiency and speed of adjustment in different regional segments of the spot FX market. Attention could be focused on disentangling the price formation dynamics with precision to determine whether a greater percentage of price adjustments take place in the regional segment of the FX market where the announcement was made. Comparisons could also be made with the reaction of London market in order to check for robustness. Furthermore, Granger causality tests could be conducted to describe the direction of spillovers and price leadership. 18

Table 1: Test of Difference in Quotes, Frankfurt and New York Frankfurt New York Time Quotes StdcFR Quotes StdcNY T-stat 12:00 1580 2.7409092 132 0.95914699 7.124014803 12:15 1582 2.8653483 182 1.2392879 7.151240857 12:30 1508 2.8999742 255 1.2673986 6.925729134 12:45 1440 2.84109 271 1.3931724 6.681729228 13:00 1364 2.7205381 322 1.3565184 6.516356191 13:15 1392 2.7316443 331 1.4381597 6.633717721 13:30 1474 2.9199975 414 1.5161322 6.58308708 13:45 1416 2.8870027 475 1.5701398 6.115190994 14:00 1381 2.7263206 499 1.6865908 6.067519609 14:15 1402 2.6231473 537 1.7837203 6.219024866 14:30 1351 2.7143168 610 1.8515628 5.306069399 14:45 1356 2.7574313 642 1.8966097 5.087626796 15:00 1336 2.30406 772 2.0742211 4.677895459 15:15 733 2.6590984 887 2.3807226 -1.076042127 15:30 630 2.3806383 968 2.517182 -2.375903799 15:45 564 2.4338207 1002 2.4676884 -3.046823985 16:00 581 2.6659351 1058 2.6235887 -3.117794133 16:15 156 1.1947469 1097 2.6366205 -5.036425669 16:30 157 1.3837318 1104 2.9082623 -4.595813772 16:45 141 1.2020902 1082 2.9874577 -4.339732221 19

Table 2: Test of Difference in Quotes, London and Frankfurt London Frankfurt Time Quotes StdcLN Quotes StdcFR T-stat 12:00 1636 3.2807823 1575 2.862367 0.412891113 12:15 1630 3.3735798 1582 2.8653483 0.319653659 12:30 1722 3.27808 1508 2.8999742 1.437693052 12:45 1786 3.7589767 1440 2.84109 2.132329228 13:00 1856 3.5144423 1364 2.7205381 3.189460786 13:15 1727 3.5365403 1392 2.7316443 2.160812894 13:30 1709 3.5821897 1474 2.9199975 1.484272914 13:45 1830 3.665684 1416 2.8870027 2.576545795 14:00 1742 3.3973989 1381 2.7263206 2.390829356 14:15 1809 3.4193412 1402 2.6231473 2.731283498 14:30 1800 3.5472722 1352 2.5716114 2.924350133 14:45 1945 3.411195 1290 2.4926014 4.402670805 15:00 1971 3.6802994 1336 2.30406 4.137113586 15:15 1930 3.6276221 733 2.6590984 6.647724879 15:30 1853 3.7289391 630 2.3806383 6.494651746 15:45 1510 4.1688909 564 2.4338207 4.419552939 16:00 1301 4.1389969 581 2.6659351 3.371996277 16:15 1183 4.3010747 156 1.1947469 3.393598639 16:30 899 4.0164937 157 1.3837318 2.639982311 16:45 559 2.6647187 141 1.2020902 2.183836189 20

Table 3: Test of Difference in Quotes, London and New York London New York Time Quotes StdCLN Quotes StdcNY T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 1636 1630 1722 1786 1856 1727 1709 1830 1742 1809 1864 1838 1971 1930 1853 1510 1301 536 526 456 3.2807823 3.3735798 3.27808 3.7589767 3.5144423 3.5365403 3.5821897 3.665684 3.3973989 3.4193412 3.3500062 3.373113 3.6802994 3.6276221 3.7289391 4.1688909 4.1389969 2.8844881 2.7357333 2.4707459 109 182 255 271 322 331 414 475 499 537 610 642 772 887 968 1002 1058 1077 1037 1047 0.85206521 1.2392879 1.2673986 1.3931724 1.3565184 1.4381597 1.5161322 1.5701398 1.6865908 1.7837203 1.8515628 1.8966097 2.0742211 2.3807226 2.517182 2.4676884 2.6235887 2.912273 2.9442769 3.0000926 5.976657999 6.297389895 7.176699093 6.571925964 7.447694253 6.788519355 6.606236717 7.00134727 6.976720473 7.205540217 7.463054534 7.155661506 6.89545895 6.17714942 5.180794771 2.74924518 1.322291125 -3.176630872 -3.009870417 -3.430966427 21

