High Frequency Trading and Its Impact on the Performance of Other Investors

Size: px
Start display at page:

Download "High Frequency Trading and Its Impact on the Performance of Other Investors"

Transcription

1 Arhus University Business and Social Sciences High Frequency Trading and Its Impact on the Performance of Other Investors Evidence from the Copenhagen Stock Exchange Master Thesis Authors: Karolis Liaudinskas MSc in Finance Mantas Malkevicius MSc in Finance Supervisor: Carsten Tanggaard Professor of Finance Department of Economics and Business August 2012

2 Abstract This paper investigates the activity of high frequency traders in the Copenhagen Stock Exchange (CSE), and focuses on the impact that HFT has on other market participants. The authors provide a thorough review of both the institutional background and the relevant academic literature. Moreover, they perform an empirical analysis examining the HFT s impact on the stock price efficiency, and the HFTs involvement in such predatory trading practices as front running and price trend creation. A unique data set, provided by the CSE, enables the authors to distinguish between the trades executed by HFTs. The methodologies applied in the research are consistent with Litzenberger et al. (2010), Brogaard (2010) and Aldrige (2011). The results of this paper suggest the following. Firstly, HFT activities tend to improve the stock price efficiency. Secondly, HFTs do not tend to get involved in price trend creation strategies. Thirdly, HFTs tend to systematically pursue front running strategies at the expense of other market participants.! ii!

3 Table of Contents! 1. Introduction... 1! 1.1. Controversy and motivation for the study!...!1! 1.2. Research question!...!3! 1.3. Empirical research and main findings!...!4! 1.4. Outline of the paper!...!4! 2. Background Information... 5! 2.1. Market microstructure!...!5! 2.2. Evolution of markets and market fragmentation!...!7! Evolution of markets!...!7! Market fragmentation!...!9! 2.3. High frequency trading and related concepts!...!11! Definitions!...!12! AT Characteristics!...!12! HFT Characteristics!...!13! A glance at HFT strategies!...!15! 2.4. Danish Stock Market!...!16! General Characteristics!...!16! Trading Practices!...!17! 2.5. Regulations in US and Europe!...!19! Differences between the U.S. and the European Market System!...!19! HFT regulations in U.S.!...!20! HFT regulations in Europe!...!22! 3. HFT Strategies... 26! 4. Literature Review... 36! 4.1. HFT effects on market quality!...!36! Theoretical models!...!36! Empirical studies!...!38! 4.2. Empirical research on HFT strategies!...!43! Initiation of price trends!...!43! Front running!...!46! 5. Data... 49! 5.1. Why Danish stock market?!...!49!! iii!

4 5.2. Dataset!...!50! 6. Methodology... 53! 6.1. Testing price efficiency!...!53! 6.2. Testing predatory trading strategies!...!60! Price trend creation!...!61! Front running strategies!...!64! 7. Results... 67! 7.1. Price efficiency!...!67! Historical CSE price efficiency analysis!...!67! CSE Price efficiency analysis. Shorter time spans!...!68! Random Walk analysis on a micro level. Relation with HFT!...!70! 7.2. Price trend creation!...!72! Main results!...!72! Weekly results!...!75! Robustness check!...!77! 7.3. Front running!...!79! Replicating the Brogaard s test!...!79! Taking the whole HFT activity into account!...!81! 8. Discussion... 83! 8.1. Implications of findings!...!83! 8.2. Answering the research questions!...!85! 8.3. External validity!...!85! 8.4. Regulatory discussions!...!85! 8.5. Data limitations and suggestions for future research!...!87! 9. Conclusion... 89! 10. Reference List... 91!! iv!

5 1. Introduction Financial markets have changed dramatically over the last few decades. Ever since computers were introduced in the industry in the 1970 s, the extent of human interaction in the trading process has been constantly decreasing (Cliff et al. 2010). Not only did the traditional trading venues become automated but also new players called electronic communication networks (ECNs) entered the market. Computerization and competition among exchanges established better conditions for traders, who were able to trade increasingly faster. In turn, traders developed their systems in order to automate not only the execution process but also decision making. All these developments created a niche in the industry and facilitated the entrance of new market participants called High Frequency Traders (HFTs). Their general trading strategy is based on the speed of execution and extremely short-term price movements, rather than fundamentals (Papagiannis, 2010) Controversy and motivation for the study High frequency trading (HFT) is a relatively new topic as it was first brought to the public s attention in 2009, when The New York Times published one of the first articles regarding the issue (Duhigg 2009). Today, however, high frequency trading is one of the most controversial and actively discussed topics in the financial world. There are a number of reasons why HFT attracts so much attention and why it needs to be studied. First of all, HFTs constitute a significant part of the total trading activity in stock, derivatives and foreign exchange markets. For instance, in the U.S. stock market high frequency traders generated around 55% of trading volume in 2011, and some sources report even higher numbers (Patterson & Eaglesham 2012). In Europe this number reached 42%, and it has been growing rapidly (Sybase 2012). Secondly, the event, known as the Flash Crash, triggered a wave of accusations towards high frequency traders. On May 6, 2010 US stock market experienced a sudden swing in prices, when the Dow Jones Industrial Average suddenly plummeted by 9% and recovered within a minute. HFTs were blamed to have used abusive trading strategies, which exacerbated price volatility and created price trends (Nanex 2010).! 1!

6 Thirdly, very little is known about how exactly HFTs operate, what strategies they employ, and what impact they have on the overall market quality. The opaqueness of the issue exists due to the fact that complicated trading algorithms, together with the speed of execution, are the key sources of competitive advantage for HFTs. Firms specializing in this type of trading tend to avoid revealing any inside information by all possible means (Friederich & Payne 2011). Fourthly, as the issue is relatively new, there has been little academic research made on HFT, although the amount of relevant literature is rapidly growing (Gomber et al. 2011). Furthermore, academics fail to arrive at a common conclusion regarding the HFT s impact on price efficiency, volatility and liquidity (Brogaard 2010) (Zhang 2010). These market characteristics significantly affect other market participants when making investment decisions (Brogaard 2010). Fifthly, the literature and empirical evidence concerning particular HFT strategies and predatory trading practices is particularly scarce. Even strategies that are generally considered to improve market quality (e.g. market making improves liquidity and statistical arbitrage improves price efficiency) are questioned by opponents of HFT (Arnuk & Saluzzi 2008). On top of that, institutional investors accuse HFTs of applying predatory strategies that directly exploit them either through price trend creation or liquidity detection and front running. A recent survey by Liquidnet (2011) revealed that 66%, 60% and 50% of institutional investors in the U.S., Europe and Asia Pacific, respectively, are concerned about HFTs. Institutional investors consist of pension funds, investment funds etc., thus the ultimate victims of predatory trading would be not only rich corporations and individuals but also ordinary people (Friederich & Payne 2011). However, there is little or no direct evidence of predatory practices being employed by HFTs (McGowan 2010). Sixthly, the extent of HFT activities and the investments made by these firms into hardware, software, co-location and human resources, suggests that they achieve very high profitability. Brogaard (2010) estimated aggregate yearly profits of HFTs operating in the U.S. to be $3 billion, while Kearns et al. (2010) stated that $3.4 billion is the upper bound of HFT profitability. According to Aite Group (2009), there are around 400 HFT firms in the U.S., which means that on average each of them earns approximately $8 million per year. The question remains whether these profits are not! 2!

