High Frequency Trading in the US Treasury Market. Evidence around Macroeconomic News Announcements

Transcription

1 High Frequency Trading in the US Treasury Market Evidence around Macroeconomic News Announcements This version: June 2012

2 High Frequency Trading in the US Treasury Market Evidence around Macroeconomic News Announcements Abstract This paper investigates the role and effect of high frequency (HF) trading in the US Treasury market around major macroeconomic news announcements. Based on tick-by-tick activities from a major interdealer trading platform, we identify HF trades and orders in the US Treasury market over the period of January 2004 to June Our results show that both HF trades and orders increase substantially following news announcements. In particular, more HF limit orders are placed following announcement with unexpected high surprises. Moreover, while HF limit orders help to reduce the bid-ask spread, there is no evidence that they help to improve the order book depth at best quotes. Finally, we provide clear evidence that HF limit orders are less informative than manual orders placed following news announcements. JEL Classification: F3, G12, G14, G15. Keywords: High Frequency Trading; Market Liquidity; Price Discovery; US Treasury Market.

3 I. Introduction Automated trading or high frequency (HF) trading, carried out by computer programs, has become prevalent in financial markets during the past decade. 1 As reported in financial press, in recent years trading records are routinely broken and millions of data messages are regularly sent per second to various trading venues 2. This anecdotal evidence is coupled with the hard fact that, in only ten years, trading latency in several markets has decreased by two orders of magnitude (Moallemi and Saglam, 2011) and trading, as well as quoting activity, takes place within a fraction of a second (see, inter alia, Clark, 2011; Hasbrouck, 2012; Scholtus and Van Djik, 2012 and the references therein). Although HF trading is one of the most significant market structure developments in recent years (SEC, 2010), the role and effect of these activities on market liquidity and price efficiency is still a relatively unexplored topic in the finance literature. A stream of very recent studies has begun to investigate the impact of HF trading in equity markets (see, inter alia, Hendershott et al., 2010; Hasbrouck and Saar, 2011; Brogaard, 2011a; 2011b; 2012; Hendershott and Riordan, 2011; Egginton et al., 2012; Bohmer et al., 2012 and the references therein). However, very little research has been carried out to understand the impact of HF activities in other important security markets. 3 Another important but relatively unexplored question is how HF trades and quotes react to information arrival. One unique feature of HF trade and quotes is its speed in reaction to information arrival. With its quick reaction speed and capacity to handle a large amount of information, it remains an open question whether HF activities improve or deplete liquidity of the market and whether they help or hinder price 1 From a semantic viewpoint, HF trading can be seen a subset of market activities carried out by computers labelled Algorithmic Trading, or AT (Chlistalla, 2011). Since this article focuses on the trading activities that are carried out by machines at a very high speed, we will refer to these activities as HF trades and orders. 2 See Speed and market complexity hamper regulation Financial Times, 7 th October, The only exception is represented by Chaboud et al. (2009) who investigate the role of algorithmic trading in the foreign exchange market.

4 discovery around information arrival. In this paper, we explore HF activities around macroeconomic news announcement in the US Treasury market. The US Treasury market is one of the world s largest financial markets with a daily trading volume that rivals the US equity market. According to Aite Group (2008), the introduction of Electronic Communication Networks (ECNs) in the early 2000s has encouraged the establishment and development of HF trading and quoting activities in the US Treasury secondary markets. Very recently ICAP, one of the major brokers in this market, quantifies that in 2009 more than 50 percent of their bids and offers are black-box-oriented and 45 percent of the overall trading in US Treasuries over their ECN, BrokerTec, is due to algorithm-based trading (Kite, 2010). Given the increasingly large role of HF trading and quoting, we fill an important gap in this literature and study the role of HF activities on market quality and the price discovery process in the US Treasury secondary market. More importantly, we focus on the impact of HF trades and orders around the times of major information arrival, macroeconomic news release. As Treasury securities have fixed cash flows, macroeconomic news announcements contain public information that is most relevant to valuation of Treasury securities. The advantage of using macroeconomic news release is that forecasts of upcoming announcement are readily available and thus we can explicitly measure public information shock. This allows us to directly investigate whether and how algorithm reacts to the extent of public information shock in terms of announcement surprise. Furthermore, around the time of announcement, market conditions, such as bid-ask spread and market depth also, experience drastic changes. The change in information environment around announcements provides a clean and unique setting to explore how HF activities relate to various market variables. The dataset used in our study is obtained from BrokerTec, a major ECN for the trading of on-the-run US Treasury notes and bonds, and contains tick-by-tick information regarding all

5 transactions and quotes over the period of January 2004 June It contains trading activities and order book information for 2-, 5- and 10-year notes. We propose a procedure to identify HF transactions and orders that are placed or modified at a speed beyond human capacity. With identified HF trades and orders, we address the following issues. First, given the relevance of macroeconomic variables in driving the price of Treasury securities (see, inter alia, Fleming and Remolona 1997; 1999; Balduzzi et al., 2001; Andersen et al., 2003; 2008 and Hoerdahl et al., 2012; Menkveld et al., 2012 and the references therein), we investigate how HF activities take place around macroeconomic news announcements. Following the arrival of new information, market participants have to assess quickly the impact of different shocks and, consequently, place new or modify existing orders. The recent development of automated news service such as Reuters NewsScope allows computer programs to react and take position in the market within a millisecond of news announcement. One of the questions studied in this paper is how HF activities react to and evolve with the arrival of public information. Second, we examine whether HF activities improve or deplete market liquidity. As HF trades and quotes respond to information shocks quickly and potentially with (cumulated) large volumes, it is important to understand whether in the US Treasury market HF activities are likely to supply or absorb liquidity in different market conditions. Third, as a logical extension of the analysis of HF activities around macroeconomic news announcements, we investigate the informativeness of HF trades and orders relative to manual orders. In addition, we are interested in whether HF trading and quoting facilitate or hinder the Treasury securities price discovery process. Our results show that both HF trades and orders increase substantially immediately following macroeconomic news announcements. In particular, HF orders are placed and cancelled much more frequently during the 15-minute post-announcement period. Moreover, there are more HF