Table 4: Test of Difference in Spreads, Frankfurt and New York Frankfurt New York Time Spreads NFR Spreads NNY T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 9.61835443 9.656131479 9.523872679 9.58125 9.573313783 9.510775862 9.500678426 9.668785311 9.656770456 9.737517832 9.797187269 9.752949853 9.782185629 9.814461119 9.853968254 9.838652482 9.903614458 9.634615385 9.713375796 10.18439716 1580 1582 1508 1440 1364 1392 1474 1416 1381 1402 1351 1356 1336 733 630 564 581 156 157 141 8.439393939 8.230769231 8.31372549 8.616236162 8.468944099 8.531722054 8.620772947 8.557894737 8.663326653 8.836126629 8.586885246 8.816199377 8.797927461 8.851183766 8.857438017 8.831337325 8.962192817 9.029170465 8.904891304 9.012014787 132 182 255 271 322 331 414 475 499 537 610 642 772 887 968 1002 1058 1097 1104 1082 3.531587365 3.277621461 3.49613736 2.659745129 2.94974375 2.641137815 2.262603643 3.357618994 2.853279449 2.813408688 3.609469666 3.149154376 3.720587511 2.900502616 2.89087264 0.329053451 2.739403977 1.294877449 1.4333544 2.390881805 22

Table 5: Test of Difference in Spreads, London and Frankfurt London Frankfurt Time Spreads NLN Spreads NFR T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 7.031784841 1636 6.979141104 1630 7.045876887 1722 6.996080627 1786 6.826508621 1856 6.996525767 1727 6.939730837 1709 6.984153005 1830 7.19804822 1742 7.239911553 1809 7.187222222 1800 7.318251928 1945 7.34906139 1971 7.534196891 1930 7.740960604 1853 8.070198675 1510 8.219830899 1301 8.267962806 1183 8.147942158 899 9.43470483 559 9.652063492 1575 9.656131479 1582 9.523872679 1508 9.58125 1440 9.573313783 1364 9.510775862 1392 9.500678426 1474 9.668785311 1416 9.656770456 1381 9.737517832 1402 9.774408284 1352 9.848837209 1290 9.782185629 1336 9.814461119 733 9.853968254 630 9.838652482 564 9.903614458 581 9.634615385 156 9.713375796 157 10.18439716 141 -3.08454528 -3.19603079 -2.865671571 -2.761318506 -3.066108759 -2.720045134 -2.778762608 -2.933102828 -2.619861457 -2.700558621 -2.888153779 -2.60898366 -2.619049977 -1.9514529 -1.745230479 -0.624650572 -1.442171657 -0.848746208 -0.912187678 -0.946924574 23

Table 6: Test of Difference in Spreads, London and New York London New York Time Spreads NLN Spreads NNY T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 7.031784841 1636 6.979141104 1630 7.045876887 1722 6.996080627 1786 6.826508621 1856 6.996525767 1727 6.939730837 1709 6.984153005 1830 7.19804822 1742 7.239911553 1809 7.186158798 1864 7.344940152 1838 7.34906139 1971 7.534196891 1930 7.740960604 1853 8.070198675 1510 8.219830899 1301 9.546641791 536 9.275665399 526 9.429824561 456 8.335443038 109 8.230769231 182 8.31372549 255 8.616236162 271 8.468944099 322 8.531722054 331 8.620772947 414 8.557894737 475 8.663326653 499 8.836126629 537 8.586885246 610 8.816199377 642 8.797927461 772 8.851183766 887 8.857438017 968 8.831337325 1002 8.962192817 1058 9.038068709 1077 9.00192864 1037 8.903533906 1047 -0.653666077 -0.76826638 -0.834814952 -1.027227629 -1.167022262 -1.051884039 -1.213701567 -1.203772529 -1.120774447 -1.259321726 -1.106522514 -1.223939065 -1.283700638 -1.184145359 -1.046568162 -0.745933982 -0.756465517 1.092756081 0.500722154 1.004140447 24