7 generated unfairly at the expense of other traders and investors (Friederich & Payne 2011). Finally, authorities in both USA and Europe have procrastinated the introduction of laws that would regulate HFT activities. Although institutional investors and other traders are pressuring the SEC and the European Commission to restrain HFTs, the opaqueness and the lack of empirical evidence concerning their activities make it difficult to formulate optimum regulations (Patterson & Eaglesham 2012), (Liquidnet 2011). Some evidence suggests that HFT tend to enhance the overall market quality, therefore, some hasty restrictions on HFTs may eventually harm all market participants (Gomber et al. 2011) Research question Due to all the mentioned reasons the topic of high frequency trading is highly controversial and further research is essential. The purpose of our paper is threefold: (1) to provide a thorough background of the HFT industry, (2) to summarize the relevant up-to-date academic literature, and (3) to contribute to the existing literature by empirically investigating the effect of high frequency trading on other market participants. In particular, due to either lacking or contradicting empirical evidence, we focus our research on HFT s involvement in predatory trading and its overall impact on the stock price efficiency. Consequently, we formulate our set of research questions as follows. (1) What effect does high frequency trading have on the stock price efficiency? (2) Do high frequency traders employ predatory trading strategies that harm other market participants? These two questions are closely related. According to Hendershott and Riodan (2011), high frequency traders can be compared to other types of intermediaries and speculators. Generally speculators tend to enhance price efficiency as they trade against mispricings and thus add more information into prices. Predatory trading and manipulative strategies, on the other hand, can worsen price efficiency (Hendershott & Riodan 2011). Therefore, the evidence of HFT having a negative impact on the stock price efficiency would suggest that HFTs tend to engage in predatory trading relatively more than in strategies improving efficiency (e.g. statistical arbitrage).! 3!

8 1.3. Empirical research and main findings In order to investigate our research questions, we apply the methodologies developed by Litzenberger et al. (2010), Brogaard (2010) and Aldrige (2011). We follow Litzenberger et al. (2010) to study the relationship between HFT and the stock price efficiency. Aldrige s (2011) methodology is employed to investigate the feasibility of pump-anddump arbitrage, and thus, the incentives of HFTs to initiate price trends. Brogaard s (2010) research is used to study the tendency of HFTs to exploit other traders by front running. One of the major advantages of our research is the superiority of our dataset. As we use the trade data of large cap companies listed on Copenhagen Stock Exchange (CSE), we explicitly observe details of each trade that occurred on the exchange in January and February, Furthermore, we are able to identify trades, which HFTs participated in, using the same approach as Nasdaq OMX Nordic. The results of our research suggest that high frequency trading generally tends to improve price efficiency. We also find that HFTs are not able to exploit other traders by creating price trends, as we do not find evidence of pump-and-dump arbitrage feasibility. However, our evidence suggests that HFTs tend to front run other traders Outline of the paper The remainder of the paper is organized as follows. Section 2 provides a thorough background of high frequency trading. Section 3 discuses the most common HFT strategies. Section 4 reviews the relevant academic literature on HFT. Section 5 presents the data used for the empirical research. Section 6 describes the methodologies employed to study both of our research questions. Section 7 presents the results. Section 8 discusses the main findings and gives suggestions for practical implications. Section 9 concludes. As our empirical research focuses on two research questions, the sections of Literature review, Methodology and Results are each divided into two broad parts, namely price efficiency and predatory trading. Furthermore, we investigate two types of predatory trading: price trend creation and front running, thus the predatory trading parts throughout the paper are divided accordingly as well.! 4!

9 2. Background Information 2.1. Market microstructure In order to understand how the majority of modern stock markets work, we provide a simple description of the market mechanism. This background is necessary to fully understand high frequency trading strategies discussed in subsequent sections of this paper. According to Kearns et al. (2010), the most widely used market mechanism is called the open limit order book. It is used to implement a type of continuous double auction and works in the following manner. Suppose a trader decides to buy 500 shares of some stock. He can submit a limit order that specifies the volume (in this case, 500) and the maximum price he is willing to pay for the stock (e.g. $10.02). Assume that currently there is no seller willing to sell for $10.02 or less. The order gets registered in the buy order book. This book consists of all buy limit orders, sorted by price. If there are orders submitted at the same price, they are sorted by time of submission, the oldest ones being the first to be executed. The highest price is called the bid and is placed at the top of the order book. The exchange also keeps a sell order book, which lists submitted orders to sell the stock. The lowest price is placed at the top of the book and is called the ask (Biais et al. 1995). The difference between the bid and ask prices is called the bid-ask spread, and these two most competitive prices are together called the inside market. All these orders lying in the limit order book are called passive orders as they supply liquidity to other market participants (Kearns et al. 2010). If a trader wants to purchase shares immediately at the best prices available, he can submit a marketable order, or equivalently, a buy limit order at the price higher than the current ask. Similarly, for an immediate sell, one should submit a sell limit order at the price lower than the current bid. Such orders are called aggressive orders as they consume liquidity. Table 1 shows an example of a limit order book. In the example, if an investor now decided to submit a buy limit order of 500 shares at $10.10, the trade would be executed immediately by filling the three sell limit orders at the top of the sell order book. As a result, our investor would purchase 100 shares at $10.08, 200 shares at $10.09 and 100 shares at $ The remaining 100 shares would be placed in the top! 5!

10 of the buy order book and the new bid would be $ The new ask would be $ Every submitted order can be canceled before it is executed (Kearns et al. 2010). Table 1. The example of the limit order book Buy orders Sell order Shares Price Shares Price 200 $ $ $ $ $ $ $ $10.11 After orders get matched, clearing houses have to make sure that securities are transferred and payments are made. Only then a trade is considered finished. It takes three days to clear and settle the trade (Menkveld 2012). Most of exchanges use this simple market mechanism, however, they often have some additional more complicated order types (Kearns et al. 2010). For instance, Nasdaq OMX Nordic, which we study in our research, offers order types such as pegged orders, hidden orders and reserve orders (Nasdaq OMX 2012d). According to Nasdaq OMX (2012d), pegged orders enable traders to price their orders relative to the current market price. The trading system adjusts the prices of such orders automatically as the Best Bid Offer (BBO) changes. Hidden orders are limit orders that are not displayed in the order book. They are executed only when there are no visible orders left at that price. Despite the priority for visible orders, hidden orders among themselves are executed like ordinary limit orders in price/time priority. Hidden orders lying in the order book are often referred to as hidden liquidity. A reserve order, sometimes also called iceberg order, allows traders to specify what portion of the total order they want to appear in the order book and what part they want to submit as a hidden order (Nasdaq OMX 2012d).! 6!

11 While the bid-ask spread is a considerable part of investors transaction costs, it is also seen as a compensation for market makers (traders who provide liquidity) for the costs incurred by holding the position (Menkveld 2012). There are three types of costs, namely (1) order-handling cost (e.g. a fee paid to exchanges for executing orders), (2) the cost of being picked off by better informed traders, and (3) a risk of a sudden price change (Madhavan 2000). Recently, however, due to intense competition, in order to attract market makers, many exchanges started paying liquidity rebates for liquidity providers, and charging liquidity takers with extra fees (Gomber et al. 2011). The other two components of the cost incurred by market makers have been shrinking as well. Introduction of computers to trading venues resulted in lower latency (faster response time), which enabled market makers to quickly update quotes in case of arrival of new public information. As a result, they are less likely to be adversely selected by better informed traders. All these changes explain the dramatic decrease in bid-ask spreads across markets over the last decade (Menkveld 2012) Evolution of markets and market fragmentation Evolution of markets According to Cliff et al. (2010), the whole history of financial markets is characterized by gradual increase in the speed of communication and data processing. Before computers were invented, people who possessed extraordinary arithmetic skills gained a significant advantage and outperformed other market participants. Means of communication have also changed dramatically. In the 19 th century, important financial information was transferred by horse-riding messengers, who were later replaced by faster carrier pigeons. The invention of telegraph reshaped the way of communication, which was further fastened by telephone (Cliff et al. 2010). Up until the last quarter of the 20 th century, the only way to trade stock was gathering into floor-based exchanges and meeting dealers physically. Only licensed brokers could access stock markets and trade on behalf of individual investors (McGowan 2010). Eventually, in the mid 1970 s, financial markets started adopting computers. In 1976, the New York Stock Exchange introduced the designated order turnaround (DOT) system, which was upgraded to Super-DOT in This system made it possible to submit buy and sell orders to specialists electronically (Gyurko 2011). In 2000, around! 7!