6 orders placed following announcements with bigger surprises. There are also significantly more HF trades of the 2-year note associated with bigger announcement surprises. The findings suggest that HF trades and orders react to both the arrival and magnitude of shock in public information. Second, we provide mixed evidence on whether HF trading activities improve or deplete market liquidity. We show that, as expected, market orders or trades tend to deplete market liquidity, whereas market orders tend to supply market liquidity. Specifically, trades tend to widen bid-ask spread, especially during the 15-minute pre-announcement period, and reduce market depth at best quotes. The later effect is highly significant during both the 15-minute preannouncement and the 15-minute post-announcement periods. On the other hand, limit order submissions tend to narrow bid-ask spread and increase market depth at best quotes. These effects are highly significant during both the 15-minute pre- announcement and the 15-minute post-announcement periods. Further examining the effect of HF trades on market liquidities, we find no evidence that a higher percentage of HF trades leads to further higher bid-ask spread. This is true during both the pre- and post-announcement periods. Similarly, we find no evidence that a higher percentage of HF trades further depletes market depth at best quotes. Again, this holds during both the pre- and post-announcement periods. The findings on the effect of HF orders on market liquidity are interesting. Our results show that while a higher percentage HF limit orders tends to further help reducing the bid-ask spread, a higher percentage HF limit orders does not further improve market depth at best quotes. Instead, our results show that a higher percentage HF limit orders has a negative effect on market depth at best quotes. This seems to suggest that HF limit orders tend to be conservative and a higher percentage of HF limit orders in the order book may subsequently cause other dealers cancel orders at best quotes. The results are similar to Hendershott et al. (2010) in the context of the US equity market, which also record that the impact of a higher message traffic on the depth of the order book is negative.

7 Finally, we employ the test proposed in Kaniel and Liu (2006) to examine whether HF trades and orders are more informative than manual orders. The test provides mixed findings across different maturities during the pre-announcement period. While there is evidence that HF market orders of the 10-year note are more informative, HF limit orders of 2-year notes are more informative. The evidence is rather consistent across different maturities during the post-announcement period. There is clear evidence that HF limit orders are less informative than manual orders, but HF market orders are more informative than manual orders although the evidence is slightly weaker. In addition, we further examine the effect of HF trades and orders on subsequent absolute mid-quotes serial correlation, a price efficiency measure proposed by Boehmer and Kelley (2010) and Boehmer et al. (2012). The results provide no consistent evidence that HF trades and orders either help to facilitate or hinder the price discovery process of Treasury securities. Our research is closely related to recent studies that have explored the role and impact of HF trades and orders to financial markets from a theoretical and empirical point of view. From a theoretical perspective, Biais, Homber and Weill (2010) show that HF trading improves the traders ability to respond to new information and thus improves informational efficiency in the market. Along the same lines, Biais, Foucault and Moinas (2010) show that HF activities reduce search cost but they induce adverse selection in equilibrium since machines process information faster than slow traders. Similar to Biais, Foucault and Moinas (2010), Javanovic and Menkveld (2011) find HF activities induce adverse selection in terms of their speed of reaction to market events. From an empirical point of view, various studies explore the rise and development of HF activity in equity and foreign exchange markets. Two issues are mainly investigated: i) the impact of HF activities on market liquidity and volatility and ii) the role of HF activities on price discovery. The first group of studies records that, on average, HF activities improve market liquidity (Hendershott

8 et al., 2011; Brogaard, 2011; 2012a; Hasbrouck and Saar, 2011; Chaboud et al., 2009). However, Boehmer et al. (2012) also suggest that HF activities reduce liquidity in small stocks and when market making is difficult. With regards to the issue of HF activities and price discovery, Chaboud et al. (2011) find evidence that, in foreign exchange markets, transactions carried out by machines are less informative than the ones generated by human traders. On the contrary, Hendershott and Riordan (2010) find that computer quoting activity improves significantly the price discovery process by contributing more to innovations to efficient prices in NASDAQ. Similarly, Brogaard (2010) finds that HF activity has greater price impact and leads price discovery. Our study is also related to the vast literature on macroeconomic news announcements and price movements in Treasury markets. Fleming and Remolona (1997) and Andersen et al. (2003; 2007) find that the largest price changes are mostly associated with macroeconomic news announcements in the Treasury spot and futures markets. Balduzzi et al. (2001), Fleming and Remolona (1999), Green (2004) and Hoerdahl et al. (2012) point out that the price discovery process of bond prices mainly occurs around major macroeconomic news announcements and the same announcements are responsible for changes in risk premia across different maturities. Menkveld et al. (2012) record similar findings for 30-year Treasury bond futures. Pasquariello and Vega (2007) find that private information manifests on announcement days with larger belief dispersion. However, the existing literature uses data sample before the introduction of electronic limit order book, during which automatic trading does not exist. With the U.S. Treasury market transition to an electronic order driven market during the late 90s (detailed in Fleming and Mizrach (2009) and Mizach and Neely (2009), this paper is thus among the first in studying how HF activities react to the arrival of public information and its impact in the Treasury market. The reminder of the article is as follows: Section 2 introduces the dataset employed in the empirical analysis and describes in detail the procedure used to compute the variables associated