Table 7: Test of Difference in Mid-prices, Frankfurt and New York Frankfurt New York Time Mid-prices NFR Mid-prices NNY T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 1.648452455 1580 1.643865377 1582 1.658273005 1508 1.652633387 1440 1.65106344 1364 1.65886385 1392 1.659905563 1474 1.642232276 1416 1.645226022 1381 1.638442177 1402 1.633498338 1351 1.638808993 1356 1.63431209 1336 1.615229969 733 1.613223019 630 1.617949891 564 1.604702376 581 1.627205736 156 1.60638908 157 1.579365178 141 1.779025121 132 1.798970012 182 1.78921062 255 1.747134249 271 1.76742559 322 1.752870899 331 1.732684911 414 1.752857358 475 1.743652336 499 1.73120063 537 1.754746427 610 1.731311699 642 1.728243229 772 1.727663363 887 1.727268952 968 1.726138776 1002 1.713879732 1058 1.707161939 1097 1.718874853 1104 1.702221846 1082 -0.90222484 -1.15558076 -1.045188585 -0.763975462 -0.973922966 -0.791553068 -0.643932877 -1.002303566 -0.896322081 -0.85539253 -1.141717471 -0.877066213 -0.919835216 -1.054073818 -1.053493041 -0.985692644 -1.00676508 -0.568750846 -0.804057486 -0.859987758 25

Table 8: Test of Difference in Mid-prices, London and Frankfurt London Frankfurt Time Mid-prices NLN Mid-prices NFR T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 1.618066259 1636 1.619126135 1630 1.620604355 1722 1.616849552 1786 1.619160614 1856 1.622590909 1727 1.622951083 1709 1.623793497 1830 1.624799024 1742 1.622008181 1809 1.619780667 1800 1.624787969 1945 1.625841958 1971 1.623238497 1930 1.619969941 1853 1.615400397 1510 1.611521829 1301 1.612090533 1183 1.612159177 899 1.61119034 559 1.644936664 1575 1.643865377 1582 1.658273005 1508 1.652633387 1440 1.65106344 1364 1.65886385 1392 1.659905563 1474 1.642232276 1416 1.645226022 1381 1.638442177 1402 1.634628668 1352 1.627568159 1290 1.63431209 1336 1.615229969 733 1.613223019 630 1.617949891 564 1.604702376 581 1.627205736 156 1.60638908 157 1.579365178 141 -0.311214637 -0.281292924 -0.43919134 -0.400138053 -0.369117327 -0.424589446 -0.437576179 -0.222879787 -0.241664545 -0.200787622 -0.171204959 -0.035050258 -0.107014168 0.092028337 0.076742935 -0.029759697 0.076429435 -0.160038637 0.070912031 0.31043975 26

Table 9: Test of Difference in Mid-prices, London and New York London New York Time Mid-prices NLN Mid-prices NNY T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 1.618066259 1.619126135 1.620604355 1.616849552 1.619160614 1.622590909 1.622951083 1.623793497 1.624799024 1.622008181 1.623201931 1.624189064 1.625841958 1.623238497 1.619969941 1.615400397 1.611521829 1.61801278 1.613770817 1.612227412 1636 1630 1722 1786 1856 1727 1709 1830 1742 1809 1864 1838 1971 1930 1853 1510 1301 536 526 456 1.784239737 1.798970012 1.78921062 1.747134249 1.76742559 1.752870899 1.732684911 1.752857358 1.743652336 1.73120063 1.754746427 1.731311699 1.728243229 1.727663363 1.727268952 1.726138776 1.713879732 1.704701242 1.702651457 1.714453918 109 182 255 271 322 331 414 475 499 537 610 642 772 887 968 1002 1058 1077 1037 1047 -1.193851765 -1.398636403 -1.475742854 -1.080421016 -1.348774194 -1.227078582 -1.09241822 -1.34532056 -1.223850312 -1.18230347 -1.404396126 -1.219314951 -1.22047486 -1.219447531 -1.2772424 -1.351278693 -1.194678396 -0.885160485 -0.881547448 -1.02066984 27