12 90% of trades at NYSE were implemented through SuperDot, and only the largest trades were still submitted and executed physically by humans (Markham et al. 2008). Once trading venues started establishing computerized communication systems, order execution became much faster and traders could choose to be connected to trading platforms instead of meeting physically (McGowan 2010). In addition to NYSE developments, NASDAQ became the first electronic market in Instead of having one specialist providing liquidity for each stock, it offered the opportunity for dealers to compete in market making activity (Markham et al. 2008). According to McGowan (2010), the introduction of computers in financial markets resulted in the development of a new trading strategy called program trading, widely used in 1980 s. The strategy consisted of buying (or selling) stock index futures (e.g. S&P 500) and simultaneously selling (or buying) corresponding equity. This double order could be programmed to be executed whenever prices of stocks and corresponding futures moved apart too much (McGowan 2010). This type of automated trading, called index arbitrage, has been considered to cause the Black Monday crash in October 1987 (Cliff et al. 2010). The Black Monday lead to an increased skepticism towards computer based trading. Nevertheless, consistently with the Moore s Law, the costs of computers dropped by half every two years (Cliff 2011). This process continuously created a soil for more intelligent programs. By the end of the century, computers became an integral part of the investment funds management. Sophisticated programs were used to identify and trade on securities, which would help to diminish or even eliminate the portfolio risk. This hedging practice gave birth to the concept of hedge funds (Cliff 2011). In the beginning of the 21 st century, automated trading systems still focused on order execution rather than decision-making (Cliff 2011). When a human made a decision to purchase or sell a particular asset, an execution was implemented by automated execution system (AES). It considered the optimal timing and portions of the order to be submitted. Eventually, financial institutions started experimenting with AES and created different sophisticated algorithms in order to find ways to further reduce the market impact. Consequently, the concept of algorithmic trading was introduced (Cliff 2011). The boom of algorithmic trading was stimulated by decimalization of stock! 8!

13 prices in US in Instead of 1/16 of a dollar per share, 1 penny became the minimum tick size, which increased liquidity in markets, and attracted more algorithmic traders (Moyer & Lambert 2009). Simultaneously with the development of AES, various models searching for arbitrage opportunities evolved (Cliff et al. 2010). The new programs were able to scan and identify profitable statistical arbitrage based on not only two but rather on thousands of assets. This breakthrough was enabled by both powerful machines used by traders to analyze markets, and the computerized trading infrastructure. Straight Through Processing (STP) was introduced in trading venues, which significantly reduced the latency of order execution. The whole process from the submission of an order to clearing and settlement became fully computerized. Furthermore, Direct Market Access (DMA) provided an opportunity for traders and investors to interact with exchanges directly without the intermediation of investment banks or brokers. Both of these factors increased the speed of execution, which is vital for exploitation of short-lasting arbitrage opportunities (Cliff et al. 2010) Market fragmentation Meanwhile, financial industry became increasingly fragmented. Market fragmentation has been characterized by both the introduction of new entities involved in securities trading, and the establishments of new electronic exchanges such as Chi-X or BATS (Better Alternative Trading System) (Financial Times Lexicon 2012). Traders were enabled to trade the same securities at different venues simultaneously, which provided more incentives to seek for pricing inconsistencies and arbitrage opportunities. Moreover, a tightened competition among market venues drove bid-ask spreads down and facilitated the establishment of new entities involved in trading (Cliff 2011). The fragmentation of the financial industry started in 1990s with the establishments of automated trading systems that disseminated orders in trading venues for third parties and dealers and could execute such orders within the networks themselves (McAndrews & Stefandis 2000, pp. 1). These systems, named electronic communication networks (ECNs), provided new opportunities to traders and facilitated the development of new trading strategies and algorithms (McGowan 2010).! 9!

14 ECNs could execute orders sent by registered individuals (without the intermediation of brokers) internally in their networks. However, it could also forward orders to primary exchanges (e.g. NYSE or NASDAQ) if they offered better deals for ECNs clients. Therefore, at first, authorities perceived ECNs as broker-dealers and did not require them to be registered as security exchanges (McGowan 2010). However, a few ECNs gained a considerable market share and were willing to register as exchanges. In 1998 the U.S. Securities and Exchange commission (SEC) issued Regulation Alternative Trading Systems (Reg. ATS), which officially acknowledged ECNs as separate trading venues. In this way they could engage into direct competition with traditional financial markets (Markham et al. 2008). However, in 2005 the SEC released a Regulation National Market System (Reg. NMS), which partly restricted the flexibility of actions available to ECNs in the U.S. For example, a Trade Through Rule (Rule 611) said that a marketable order always has to be forwarded to and executed at a trading venue, which offers the best price nationally (McGowan 2010). Nevertheless, it did not discourage ECNs from the competition. For instance, BATS achieved 10% market share in US equities in the first couple of years of operation, as it became a licensed exchange in 2007 (Gyurko 2011). A successful example of an ECN in Europe is Chi-X that was launched on April 16, 2007 and has been successfully competing with other exchanges. It is considered to be one of the fastest trading platforms in the investment industry with the latency of only two milliseconds. During the first year of operations in Europe the venue traded stocks in six Western European countries and captured 4.7% of all trades. In 2011, Federation of European Exchanges acknowledged Chi-X as the largest equity market in Europe (Menkveld 2012). A typical ECN provides automatized fast order execution at low cost (Stoll 2006). However, today there is a wide variety of ECNs that differ from each other with respect to (1) targeted clientele, (2) order routing strategies (e.g. some ECNs simply route orders to other networks), (3) speed, (4) quality and certainty of execution, and (5) accessibility to limit-order books (Gyurko 2011). Another type of venue, which evolved as a solution to reduce the market impact, is a dark pool. Institutional investors sometimes wish to sell or buy a large block of a financial instrument, and doing so on typical exchanges or ECNs would result in a! 10!