9 with HF intensity. Section 3 discusses the empirical results and a final section concludes. II. Data and HF Trades and Orders II.1 Data The data on US Treasury securities used in this article is obtained from BrokerTec, an interdealer ECN in the secondary wholesale US Treasury securities market. Prior to 1999, the majority of interdealer trading of US Treasuries occurred through interdealer brokers. After 1999, two major ECNs emerged: espeed and BrokerTec. Since then, the trading of on-the-run Treasuries has migrated to the electronic platforms (Mizrach and Neely, 2009; Fleming and Mizrach, 2009). 4 In our empirical investigation we use data on 2-, 5- and 10- year Treasury notes from the BrokerTec limit order book over the period January 5 th, 2004 to June 29 th, The data set contains the tick-by-tick observations of transactions, order submissions, and order cancellations. It includes the time stamp of transactions and quotes, the quantity entered and/or deleted, the side of the market and, in the case of a transaction, an aggressor indicator. The data on macroeconomic news announcements and the survey of market participants are obtained from Bloomberg. Balduzzi et al. (2001) show that professional forecasts based on surveys are neither biased nor stale. In our empirical investigation we use a set of 17 key macroeconomic news announcements at 8:30 a.m. ET from Pasquariello and Vega (2007). In line with much empirical literature on announcement news and asset prices (see Andersen et al. 2003; 4 According to Barclay et al. (2006), the electronic market shares for the 2-, 5- and 10-year bond are, respectively, 75.2%, 83.5% and 84.5% during the period of January 2001 to November By the end of 2004, the majority of secondary interdealer trading occurred through ECNs with over 95% of the trading of active issues. BrokerTec is more active in the trading of 2-, 3-, 5- and 10-year Treasuries, while espeed has more active trading for the 30-year maturity.

10 2007 and the references therein), we construct standardized news surprises to measure public information shocks as follows: where is the actual value of announcement k on day t, is the median forecast of the announcement k on day t and is the time-series standard deviation of. The full list of macroeconomic news announcements, together with some descriptive statistics computed over the sample period, is reported in Table 1. The summary statistics of the trading activities around news announcements are reported in Table 2. Panel A reports the results during the pre-announcement period (8:15 am to 8:30 am), and Panel B reports the results during the post-announcement period (8:30 am to 8:45 am). The results show that during the pre-announcement period, the 2-year note is the most liquid market followed by the 5-year note and then the 10-year note. In fact, the 2-year note records the largest trading volume, the deepest depth of the order book (at both the best quotes and overall) and the smallest spread. The other two maturity tenors are not very different from each other in that they record comparable levels of trading volumes, depths and spreads. The patterns highlighted in Panel A are also confirmed in Panel B, where the variables of interest are recorded during the 15-minute interval following major macroeconomic news announcements. We construct abnormal values for those variables for use in the subsequent empirical analysis in order to take into account for potential time-series effects that might have occurred over the sample period. More specifically, we define abnormal spread, P D, as P D 1M(i) P D 1M( i) NON =1 j=1 ] (1) 5 [ 1 30 P D 1M(i j) where P D 1M( i) denotes the average bid-ask spread within i-th minute interval on

11 announcement day t and NON P D 1M(i j) denotes the average bid-ask spread in the past t-k no-event day during (i-j)th intervals, where j=0,1, 30 and k=1,,5. Similarly, abnormal depth measures are computed as BST DPTH 1M(i) BST 1 DPTH 1M(i) 30 BST NON 5 [ 1 DPTH 5 =1 30 j=1 1M(i j) ], (2) ALL DPTH 1M(i) ALL 1 DPTH 1M(i) 30 ALL NON 5 [ 1 DPTH 5 =1 30 j=1 1M(i j) ], (3) where DPTH BST ALL 1M(i), DPTH 1M(i) denote abnormal depth recorded at the best quote and the BST ALL overall depth, respectively and DPTH 1M(i), DPTH 1M(i) are the average of the visible depth at the best quote and the overall depth recorded within i-th minute interval on announcement day t. Finally, abnormal volatility is defined as, VOL 1M(i) VOL 1M(i) 1 30 NON 5 [ 1 VOLA 5 =1 30 j=1 1M(i j) ], (4) where VOL 1M(i) denotes volatility over i-th minute interval on announcement day t. For each order, we calculate the most recent change in mid-quote right before the order was submitted. VOLA is calculated as the average of absolute change in mid-quote corresponding to all orders within i-th minute interval on announcement day t. The behavior of these trading variables on announcement and non-announcement days are plotted in Figure 1. Bid-ask spreads are higher on announcement days vis-à-vis non announcement days. On announcement days, the change in spreads occurs at the announcement time and the value of the spreads reverts to pre-announcement values almost instantaneously. The depth of the order book (both at the best quote and overall), decrease substantially on