Table 10: Test of Difference in Volatility, Frankfurt and New York Frankfurt New York Time Volatility NFR Volatility NNY T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 3.097029077 3.027581783 4.100195185 3.628998609 3.588002874 3.602442334 3.937146893 3.955829109 4.027104137 4.156065089 4.336046512 5.203148426 4.566848568 4.666666667 7.027777778 6.356687898 6.514184397 6.724770642 7.011363636 6.705607477 1580 1582 1508 1440 1364 1392 1474 1416 1381 1402 1351 1356 1336 733 630 564 581 156 157 141 6.406593407 6.064171123 6.773662551 5.150337838 4.587613293 4.605072464 4.849473684 4.728456914 4.554003724 4.724960254 4.724137931 4.587719298 4.422773393 4.327995868 4.089041096 4.0625 4.16728281 4.040389972 4.049662488 3.863896848 132 182 255 271 322 331 414 475 499 537 610 642 772 887 968 1002 1058 1097 1104 1082 3.309564329 3.03658934 2.673467366 1.521339229 0.999610419 1.00263013 0.912326792 0.772627804 0.526899587 0.568895166 0.388091419 -0.615429128 -0.144075174 -0.338670799 -2.938736682 -2.294187898 -2.346901588 -2.68438067 -2.961701148 -2.841710628 28

Table 11: Test of Difference in Volatility, London and Frankfurt London Frankfurt Time Volatility NLN Volatility NFR T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 2.966564417 1636 2.903640929 1630 4.126805315 1722 3.368131868 1786 3.279675738 1856 3.313341135 1727 3.75 1709 3.70608496 1830 3.774184632 1742 3.916388889 1809 4.053213368 1800 4.409736308 1945 4.021502591 1971 4.597949271 1930 4.877455566 1853 4.541156841 1510 5.661001789 1301 6.003731343 1183 5.378326996 899 5.589912281 559 3.097029077 1580 3.027581783 1582 4.100195185 1508 3.628998609 1440 3.588002874 1364 3.602442334 1392 3.937146893 1474 3.955829109 1416 4.027104137 1381 4.156065089 1402 4.336046512 1351 5.203148426 1356 4.566848568 1336 4.666666667 733 7.027777778 630 6.356687898 564 6.514184397 581 6.724770642 156 7.011363636 157 6.705607477 141 0.13046466 0.123940854 -0.026610129 0.260866741 0.308327135 0.289101199 0.187146893 0.24974415 0.252919505 0.2396762 0.282833144 0.793412117 0.545345977 0.068717395 2.150322212 1.815531057 0.853182608 0.721039299 1.63303664 1.115695196 29

Table 12: Test of Difference in Volatility, London and New York London New York Time Volatility NLN Volatility NNY T-stat 12:00 12:15 12:30 12:45 13:00 13:15 13:30 13:45 14:00 14:15 14:30 14:45 15:00 15:15 15:30 15:45 16:00 16:15 16:30 16:45 2.966564417 2.903640929 4.126805315 3.368131868 3.279675738 3.313341135 3.75 3.70608496 3.774184632 3.916388889 4.053213368 4.409736308 4.021502591 4.597949271 4.877455566 4.541156841 5.661001789 6.003731343 5.378326996 5.589912281 1636 1630 1722 1786 1856 1727 1709 1830 1742 1809 1800 1945 1971 1930 1853 1510 1301 1183 899 559 6.406593407 6.064171123 6.773662551 5.150337838 4.587613293 4.605072464 4.849473684 4.728456914 4.554003724 4.724960254 4.724137931 4.587719298 4.422773393 4.327995868 4.089041096 4.0625 4.16728281 4.040389972 4.049662488 3.863896848 132 182 255 271 322 331 414 475 499 537 610 642 772 887 968 1002 1058 1097 1104 1082 3.440028989 3.160530194 2.646857237 1.78220597 1.307937555 1.291731329 1.099473684 1.022371954 0.779819092 0.808571365 0.670924563 0.17798299 0.401270803 -0.269953404 -0.78841447 -0.478656841 -1.493718979 -1.963341371 -1.328664508 -1.726015433 30

Figure 3.1: Quote Frequency; Frankfurt, London & New York 7000 -6000 -5000 -04000 -a 0 E 3000 z 2000- 1000 __ _ _ _ _ _ 0 Time (GMT)

Figure 3.2: Midprice Volatility; Frankfurt, New York 45 40 35 30 i 25 20 E 20 15 10 o0 &o ~ ~ ~ ( N& ~.. I Deiations. Standard Deviations

Figure 3.3: Bid-ask Spreads; Frankfurt, London & New York 40 35 30 25 Lu Q., 20 0) 15 10 0 -1- a G 1i _____ i; II i ir 1 New TorKso I i; i; i;. N %Zp t..s.N %Z. N, t. d N qp.t..s.N %. s.0 N %p t..s.N % N qp @'& do & 6 3 & 6 6 & S 9- 9- b N N rV I e i N N - N N @ 69- N P + ax q Time (GMT)