15 strong impact on the asset price. In order to reduce the market impact, investors can divide a large order into many small orders and submit them for execution one by one. Alternatively, they can execute trades in dark pools, as they offer anonymity of traders and prevent information leakage (Gyurko 2011). Furthermore, limit orders in dark pools are not quoted in the order books, thus execution is uncertain and unpredictable. Finally, in dark pools trades are executed at prices that are determined at primary exchanges rather than within dark pools (Gyurko 2011). All these features of dark pools contribute to the reduction of market impact when selling or purchasing large blocks of stocks. All in all, the combination of (1) cheap computer power, (2) rapid development of sophisticated trading programs, (3) direct access to trading venues, (4) favorable regulations and (5) fragmentation of the financial industry created new trading opportunities for market participants. It lead to the development of new trading strategies, which were based on expected short-term asset values. As a result, these rapid changes in the financial industry initiated a race among traders, where the key competitive advantage is the high speed of data analysis and order execution. In order to completely eliminate the intervention of humans and further increase the speed, trading algorithms were programmed not only to execute orders but also to make trading decisions based on observed order flows. These traders held positions for a matter of seconds or milliseconds and became known as High Frequency Traders. (Cliff 2011) 2.3. High frequency trading and related concepts Today high frequency trading is dominating in financial markets in terms of trading volume both in the U.S. and Europe (McGowan 2010). According to the TABB Group, high frequency trading generated 61% of U.S. stock-trading volume in 2009 and dropped marginally to 56% in 2010 and 55% in 2011 (Patterson & Eaglesham 2012). These numbers are impressive, considering that HFT firms 1 constitute only 2% of the overall 20,000 companies involved in trading in the U.S. (Aite Group 2009). Meanwhile in Europe in 2010, high frequency traders were involved in roughly 33% of trades, which made up around 32% of the total trading volume. In 2011, the share of HFT reached 42% of trading volume, while it is projected to increase up to 45% in 2012 (Sybase 2012).!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 1!E.g.!Getco LLC, Knight Capital Group, Citadel LLC, Jump Trading etc.! 2 In this paper HFT(s) can also refer to high frequency trader(s).! 11!

16 Although the involvement of HFT has been growing the opaqueness of HFT strategies makes it difficult not only to discuss their overall impact on financial markets but also to define the HFT itself (Friederich & Payne 2011). The purpose of this section is to define high frequency trading and introduce its key characteristics Definitions Due to the fact that high frequency trading (HFT) 2 is a rather new, rapidly changing and relative term, there is no general agreement on its single definition. As a result, academics often refer to each other and use similar definitions in order to remain consistent while defining HFT (Gomber et al. 2011). In order to be consistent with the existing literature, in our paper we define HFT and algorithmic trading (AT) in line with Brogaard (2010) and Hendershott & Riordan (2011). According to Brogaard (2010), high frequency trading is a type of investment strategy that is engaged in buying and selling financial assets very rapidly by using computer algorithms, and holding those assets for a very short period (a matter of seconds and microseconds). HFT firms are defined as firms engaged in proprietary high frequency trading, who tend to hold neutral positions in assets overnight. Algorithmic trading is defined as the use of computer algorithms to automatically make trading decisions, submit orders, and manage those orders after submission (Hendershott & Riordan 2011, pp. 2). HFT and AT are similar in a sense that both use automated decision making technology, however AT investment horizon is not specified and could be any period of time from seconds to years. Meanwhile HFT tend to hold positions only for extremely short periods. Therefore, HFT is a subset of algorithmic trading (Brogaard 2010) AT Characteristics Most non-hft algorithmic trading strategies are employed by institutional investors, who attempt to reduce the market impact of large orders. For instance, if an institutional investor decides to sell a large block of stock and submit the whole block for execution, it should significantly decrease the price in order to encourage other market participants!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 2 In this paper HFT(s) can also refer to high frequency trader(s).! 12!

17 to absorb the increased supply (Angel et al. 2010). Alternatively investors can use algorithms that divide large trade orders into a number of smaller orders and submit them one by one into the market for execution. The most common non-hft algorithms are classified into four generations by Almgren (2009). Order execution with first generation algorithms is based on benchmarks that depend on the existing situation in a market. They do not depend on the actual order characteristics or the situation in the order book. For instance, Participation Rate Algorithms could be programmed to participate in a market by constantly trading 3% of the total market volume until the desired position is liquidated or built. Another example is Time Weighted Average Price (TWAP) algorithms, which can split an order and submit smaller orders for execution in equal time intervals. The length of time intervals and the order size are usually predefined by humans. A third example is Volume Weighted Average Price (VWAP) algorithms. They attempt to execute orders at a price, which would be the same or better than a volume weighted average price observed in the market. (Almgren 2009) The second generation algorithms are more intelligent as they attempt to tailor a specific benchmark for every individual order and take the timing risk (potential negative price movements during the execution process) into account. The third generation algorithms constantly re-evaluate their order execution schedule and adapt to new market conditions. The fourth generation is programmed to automatically react to the news concerning particular instruments. These algorithms can employ text-mining techniques or simply subscribe to low latency electronically processable news feeds provided by news agencies and exchanges. (Gomber et al. 2011) These non-hft algorithmic trading strategies are relatively easy to predict and exploit, and high frequency traders are often blamed to be engaged in such activity. The next two sub-sections present HFT characteristics, which distinguish it from other types of trading, and shortly introduce to the main strategies employed by HF traders HFT Characteristics Generally speaking, high frequency trading specializes in capturing small gains on short-term fluctuations of asset prices. Most HFT firms attempt to find temporary mispricings and other inefficiencies in financial markets and exploit them before they! 13!

18 disappear. Other HF traders are engaged in market making at high frequency. As a result, one of the main characteristics of HFT is high capital turnover. Other characteristics include a dependence on low latency, a relatively short shelf life of algorithms, and participation on multiple trading venues (Iati 2009). Latency is a speed factor, which describes the delay experienced in a system. The lower latency in trading systems results in faster execution of trades. Ultra-low latency, which provides the ability to execute trades in less than 1 microsecond, is a vital component of all high frequency trading strategies. In order to earn profits, a high frequency trader has to process information through his algorithms microseconds faster than his competitors. Therefore, HF traders constantly upgrade their hardware in order to remain competitive. (Mackenzie 2009) Furthermore, in order to further reduce latency, various trading venues started providing facilities for HFT firms to co-locate their servers next to the exchanges. In this way, real-time market information can reach a high frequency trading platform instantaneously. The platform can process the information, create orders and submit them back to the exchange s server for execution. All this procedure can be done in less than a millisecond (Iati 2009). Profits earned from a single trade usually amounts to pennies or less. Therefore, HFT firms execute thousands of trades everyday in order to earn significant profits (McGowan 2010). HFT firms also purchase real estate in different cities next to the buildings of different exchanges and set up their offices. Consequently, real estate prices around exchanges skyrocketed; however, HFT firms are still willing to pay that money. This fact testifies the significance of benefits provided by co-location (Mackenzie 2009). Although low latency and speed of execution is a vital source of competitive advantage in HFT business, firms also compete in recruiting specialists, who would be able to create and update competitive trading algorithms. Traders that graduate from the best schools with mathematics and computer science degrees are particularly demanded (Wahba & Chasan 2009). It is important to be in possession of intelligent specialists, since the shelf life of trading algorithms is very limited. In order to retain a competitive advantage, HFT firms have to update their codes regularly - sometimes a few times per week (McGowan 2010). There are two reasons why the usefulness of algorithms dilutes over time. Firstly, high frequency trading strategies rely on correlations among markets! 14!