12 announcement days around announcement times. Different from bid-ask spreads, the decline is more pronounced for the 2-year note than the 5- and 10-years notes. In fact, around announcement times the average depth of the limit order book for the 2-year note decreases between 50 percent (overall) and 80 percent (best quote). However, in both cases the recovery occurs within 10 minutes from the announcement time. With regards to trading volume, as in much empirical literature on trading and macroeconomic news announcements (Fleming and Remolona, 1997; 1999), on announcement days trading volume is higher and the difference peaks around announcement times for all maturity tenors. However, the difference is larger for the 2-year note than for the other two bond maturities. II. Identification of HF Trades and Orders Our dataset does not contain information about whether a trade or order is placed through computer program or manually. However, the dataset records reference numbers that provide information on the submission timing of an order and its subsequent alteration, cancellation or execution. Using this useful piece of information, we identify high frequency (HF) activity by looking at the reaction time of order and transactions to changes in market conditions. We select those trades/orders that are placed to the market at a speed deemed beyond human reaction. More specifically we label trades and orders as originating from computers if the following conditions apply: High Frequency Market order (HFMO) if a market buy (sell) order is placed to hit the best ask (bid) quote within a second of the changes of the best quotes High Frequency Limit order

13 o HFLO1 if a limit order at the best quote is modified within one second of changes in best quotes on the same side of the market o HFLO2 if a limit order at the best quote is modified within one second of changes in best quote on the opposite side of the market o HFLO3 if a limit buy (sell) order placed at the second best quote is modified within one second of the changes of the buy (sell) best quotes. o HFLO4 if a limit order is cancelled or modified within one second of its placement regardless of market condition changes. We note that the procedure described above is designed to infer HF activities on the basis of the speed at which trades are executed and orders are submitted to or withdrawn from the market. However, some caveats are in order. First, manual orders can be mistakenly identified as HF orders if manual orders are placed earlier but arrive exactly one second before market conditions change. As a consequence, some manual trades and orders may be misidentified as HF trades and orders. Second, while tick-by-tick data is available in our dataset, we are cautious about using ultra high frequency data because of the concerns of market microstructure effects. In fact, due to discrete tick size, market microstructure noise may be aggravated as the sampling frequency increases. Therefore, we aggregate our data at the 1-minute frequency, consistent with existing empirical studies (see, inter alia, Fleming and Remolona, 1999; Balduzzi et al., 2001 and the references therein). Having constructed orders and trades that are likely to be generated by computers, we define a measure of abnormal HF activities around macroeconomic news announcements. Similar to

14 liquidity variables, the abnormal values are computed as the differences between actual HF trades and orders during the time interval and normal HF trades and orders, where the latter are defined as the average one-minute HF trades and orders over the past 5 days with no major news announcements or economic events. More formally, we define abnormal HF trades in the i-th minute interval on day t as TRADE HF 1M(i) TRADE 1 HF 1M(i) 30 TRADE NON 5 [ 1 HF 5 =1 30 j=1 1M(i j) ], (5) where HF TRADE 1M(i) is the number HF trades within the i-th minute interval on announcement day t, TRADE HF NON 1M(i j) denotes the HF trades identified in the past t-k no-event day during (i-j)th intervals, TRADE NON where j=0,1, 30 and k=1,,5. The term 1 5 [ 1 HF 5 =1 30 j=1 1M(i j) ] denotes the average of past HF trades recorded over the past five no-event days during the 30-minute interval prior to the i-th minute interval. Similarly, we define abnormal HF orders in the i-th minute interval on announcement day t as 30 ORDER HF 1M(i) ORDER 1 HF 1M(i) 30 ORDER NON 5 [ 1 HF 5 =1 30 j=1 1M(i j) ], (6) where HF ORDER 1M(i) denotes the number of HF orders within the i-th minute interval on day t. Equations (5) and (6) define measures of HF intensity that take into account potential time trends and intraday seasonality in HF trades and orders and the normal trading activities of days with no major news events. In essence, abnormal HF activity describes how HF activity, computed over a 1-minute interval, deviates from its average of the previous 5 days without prescheduled announcements and without significant events during the previous 30-minute intervals.

15 A summary of identified high frequency trades and orders is reported in Table 3. Panel A reports the average number of HF trades (inferred from market orders) over a 1-minute interval over the full sample period and across different macroeconomic news announcements. For all maturity tenors, average HF trades range between 13 (38) and 35 (87) per minute before (after) news announcements. The HF intensity seems more pronounced during the 15 minutes after the announcement with a value that is 2 to 3 times larger than that prior to news announcements. Panel B shows the average number HF orders which are consistent with the filter discussed earlier in this Section. The values are disaggregated by the type of filter (HFLO1 to HFLO4) and then aggregated (all HF orders). Across different types of potentially HF orders, limit orders that are cancelled or modified within one second of their placement regardless of market condition changes (HFLO4) exhibit the largest number. However, regardless of the filter applied and across different maturity tenors, even the figures reported in Panel B confirm that HF quoting activity is larger after macroeconomic news announcements with multipliers that range between 3 and 4 times the numbers of HF orders before the same announcements. If we take into account potential time trends and seasonality in computing the figures in Panels A and B and compute abnormal HF orders and trades as in Equations (1) and (2), the difference between pre and post announcement period, reported in Panel C, is even more striking. The results are corroborated by the patterns plotted in Figure 2. In fact, across all maturities, HF orders and trades are substantially larger at and after the announcement time than the period preceding the announcements. The shift is larger for the 5- and 10-year notes which record increments of 500 percent for HF orders and trades at the announcement time in comparison to pre-announcement periods. The 2-year note, although smaller in magnitude, records increments of percent at the announcement time. In all cases the levels of HF orders and trades revert to their pre-announcement levels within 15 minutes after the announcement time but taking into