19 and individual securities. In today s extremely volatile markets, various market and stock characteristics change rapidly, thus financial engineers have to react and adjust codes accordingly. Secondly, trading algorithms have to be altered regularly because of the threat of reverse engineering by competitors. If a rival trading firm learns a trading pattern of its competitor, it can exploit this knowledge to its advantage. The most profitable strategies can become the riskiest in one day, if rivals manage to predict algorithm s behavior. As a result, in order not to become a victim of reverse engineering and to retain the competitive advantage, a firm has to constantly alter its codes. (McGowan 2010) As mentioned earlier, another major characteristic of HFT is participation on multiple trading venues. HFT firms trade in different asset classes including stocks, options, futures, currencies etc. Moreover, they trade the same assets on different exchanges simultaneously (McGowan 2010) A glance at HFT strategies The strategies employed by HF traders are too opaque and diverse to discuss them all (Gomber et al. 2011). However, some of them are well known and not new to the market. In fact, most of the known HFT strategies are old strategies implemented with better technology and at higher speed (Friederich & Payne. 2011). Gomber et al. (2011) suggests the following classification of strategies pursued by HF traders. We enrich this list with strategies identified in other sources (McGowan 2010; LSEG 2010; Menkveld 2012; Brogaard 2010; Angel et al. 2010; Egginton et al. 2012; Arnuk & Saluzzi, 2009). 1. Electronic liquidity provision a. Market making b. Liquidity rebate trading 2. Market Arbitrage a. Market neutral arbitrage b. Cross asset, cross market & ETF arbitrage 3. Liquidity detection a. Pinging/Sniffing/Sniping b. Quote Matching! 15!

20 4. Other a. Latency arbitrage b. Short Term Momentum c. Predatory algorithms d. Quote stuffing e. Spoofing and Layering Market making and market arbitrage strategies are considered to be beneficial to the market as a whole, since these strategies either provide liquidity and narrow bid-ask spreads (MacGowan 2010) or increase price efficiency (Hendershott & Riodan 2011). However, other practices such as liquidity rebate trading, pinging, quote matching, latency arbitrage, predatory algorithms, quote stuffing, spoofing and layering are very questionable as they are considered to help HF traders unfairly exploit their speed advantage at the expense of slower institutional investors and other traders. Some of those strategies, namely, quote stuffing, spoofing and layering, are already scrutinized and planned to be banned by authorities in US and Europe (Gomber et al. 2011). However, even market making and market arbitrage strategies can gain a negative perception concerning fairness, if they are combined with liquidity detection techniques, such as pinging (Arnuk & Saluzzi 2009). Our paper provides a thorough description of HFT strategies. It is enriched with particular examples and explanations of how they work. This discussion, however, is rather extensive; therefore, we provide it separately in the next section named HFT Strategies Danish Stock Market General Characteristics The Danish stock market is quite fragmented with 62.89% of all trades being executed in lit stock exchanges, 2.72% attributable to dark pool and the rest 34.39% traded over the counter. What concerns exchange-traded stocks, in January, 2012, 48.49% of all Danish stocks were traded on Copenhagen Stock Exchange (CSE) and the rest 14.42% were spread among 5 other alternative market providers with Chi-X gaining the largest share of 8.59% (Nasdaq OMX 2012a).! 16!

21 The exchange controlling the highest share of Danish stocks traded, Copenhagen Stock Exchange (CSE), is a part of NASDAQ OMX Group, which is the largest single cash equities securities market in the world in terms of share value traded (Nasdaq OMX 2012b). In the exchange shares, bonds, treasury bills, financial futures and options are traded. Currently there are 175 companies listed in the CSE with an average equity turnover of around 2.6 billion Danish kroner per day (Nasdaq OMX 2012c). The exchange maintains the European Best Bid and Offer, the European version of the National Best Bid Offer in the U.S. In the CSE companies are categorized into three segments, namely Large Cap, Mid Cap and Small Cap firms, according to their market capitalization level. Large Cap segment includes companies with market capitalization of 1 billion Euros or more, while Mid Cap firms have capitalization level between 0.15 and 1 billion Euros, and the Small Cap segment comprises of companies with market capitalization of less than 0.15 billion Euros Trading Practices During regular trading day, there are four main trading sessions in the CSE, namely Opening, Continuous Trading, Closing and After Market. The trades are executed using one of the most advanced trading platforms called INET. It was introduced on the 8 th of February, 2010 both in Nordic and Baltic branches of Nasdaq OMX, CSE included. Since then all equity markets operated by Nasdaq OMX around the world have been using the same trading platform that enables investors to access distant markets easier. The INET system is able to handle around one million messages per second at extremely low latency level (Nasdaq OMX 2010). Currently the number of trades in Nasdaq OMX Copenhagen is around 50,000 per day (Nasdaq OMX 2012c). The majority of trades, around 95%, comprises of trades in large capitalization companies. The Figure 1 summarizes the evolution of average daily trades executed and average trade size in the CSE during past few years.! 17!

22 Figure 1. The average trade size and average dauly number of trades executed in CSE. Average trade size expresed in DKK. Data source: Nasdaq OMX.! Average'Trade'Size'in'CSE' ,000 50,000 40,000 30,000 20,000 10,000 Average'Daily'number'of' Trades'in'CSE' As it can be seen from the Figure 1, the daily trade number has experienced a large growth, accounting to 139% from the year On the other hand, average daily trade size has diminished considerably by 74%. Such changes can be attributed mainly to the increased market fragmentation and lower displayed liquidity due to the tick-size change in 2010 (increased up to 4 decimals) (Nasdaq OMX, 2012a). What concerns order size, electronic trading systems often slice the order into smaller parts. Quite similarly, smart order routing, also used in the CSE, splits an order into smaller bits and then searches for the best execution options (Hatrick & Deliya 2008). Thus, the evolution of electronic trading has reduced the average size of the order considerably. In the CSE, the order flow comes from personal broker accounts, routing accounts or algorithmic accounts. Volume traded from algorithmic accounts increased rapidly during the last couple of years in line with the explosion in the number of trades executed daily in the exchange (Nasdaq OMX 2012a). From the legal perspective, in order to deploy such AT practices in the CSE, investors have to obtain automated trading rights. It entitles investors to trade through automated trading facilities using software that automatically generates orders in response to specific pre-programmed factors. A special form of an automated trading account, called AUTD, can also be set up, which entitles investors to a specific discount on a current stock price. However, the AUTD account has to be used purely for automatic trading and other execution practices are not allowed (Nasdaq OMX 2012d).! 18!

23 Co-location services that enable exchange customers to reduce the response latency to the minimum, is of high importance to Algorithmic and HF traders. Nasdaq OMX Nordic, including Copenhagen OMX, offers its customers to acquire space for their servers in the same datacenter where the exchange s central matching machine operates as well as other support services (Nasdaq OMX 2012e). The extremely low-latency execution system INET, co-location services and the market fragmentation offering profitable trading opportunities, make the CSE attractive to HF traders. As a result, an increasing trend of HFT activity has been observed lately. According to Nasdaq OMX data, the percentage of HFT shares in total equity turnover almost tripled during last year and amounted to 7.5% (Nasdaq OMX 2012a) Regulations in US and Europe In this section we provide an overview of the main regulatory practices implemented and the initiatives proposed in both the U.S. and Europe in order to control the activities of HFTs. Closely following the paper of Gomber et al. (2011), the overall differences between the two market systems are discussed. Then the overview of proposals on how to manage risk and ensure fairness in both markets is presented Differences between the U.S. and the European Market System One key difference between the two market systems is the approach to the best execution regime. While the U.S. practices a rules-based approach, Europe favors a principle-based method. There are two main features enforced by the U.S. Regulation Market System that guarantee fair execution of trades at the best prices prevalent among exchanges. The first one is the National Best Bid Offer (NBBO), which is the aggregated best bid/ask spread for any stock traded around U.S. Every marketplace has to distribute their best quote for every security traded which then is aggregated nationwide to arrive at the NBBO. To ensure that trades are always executed at the NBBO, the regulation 611 REG NMS, also known as the order protection rule, was implemented (SEC 2005). It forbids the exchanges to trade at the prices worse than those quoted in the NBBO. In case the marketplace is not able to match the NBBO, it has to route the order to another exchange where such a quote is available. Once implemented, this rule has increased the competition among marketplaces that resulted in better quotes for investors (Gomber et al. 2011).! 19!

24 In Europe, Markets in Financial Instruments Directive (MiFID) requires investment firms to ensure that trade is executed at the most favorable terms for their customers (European Commission 2004). It differs from the U.S. not only by the fact that it is not a strict rule, but also that the party responsible for the fair execution is an investment firm. In contrast, in the U.S. the responsible party is the exchange. Nevertheless, both regimes have made it easier for new marketplaces to compete with the old established ones and thus, have improved the prices offered to customers HFT regulations in U.S. In the following section the most debatable areas concerning HFT regulations in the U.S. are presented and the attitude regulatory bodies have towards them is described. Flash orders There is an exception to the order protection rule stating that a trade can be executed at the least aggressive quote 3 present over the previous 1 second at the exchange where the trade is being executed (McInish & Upson 2012). It means that for one second the order does not have to be routed to other exchanges in case there is not enough liquidity at the NBBO locally. Due to this exception, some exchanges within the U.S. have started to use so called flash orders (Gomber et al. 2011). They help exchanges increase the probability that orders are executed locally and not routed to any other exchanges. A flash order is a marketable order, which can be converted into a limit order for a few milliseconds by the exchange (Gomber et al. 2011). Consider the following example. If an investor chooses to place a marketable order, which cannot be executed instantly in the exchange due to a lack of liquidity at the prevailing NBBO, the order is not immediately routed to the other marketplace. It stays in the original exchange for several milliseconds more. During this short period of time, the exchange changes the marketable order into a limit order at the prevailing NBBO and expects that some traders would consume the supplied liquidity. If the counterparty is found, the order is executed inside the exchange where the order was initially placed, even though in the beginning there were no offers that would comply with the order protection rule (Gomber et al. 2011). Since such a trading opportunity lasts just for several!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 3 It is called the Flicker price! 20!

25 milliseconds, only HFTs can react and participate in it. Thus, such trading orders are generally perceived to be unfair to regular investors. Even though the actual impact of these orders is still unclear, it is argued that they may impair the price discovery process. Furthermore, such a practice can worsen limit-order generated liquidity provision. The reason is that limit orders become less likely to be executed against orders being routed from other marketplaces that are using flash orders. Therefore, investors might have less incentive to place limit orders and, consequently, liquidity may decline (Gomber et al. 2011). As a result, the SEC proposed to ban flash orders (SEC 2009). However, no actual measures have been implemented yet. Co-location Services In June 2010, the Commodity Futures Trading Commision (CFTC) proposed the regulation which should ensure fair access to co-location services for market participants (CFTC 2010). The commission argues that the possibility to locate a trader s server to the exchange s servers as close as possible in order to reduce latency should be accessible to everyone. However, in practice such places are very expensive to rent and mostly occupied by HF traders that are willing to invest highly in it to gain the crucial competitive advantage (McGowan 2010). The CFTC proposes that uniform fees for co-location and related services should be established to assure equitable pricing to all market participants. The commission also suggests that exchanges should report information about the latency of different classes of investors and update it regularly. Circuit Breaker A circuit breaker is defined as a procedure that halts temporarily, or, under extreme circumstances, closes a market before the normal close of the trading session, if a stock experiences a severe and unanticipated price decline (SEC 2012). The thresholds to trigger a circuit breaker are set to 10%, 20% or 30% decline of stock market value within five minutes. Depending on the magnitude of a drop, trading can be halted for a time period between one hour and one day. Stub quotes Sometimes market makers could quote prices of securities that are unreasonable and far! 21!

26 away from the fair market price in order to comply with existing quotation obligations. However, in the event of liquidity shortage, such offers could be executed, even though market makers did not have any intention of that. After the Flash Crash in 2010, when such unintended and harmful executions were witnessed, the SEC banned stub quotes and started to require market makers to quote prices around the current level of NBBO (SEC 2010a). Consolidated Audit Trail Consolidated Audit Trail system s main aim is to increase regulator s ability to monitor exchanges for any abuses conducted or unusual market events that could trigger turmoil. The system is proposed to be built on one unified database which would enable regulators to access detailed information about the orders from their origination up to execution (SEC 2010b) HFT regulations in Europe Primarily due to the Flash Crash in the U.S., HFT activities came under the radar of European regulatory bodies too. Consequently, several documents and reports were issued in 2010 that were supposed to draw the guidelines for the future regulations and to highlight the points to consider before taking them into effect. CESR Technical Advice to the European Commission The Committee of European Securities Regulators (CESR) released a survey open to all market participants in 2010 regarding problems in European equity markets (CESR 2010a). Among other questions there were several relating to HFT with the purpose to better understand the attitude investors have towards it. The results were quite mixed, with market participants supplying both positive and negative opinions about the impact of HFT on liquidity, volatility, price efficiency in the markets, and possible abuse of other investors due to predatory strategies used (CESR 2010b). Based on the results, the action plan, mainly addressing the following three key points, was proposed. Firstly, The CESR expressed the necessity to research HFT further to better understand possible risks to markets arising from trading strategies their algorithms use. Secondly, the CESR proposed developing some particular guidelines that would help to control HFT activities and assure a fair and safe environment in the markets. Thirdly, the! 22!

27 commission proposed to amend some parts of the Markets in Financial Instruments Directive (MiFID), which aims to harmonize regulation for investment services around the European Economic Area (European Commission 2004). It was first published in 2004 and then amended in 2008 (European Commission 2008). Among the proposed amendments was the Article 2(1)(d) exempting investors, who do not provide any investments services and deal on their own account, from the compliance with MiFID. Since one of the key features defining HF traders is that they are proprietary traders, i.e. trade on their own account, the exemption granted by Article 2(1)(d) essentially means that these traders are treated as regular investors and, thus, face few regulatory obligations. By proposing to amend Article 2(1)(d), the CESR tries to put much more regulatory restraints on HF traders. Report on Regulation of Trading in Financial Instruments In the report, prepared by the Committee on Economic and Monetary Affairs (European Parliament 2010), the guidelines for regulations regarding trading in financial instruments were developed. What concerns HFT, it was decided that there is a need for specific HFT regulations to be implemented. Among those, the most radical one would be to require regular reviews of HFT algorithms the trading entity uses to ensure that they comply with the standards. This would compromise the essence of the HFT business, since the competitive advantage lies primarily in the way algorithms are programmed and in the speed they can operate. Furthermore, the report suggests that the practices of quote stuffing and layering, which are frequently named among strategies HFT firms use 4, should be defined as market abuse. The mechanism to precisely detect such deviations in real-time would be complicated to implement. However, the report, on a more general scale, proposes that all exchanges would be required to have technical ability to recreate order books after unusual event, if necessary, making it easier to eventually track the firms that used prohibited strategies. What concerns co-location services, the Committee s position is quite similar to that of the CFTC in the U.S. It requires practices that would guarantee fair and nondiscriminatory access for all participants to such services. Moreover, the exemption of proprietary traders from conforming to MiFID is also highlighted in the report and is described as a matter of urgency. Such high attention to the problem is specifically!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 4 For explanation and further discussion about the strategies HFTs use, please look the section 3, HFT Strategies.! 23!