16 account of normal market conditions, the abnormal HF orders and trades remain higher than that of pre-announcement period during the 15 minute interval after announcement III. Empirical Analysis In this section, we address the following issues around public information arrival: i) the use or occurrence of HF activity under different market conditions, ii) the effect of HF trades and orders on subsequent market liquidity, and iii) the informativeness of HF trades and orders as well as the effect of HF trades and orders on market efficiency. III.1. HF Trading Activities and Market Conditions The first issue we investigate relates to the relation between market conditions and HF intensity during the pre- and post-announcement period. Put differently, under which circumstances do HF trades and orders occur more often? We address this issue by regressing the abnormal levels of HF trades and orders against various variables related to market conditions. We separate the analysis for the pre- and post-announcement period and run the following panel regression for both abnormal HF trades and orders. In the former case, the equations we estimate are as follows: TRADE HF 1M(i) ORDER HF 1M(i) α + β 1 VOL 1M(i) α + β 1 VOL 1M(i) BST + β 2 DPTH 1M(i) BST + β 2 DPTH 1M(i) ALL + β 3 DPTH 1M(i) ALL + β 3 DPTH 1M(i) + β 4 SPRD 1M(i) + β 4 SPRD 1M(i) + ε 1M(i) + ε 1M(i) where the relevant variables are defined as discussed in Section 2. The specification for post-announcement period is similar to Equations (7), but we also incorporate absolute news (7) surprises, denoted by to control for the effect of public information shock on HF orders

17 and trades: 5 TRADE HF 1M(i) α + β 1 VOL 1M(i) BST + β 2 DPTH 1M(i) ALL + β 3 DPTH 1M(i) + β 4 SPRD 1M(i) + γ + ε 1m(i) ORDER HF 1M(i) α + β 1 VOL 1M(i) BST + β 2 DPTH 1M(i) ALL + β 3 DPTH 1M(i) + β 4 SPRD 1M(i) + γ + ε 1m(i) (8) The results of the regressions are reported in Table 4. Panel A reports the results for the pre-announcement period and Panel B reports the regression results for the post-announcement period. The results show that first, HF trades are more likely to occur when market volatility (VOLA) is high during the pre-announcement period. The relation is rather weak and mixed during the post-announcement period. Interestingly, during both pre- and post-announcement periods, HF limit orders more likely occur when market volatility is low. The negative relation between market volatility and HF limit orders alleviates the concern that our identification procedure more likely identifies limit orders submitted in a volatile market as HF orders. Second, while both HF trades and limit orders are positively related to bid-ask spread during the pre-announcement period, the relations tend to be negative during the post-announcement period. This suggests that dealers submit more market orders and limit orders when the bid-ask spread is wide during the pre-announcement period, whereas they tend to submit more market orders and limit orders when the bid-ask spread is narrow during the post-announcement period. That is, more HF trades and orders are placed during the pre-announcement period even under less favorable market liquidity conditions. Third, almost the opposite patterns are observed for the relation between HF trades and orders and overall depth of the limit order book. That is, while 5 If there are multiple announcements occurring at 8:30 on day t, we use the maximum value of absolute standardized surprise.

18 both HF trades and limit orders are positively related to overall depth of the limit order book during the pre-announcement period, the relations are negative during the post-announcement period. This suggests that dealers submit more market orders and limit orders when the overall limit order book is deep during the pre-announcement period, whereas they tend to submit more market orders and limit orders when the overall limit order book is shallow during the post-announcement period. Fourth, both HF trades and orders are negatively related to the depth at the best quotes during the pre- and post-announcement periods. The only exception is HF orders of the 10-year note during the post-announcement period. This suggests that when there is more depth at the best quotes, HF trades and orders are more likely used. Finally, the results in Panel B show an overall positive relation between HF trades and orders with standardized announcement surprises. The positive relation is highly significant for HF orders. This suggests that more HF limit orders are placed following announcement with unexpected high surprises. III.2. The Effect of HF Trades and Limit Orders on Subsequent Market Liquidity While the results in the above subsection may provide some interesting findings on the use of HF trades and orders under different market conditions, a potential drawback of the analysis is that it does not take into account of the interactive relations between HF activities and market conditions. In this section, we examine the potential impact of HF activities on subsequent market liquidity. Specifically, we focus on the effect of HF trading activities on bid-ask spread and the depth standing at the best quotes. We first examine how the number of orders submitted and transactions affect spread and depth at the best quotes, P D 1M(i+1) BST DPTH 1M(i+1) α + φ 0 O D α + φ 0 O D 1M(i) + γ 0 T D 1M(i) + β P D 1M(i) BST 1M(i) + γ 0 T D 1M(i) + βdpth 1M(i) + ε 1m(i) + ε 1m(i)