28 related to the fast-growing presence of HF traders, who mostly operate as proprietary investors and, thus, are not regulated at the moment. Finally, the Committee proposes that the ESMA should investigate the recipients of market maker rebates and develop a way to make sure they comply with formal market maker obligations. Since HF traders executes plethora of trades each day, they are among largest recipients of liquidity rebates. However, they are not legally obliged to guarantee liquidity provision. Thus, at the times of market turmoil and low liquidity, HFTs can step out off the market and distort market s conditions even more. The Committee proposes that if an HFT firm on average executes sufficiently large number of trades, it has to provide continuous liquidity and in essence act as the market maker. (Gomber et al. 2011) Review of the Markets in Financial Instruments Directive (MiFID) In 2010 the European Commission revised the Market in Financial Instruments Directive (MiFID) The European Union law regulating investment services in the member states of the European Economic Area (EU plus Iceland, Liechtenstein and Norway). They concluded that the law needs to be updated due to rapid technological developments, market fragmentation and the emergence of high frequency traders (Gomber et al. 2011). The potential MiFID amendments proposed by the Commission included the following: Redefining algorithmic trading and high frequency trading All persons engaged in HFT to the extent greater than some particular limit should fulfill licensing requirements Requirements for HFTs to control risk that emerges from trading system errors Requirements for exchanges to control risk by circuit breakers, and to assure fair and equal possibilities for traders to co-locate their servers (Gomber et al. 2011) The Commission is still considering three other modifications of the MiFID: To ensure that HFTs would provide liquidity continuously Introducing a minimum time period between the submission and the cancelation of orders Introducing a maximum level of the ratio of orders to transactions (Gomber et al. 2011)! 24!

29 To conclude, lately there have been a lot of discussions and propositions on the best way to regulate HFT activities. While there were quite many regulatory changes implemented in the U.S. after the Flash Crash, Europe is still in the initial phase of deciding how to ensure sufficient legal control over HFT. There have been several plans of action proposed by European regulatory bodies on how to deal with the problem. However, to date none of them has been implemented yet, showing the complexity of the question. The point worth mentioning is that the need for HFT regulations is taken for granted without questioning whether the current regulations on financial markets are sufficient enough. This is somewhat surprising, since the area is still not thoroughly researched and, in fact, there is no globally accepted definition what actually constitutes HFT activities and strategies. However, it is clear that some actions, at least in clarifying the situation and harmonizing global perception of terminology used, have to be taken.! 25!

30 3. HFT Strategies According to the report written by the London Stock Exchange Group (2010), it is extremely complicated to identify specific HFT strategies, because all high frequency traders differ significantly in their approach to trading and they keep their algorithms strictly in secret. Nevertheless, all HFT algorithms can be generalized as being programmed to observe trading signals in financial markets, accumulate historical data on trades and quotes, and predict the short-term value of financial instruments (LSEG 2010). Tabb Group (2009) describes HFT algorithms as fully automated decision making mechanisms that employ a wide variety of different strategies, most of which can be attributed either to electronic market making or arbitrage exploitation. According to the Group, HFT firms often seek to take advantage of market liquidity imbalances and short-term mispricing of financial instruments. The common characteristics of all HFT strategies include ultra-fast trading, short holding periods, constant presence in order books and closed positions by the end of the day (Tabb Group 2009; LSEG 2010). Some of the most popular HFT strategies include automated market making, market arbitrage, liquidity rebate trading, latency arbitrage and the use of predatory algorithms (McGowan 2010; Abbink 2011; Arnuk & Saluzzi 2008; Arnuk & Saluzzi 2009; BMO Capital Markets 2009). McGowan, Arnuk & Saluzzi, and members of the BMO Quantitative Execution Desk discuss these strategies in detail and provide numerical examples illustrating how HFT firms can employ them. Automated market making According to McGowan s (2010) paper The rise of computerized high frequency trading: use and controversy, market making is generally considered to be a strategy that improves market quality and thus, is favorable to all investors. Proponents of HFT often argue that the involvement of HFT in the market making activity has provided additional liquidity in financial markets and narrowed bid-ask spreads (McGowan 2010). Consistently with other market makers, HFTs provide liquidity by placing limit orders on both sides of the order book, and offering liquidity takers to buy stock above the current market price and to sell their stock below the current market price. HFTs! 26!

31 attempt to earn the bid-ask spread automatically by using sophisticated algorithms that enable them to constantly remain in the top of both sides of the order book and be the first to provide stocks to liquidity takers (McGowan 2010). According to proponents of HFT and some academics, market making is the main and most commonly employed strategy by high frequency traders. For instance, Menkveld (2012) finds in his case study of one large HF trader that 80% of its trades were passive. The author concludes that in general HFTs can be called the modern market makers. However, as we discuss further, both passive and active trading can be applied by HFTs in an exploitative manner if combined with liquidity detection techniques. Arnuk and Saluzzi (2008) provide a theoretical explanation and a numerical example illustrating this view in their paper Toxic Equity Trading Order Flow on Wall Street. First of all, co-location and superior technology, which result in ultra-low latency, enable HFT firms to identify algorithms run by institutional investors. The main technique used by HFTs to sniff out large orders lying in the order book is called pinging and it involves ultra-fast placing and canceling of small orders (Arnuk & Saluzzi 2008). For instance, assume that the current bid-ask spread for a stock stands at $10.00-$10.06 and there is an institutional investor, whose trading algorithm is programmed to buy shares bidding $ In order to reduce market impact, the algorithm divides the limit order into a number of smaller limit orders and feeds them into the market. Assume further that the investor s trading algorithm is programmed to take liquidity and buy shares immediately if they are offered in the market for up to $10.03, however this information is not publicly available. A high frequency trader involved in market making can learn this information by pinging, which works in the following way. It places a limit order to sell 100 shares at $10.05, and, since nothing happens, cancels it immediately. Then it repeats the same procedure at $ When the HFT places an order at $10.03 and it is executed, it learns that there is an institutional algorithm willing to pay up to $10.03 for a share. Then the HFT starts front-running the institutional investor: it places a buy limit order at $10.01 in order to buy more shares before the institutional investor can (the investor s limit orders are placed at $10.00), and then sells that stock to him at $ (Arnuk & Saluzzi 2008) This is an example of how automated market makers, who have an execution speed advantage, can exploit non-hft investors. All this can happen many times in particles! 27!

32 of seconds. As a result, some believe that HFT s input in providing liquidity and reducing bid-ask spreads is outweighed by unfair exploitation of other traders (McGowan 2010). Finally, some argue that the extensive HFTs liquidity provision alone can hurt other investors. Recently, Pragma Securities issued a report suggesting that due to intensive competition in liquidity provision, it is difficult for institutional investors to execute trades by placing limit orders in the order book. As a result, they are often forced to cross the spread and take liquidity in order to sell or acquire a desired amount of stock. According to the report, the effective spread actually widens, because the overall trading costs for non-hft traders increase (Pragma Securities 2012). Market arbitrage Another HFT strategy discussed in McGowan s paper is market arbitrage. HFT firms are highly involved in profiting from different prices, rates or other conditions observed in different markets or maturities. For instance, high frequency traders can identify different prices of the same stock in different venues and exploit the discrepancies observed. Although the price differentials might exist only for a second or less, HFTs can use their speed advantage and be the first to profit from the spreads before they disappear (McGowan 2010). According to TABB Group, the total profit that a few hundred HFT firms earn in US from market arbitrage strategies during a year makes up around $21 billion (Tabb Group 2009). There are two types of arbitrage opportunities that HFTs pursue. Firstly, market neutral arbitrage involves holding a long position in an asset, which is perceived undervalued, and a short position in a different but closely related asset, which is considered overvalued. In this way, one offsets overall market movements. An HFT liquidates the positions when the prices of the assets normalize and realizes the profit (Gomber et al. 2011). Secondly, cross asset and cross market arbitrage strategies take advantage of observed mispricings among assets and markets. For instance, a trader can find an opportunity to purchase stock cheap in one exchange and quickly sell it expensive in another trading venue. Market fragmentation has contributed to the growth of cross market arbitrage opportunities as discussed in the Background Information section.! 28!