19 (9) where O D denotes the number of limit orders submitted with the i-th minute interval, T D is the number of market orders submitted within the i-th minute interval and the dependent variables P D and DPTH BST are as defined in Section 2. We expect that φ 0 < 0 and γ 0 > 0 for P D since market orders erode depth and widen spread and φ 0 > 0 and γ 0 < 0 for DPTH BST since limit order submission deepens depth of order book and narrows spread. Next, we examine the additional impact of HF orders and trades on subsequent market liquidity. Formally, we estimate the following regressions: P D 1M(i+1) BST DPTH 1M(i+1) α + (φ 0 + φ HF P ORDER )O D + β P D 1M(i) + ε 1m(i) α + (φ 0 + φ HF P ORDER )O D BST + βdpth 1M(i) + ε 1m(i) 1M(i) + (γ 0 + γ HF P TRADE )T D 1M(i) 1M(i) + (γ 0 + γ HF P TRADE )T D 1M(i) (10) where P ORDER (P TRADE ) indicates the proportion of HF orders (trades) out of the total number of orders submitted (trades executed). If HF trades and orders improve liquidity, we expect that φ hf < 0 and γ hf < 0 for P D, and φ hf > 0 and γ hf > 0 for DPTH BST. Again, as in Section III.1, we separate the estimation for the pre- and post-announcement periods to examine whether the arrival of public information changes the relationship. The results of the above regressions are reported in Table 5. Panel A reports the results for the abnormal quoted spread regressions during both pre- and post-announcement periods, and Panel B

20 reports the results for the abnormal depth at the best quotes regressions during both pre- and post-announcement periods. The results are summarized as follows. First, both abnormal spreads and abnormal depth at the best quotes are highly persistent as the coefficients of lagged variables in their respective regressions are positive and highly significant. It is thus important to include the lagged abnormal spreads and abnormal depth at the best quotes as control variables in their respective regressions. Second, as expected market orders or trades tend to widen bid-ask spread, whereas limit order submissions tend to reduce bid-ask spread. In the abnormal spread regressions, the coefficient of order submission is significantly negative during both pre- and post-announcement periods. On the other hand, the coefficient of trades is significantly positive during the pre-announcement period but insignificant during the post-announcement period. Third, again, as expected, market orders or trades tend to deplete the depth at best quotes, whereas limit order submissions tend to improve the depth at best quotes. In the regressions of abnormal depth at best quotes, the coefficient of order submission is positive during both pre- and post-announcement periods, and the coefficient of trades is negative during both pre- and post-announcement periods. In all cases, the coefficients are highly significant. Next, we examine the additional effect of HF trades and orders on subsequent market liquidity. As noted earlier, φ HF P TRADE captures the additional effect of HF trades on subsequent market liquidity, whereas φ HF P ORDER captures the additional effect of HF order submission on subsequent market liquidity. In the abnormal spread regressions, φ HF P TRADE is negative in all regressions. That is, HF trades reduce the widening effect of trading on subsequent bid-ask spread. Nevertheless, the effect is stronger during the post-announcement period as the coefficient is significant both 2-year and 10-year notes. On the other hand, HF limit orders tend to reinforce the effect in narrowing subsequent bid-ask spread. The result is rather strong as the coefficient φ HF P ORDER is negative and highly significant in five out of six regressions. Now we turn to the

21 regressions of abnormal depth at best quotes. φ HF P TRADE is positive in five of the six regressions. That is, HF trades tend to reduce the depleting effect of trading on depth at best quotes. Nevertheless, the results are rather weak, especially during the post-announcement period. However, the coefficient φ HF P ORDER is negative and highly significant in all six regressions. That is, HF limit order submission significantly reduces the improving effect of limit orders on the depth at best quotes. The finding holds both before and after announcement. Coupled with results in last subsection on negative contemporaneous relationship with market depth, this finding may suggest that HF orders tend to be conservative and avoid competition with the positions at the best quotes. In fact, Hendershott et al. (2010) in the context of the US equity market also record that the impact of a higher message traffic on the depth of the order book is negative. III.3. The Informativeness of HF Trades and Orders and the Effect on Price Efficiency In this section we investigate whether HF activity promotes or hinders the price discovery process of Treasury securities. The relationship between price discovery and trading activity has been largely explored in the literature and several approaches have been proposed. In our empirical investigation we examine this important issue from two angles: first, we compare HF orders and trades against slow (or manual) orders and trades to assess which group is more informative. We do this by using the test proposed in Kaniel and Liu (2006). More specifically, we divide the whole population of orders and trades into two samples. The first sample consists of the HF orders (trades) while the second sample consists of orders (trades) which are submitted to the market with a delay of 3 seconds or more following market changes. We exclude orders which are submitted more than 1 second but less than 3 seconds following market changes. Intuitively, the Kaniel and Liu (2006) test the informativeness of orders (trades) from the two samples by comparing the actual percentages of orders (trades) in the right side of the market or predicting the correct direction of the market. Correct direction means that a buy (sell) order is followed

22 by higher (lower) mid-quote in the future. If one sample has significantly larger number of quotes in the right of the market than expected, then the sample is relatively more informed than the other sample. More specifically, define P man (1 P man ) as the probability that a submitted order (trade) is a manual or HF order (trade), respectively; n the total number of times the quote midpoint at 9:15 a.m. is in the correct direction (that is above the one at submission for a buy order and below the one at submission for a sell order) following a submission of either a manual or a HF order (trade) and n man the number of midpoint changes in the correct direction that follow manual orders. Under the null hypothesis, Kaniel and Liu (2006) show that out of these n quotes, n man or more is followed by manual order is given by n man np man φ 1 N [ ] (11) n P man (1 P man ) If the probability is lower (higher) than 5% (95%), we reject the null hypothesis of equal informativeness of HF orders (trades) and manual orders (trades) in favor of the alternative that manual (HF) orders (trades) are more informative. In implementing the test, we also divide the orders according to their size: small size (in the bottom tercile), medium size (in the middle tercile) and large size (in the top tercile). The results of the Kaniel and Liu s (2006) testing procedure are reported in Table 6. Panel A reports the probabilities computed for both orders and trades for all bond maturities during the pre-announcement period. Panel B reports the probabilities computed for both orders and trades for all bond maturities during the post-announcement period. We also divide both trades and orders into three size groups. The results for market orders or trades are less conclusive, especially during the pre-announcement period. However, during the post-announcement period, the test favors the informativeness of HF trades. On the other hand, the results for limit orders strongly

23 suggest that manual orders overall tend to be more informative. In particular, during the post-announcement period, the results are rather consistent across bond maturities and order sizes. Overall the results reported in Table 6 are similar in spirit to the ones in Chaboud et al. (2009) who recorded that the share of variance in returns that can be attributed to HF trading is surprisingly small when compared to the share of variance attributed to human trading activity. We assess and complement the above findings by further examining the effect of HF trading on subsequent mid-quotes serial correlation, as suggested in Boehmer and Kelley (2010) and Boehmer et al. (2012). The intuition is that, if prices follow a random walk, quote mid-points autocorrelation should be equal to zero at all horizons. Deviations from zero imply predictability. To put this framework to the data, we compute the quote mid-point return serial correlation within each 5-minute intervals using 1-minute interval data after any macroeconomic news announcement in our sample. As in Boehmer et al. (2012), we estimate the following equation: ORDER C 5m(i) α + γ 1 HF 5m(i 1) ALL + β 3 DPTH 5m(i 1) TRADE + γ 2 HF 5m(i 1) + β 4 P D 5m(i 1) + β 1 VOL 5m(i 1) + γ + ε 5m(i) BST + β 2 DPTH 5m(i 1) where C 5m(i) denotes the absolute value of the mid-quote serial correlation coefficient and the other variables are as discussed in Section 2. The results of the estimation are reported in Table 7. The effect of HF activity on price informativeness is mixed. In fact, abnormal HF orders and trades are only significant at the 5 percent level for the 2-year note. For the remaining bond maturities, there is no statistically meaningful relationship between HF trades and orders and absolute mid-quote return serial correlation. Even in the case of the 2-year note only HF orders are correctly signed. In fact, larger the abnormal HF quoting activity is associated with smaller absolute serial correlation. The sign of (12)

24 the coefficient associated with abnormal HF trades is, however, positive suggesting that larger the HF trading activity is associated with smaller informational efficiency. IV. Conclusion This article explores the role of HF activities and their effects on market quality and the price discovery process of US Treasury securities around macroeconomic news announcement. Using a detailed dataset provided by BrokerTec, we propose and construct measures of HF activity by looking at orders and transactions that have been recorded at a speed that is beyond human capability. Using these new measures we assess i) how HF trades and orders take place around macroeconomic news announcements, ii) whether HF trades and orders increase or deplete market liquidity and iii) the role of HF activities in the price discovery process for US Treasury securities. Our results are as follows: regarding (i), we find that both HF trades and orders submission go up with the arrival of public information and remain high for the next fifteen minute interval. The submission of HF orders increases significantly with the size of public information shock. Regarding (ii), A higher percentage of HF trades does not widen bid-ask spread and does not deplete market depth at best quotes. While a higher percentage HF limit orders tends to further help reducing the bid-ask spread, a higher percentage HF limit orders has a negative effect on market depth at best quotes. Regarding (iii), there is mixed evidence that HF activities contribute to price discovery. We find that HF orders are not better informed than the slow orders. However, there is mixed evidence that HF trades are more informative.

25 References Andersen, T.G., T. Bollerslev, F.X. Diebold and C. Vega, (2003) 'Micro Effects of Macro Announcements: Real-Time Price Discovery in Foreign Exchange, American Economic Review 93, Andersen, T.G., T. Bollerslev, F.X. Diebold, and C. Vega (2007). Real-Time Price Discovery in Global Stock, Bond and Foreign Exchange Markets, Journal of International Economics 73, Aite Group (2008), U.S. Electronic Fixed Income Trading Platforms: The World Ain t Standing Still, Boston. Balduzzi, P., E. Elton, and C. Green (2001), Economic News and Bond Prices: Evidence From the U.S. Treasury Market, Journal of Financial and Quantitative Analysis 36, Barclay, M., T. Hendershott, and K. Kotz (2006), Automation versus Intermediation: Evidence from Treasuries Going Off the Run. Journal of Finance 61, Bias, B, J. Hombert and P. Weill (2010), Trading and Liquidity with Limited Cognition, NBER Working Paper No Biais, B., T. Foucault and S. Moinas (2010), Equilibrium Algorithmic Trading, Working Paper, HEC Paris. Boehmer, E., and E. Kelley, (2009), Institutional Investors and the Informational Efficiency of Prices. Review of Financial Studies 22, Boehmer, E., Fong, K. and Wu, J. (2012), International Evidence on Algorithmic Trading,Working Paper, EDHEC Business School. Boni, L. and Leach, C.J. (2002), Supply Contraction and Trading Protocol: An Examination of Recent Changes in the U.S. Treasury Market, Journal of Money Credit and Banking 34, Boni, L. and Leach, C.J. (2004), Expandable limit order markets, Journal of Financial Markets 7,

26 Brogaard, J. (2011), High Frequency Trading and Market Quality, Working Paper, Northwestern University.. Brogaard, J. (2012a), The Activity of High Frequency Traders,. Working Paper, Northwestern University.. Brogaard, J. (2012b), High Frequency Trading and Volatility,. Working Paper, Northwestern University.. Chaboud, A., B. Chiquoine, E. Hjalmarsson and C. Vega (2009), Rise of the Machines: Algorithmic Trading in the Foreign Exchange Market, Working Paper, Federal Reserve Board. Chlistalla, M. (2011), High Frequency Trading. Better than Its Reputation?, Working Paper, Deutsche Bank Research Clark, E. (2011), The Algorithmic Hare and the Legal Tortoise: High Frequency Trading and the Challenge for Regulators, Working Paper, Griffith University Dichev, I. D., Huang, K and Zhou, D. (2011), The Dark Side of Trading, Working Paper, Emory University. Egginton, J.F., van Ness, B.F., van Ness, R.A. (2012), Quote Stuffing,Working Paper, University of Mississippi. Fleming, M., and E. Remolona (1997), What Moves the Bond Market?, Federal Reserve Bank of New York Economic Policy Review 3, Fleming, M., and E. Remolona (1999), Price Formation and Liquidity in the U.S. Treasury Market: The Response to Public Information. Journal of Finance 54, Fleming, M. J., and B. Mizrach (2009), The Microstructure of a U.S. Treasury ECN: The BrokerTec Platform. Working Paper, Federal Reserve Bank of New York. Green, T. (2004), Economic News and the Impact of Trading on Bond Prices, Journal of Finance 59,

27 Hasbrouck, J. (2012), High-Frequency Quoting: Measurement, Detection and Interpretation, Working Paper, New York University Hasbrouck, J. and G. Saar (2010), Low Latency Trading, Working Paper, New York Univeristy Hendershott, T., C. Jones and A. Menkveld (2011), Does Algorithmic Trading Improves Liquidity?, Journal of Finance 66, 1-33 Hendershott, T. and R. Riordan (2010), Algorithmic Trading and Information, Working Paper, University of California at Berkeley Hoerdahl, P., Remolona, E. and Valente, G. (2012), Macroeconomic Announcements and Risk Premia in the Yield Curve, Working Paper, BIS Jovanovic, B. and Menkveld, A. (2011), Middlemen in Limit-Order Markets, Working Paper, New York University Kaniel, R., and H. Liu (2006), So what orders do informed traders use? Journal of Business 79, Kite, S. (2010), Algos Take Hold in Fixed-Income Markets, Securities Technology Monitor, February. Menkveld, A.J., A. Sarkar and M. van der Wel, (2012), Customer Order Flow, Intermediaries, and Discovery of the Equilibrium Risk-free Rate, Journal of Financial Quantitative Analysis forthcoming. Mizrach, B., and C. Neely (2008), Information Shares in the U.S. Treasury Market, Journal of Banking and Finance 32, Mizrach, B. and C. Neely (2009), The Microstructure of the U.S. Treasury. In Encyclopedia of Complexity and Systems Science, R. A. Meyers (ed.), New York, NY: Springer-Verlag. Moallemi, C.C. and Saglam, M. (2011), The Cost of Latency, Working Paper, Columbia University.

28 Pasquariello, P. and C. Vega (2007), Informed and Strategic Order Flow in the Bond Markets, Review of Financial Studies 20, SEC (2010). Concept Release on Equity Market Structure. Release No ; File No. S Scholtus, M. And D. van Dijk (2012), High-frequency Technical Trading: The Importance of Speed, Tinbergen Institute Discussion Paper No /4. Zhang, F. (2010), High-Frequency Trading, Stock Volatility, and Price Discovery,Working Paper, Yale School of Management.

29 Table 1 Macroeconomic News Announcements This table reports the list of macroeconomic news announcements included in our analysis. N denotes the total number of announcements during the period from January 5, 2004 to June 29, 2007; Day denote the weekday or day of the month of announcement; σ denotes the standard deviation of announcement surprises; N SUR denotes the number of announcements with surprise greater than one or two standard deviations. Announcements N Day N SUR >σ N SUR >2σ Business Inventories 13 Around the 15th of the month 4 1 Change in Nonfarm Payrolls 43 1st Friday of the month 14 3 Consumer Price Index 43 Around the 13th of the month 16 0 Current Account weeks after quarter-end 5 0 Durable Orders 43 Around the 26th of the month 14 2 GDP Advance 15 3rd/4th week of the month for prior quarter 6 3 GDP Final 14 3rd/4th week of 2nd month following the quarter 2 1 GDP Preliminary 14 3rd/4th week of 1st month following quarter 2 1 Housing Starts 43 2 or 3 weeks after the reporting month 10 4 Initial Jobless Claims 187 Thursday weekly 47 9 NY Empire State Index 42 15th/16th of the month 17 2 PCE 43 Around the 1st business day of the month 5 5 Personal Income 43 Around the 1st business day of the month 5 2 Producer Price Index 44 Around the 11th of each month 2 2 Retail Sales 43 Around the 12th of the month 2 2 Trade Balance 43 Around the 20th of the month 12 2