33 Similarly, cross asset arbitrageurs can profit from mispricings between a derivative or ETF (Exchange Traded Fund) and a relevant underlying asset. (Gomber et al. 2011) Generally, arbitrage has been perceived beneficial to financial markets, since it eliminates mispricings and various inconsistencies among securities in different venues, and thus contributes to the price discovery process (McGowan 2010; LSEG 2010; Hendershott & Riodan 2011). However, today arbitrage strategies executed by HFT firms are gaining a negative perception. True high frequency traders are often considered to ignore economic theories and characteristics of the financial assets they trade. Instead, they are blamed for focusing on the structure of markets and creating mispricing themselves, which they could later exploit at the expense of retail and institutional investors (Papagiannis 2010). One of the arbitrage strategies that has been most scrutinized is latency arbitrage. Latency arbitrage Arnuk and Saluzzi (2009) highlight the unfair access to financial markets that high frequency traders have. The source of latency arbitrage is high-speed hardware, software and co-location of HFT firms servers next to exchanges servers, which reduce latency and increase the speed of order execution. Theoretically, any institution can rent out space next to the market centers servers. Practically, however, HFT firms are able to exploit this advantage most effectively (Arnuk & Saluzzi 2009). The speed advantage created by co-location, technology and access to raw data feeds from exchanges enable HFT firms to construct their own order books before they are publicly available from the Security Information Processor (SIP) quote (Patterson 2010; Arnuk & Saluzzi 2009). Effectively it means that high frequency traders can see the order books a few microseconds into the future (Mehta 2009). Arnuk and Saluzzi (2009) provide a numerical example of the predatory latency arbitrage practice. According to the authors, orders placed by institutional algorithms are often driven by volume weighted average price (VWAP) formulas. As a result, those orders adjust automatically if the bid-ask spread shifts upwards or downwards (Arnuk & Saluzzi 2009). Assume that an HFT firm pings stock prices and identifies an institutional investor s order to buy shares at the price related to volume weighted average price. The current bid-ask spread for stock A is $ $ Assume! 29!

34 further that there is an order that is about to change the bid-ask spread to $ $10.04, and the HFT firm is the first to see that. The arbitrager immediately purchases all available stock at $ When the spread moves up, as predicted by the HFT, the arbitrager places sell orders at $10.03 and $ The institutional investor is forced to purchase available stock at these prices, since the VWAP moves up and the institutional algorithm follows VWAP formulas. As a result, the HFT firm earns a risk-free arbitrage profit, which amounts to 1-2 cents per share. However this is done at the expense of the institutional investor. (Arnuk & Saluzzi 2009) Arnuk and Saluzzi (2009) estimate that the predatory latency arbitrage technique earns $6-$12 million a day in US stock exchanges. However, the main concern is the fairness of the strategy. By facilitating co-location stock exchanges do not seem to provide equal trading conditions to all market participants. (Arnuk & Saluzzi 2009) John Abbink (2011) reports a slightly different way of exploiting latency arbitrage. According to him, large institutional orders are usually placed on multiple venues in order to reduce market impact. However, due to different latency, some venues receive an order earlier than others. Even though the difference might be only a fraction of a second, this time is enough for high frequency traders to anticipate the order flow and exploit it. Abbink (2011) suggests that placing orders across venues in particular sequences would result in simultaneous appearance of orders in all market centers. This could partly eliminate latency arbitrage opportunities (Abbink 2011). However, market centers are not expected to be in favor of eliminating latency arbitrage, because co-location seems to be a highly profitable business. The demand for co-location far exceeds the supply. As a result, the NYSE Euronext has invested around $1 billion in building huge liquidity centers in New Jersey, USA, and Basildon, UK, and plans to build more in the future. NYSE Euronext expects to collect $1 billion in revenues per year from these co-location facilities (Rath 2010). Liquidity rebate trading Liquidity rebate trading is another trading strategy discussed by McGowan (2010). This strategy enables HFT firms to make money by providing liquidity to markets, and collecting liquidity rebates paid by exchanges (Interactive brokers 2012). Market centers are willing to pay rebates, which amount to a fraction of a penny per share, as! 30!

35 they attract trading volume (Bunge 2010). Effectively, liquidity rebates are paid by liquidity takers through extra charges imposed by trading venues (McGowan 2010; Gomber et al. 2011). Since every liquidity provider is entitled to liquidity rebates, it is hard to imagine how they might induce exploitation of some market participants. However, low-latency and co-location enable HFT firms to employ techniques that take advantage of these rebates to the detriment of other traders. Arnuk and Saluzzi (2008) give one more numerical example that explains how it can be implemented in practice. Assume that the institutional investor has programed its algorithm to purchase shares in the price range of $10.00 to $ The algorithm attempts to place buy limit orders at the top of the order book but it cannot exceed $ Also, it is set to consume liquidity and purchase shares immediately if they are offered at $10.03 or below. Suppose that an HFT recognizes the institutional algorithm by pinging and that the current best bid price is at $ Then the liquidity rebate trader front runs the institutional by placing a buy limit order at a marginally higher price of $10.01 in order to absorb all liquidity demand that would have sold shares for $ The HFT firm receives a liquidity rebate of around ¼ of a penny for every stock bought due to providing liquidity. Then the firm turns around and places a limit order to sell the shares at $10.01, which are likely to be bought by the institutional investor s algorithm (recall that it is programmed to purchase shares offered at prices below $10.03). After the order is executed, the HFT firm collects another liquidity rebate for providing liquidity. Although the HFT firm does not profit from the increase in the stock price, it earns ½ of a penny for every stock it purchases and sells. Meanwhile, the institutional investor is forced to pay $0.01 per share more than it could have paid in the absence of the liquidity rebate trader. (Arnuk & Saluzzi 2008) This strategy is also criticized in the report The impact of high frequency trading on the Canadian market written by BMO Capital Markets (2009). It can be seen from the example above that this strategy intentionally locks the market, meaning that for a fraction of a second the bid and ask prices become equal ($10.01). While some believe that locked markets are perfect markets, the authors of the report argue that it creates unfair overcompensation for passive liquidity providers and hurts other liquidity seeking investors (BMO Capital Markets 2009).! 31!

36 Predatory algorithms and short-term momentum The report of BMO Capital Markets (2009) also discusses HFT strategies that involve the use of predatory algorithms. These predatory programs are designed to spot algorithms programmed by institutional investors (e.g. by pinging ) and predict their actions. According to the paper, some institutional algorithms are relatively easy to identify. For example, they may be driven by volume weighted average price (VWAP) formulas, or even directly pegged to a bid-ask spread. High frequency traders are actively searching for these types of algorithms and once they are spotted, HFT firms are able to take advantage of them (BMO Capital Markets 2009). According to Arnuk and Saluzzi (2008), in the U.S. in 2008 more than 50% of orders executed by institutional algorithms were pegged to the National Best Bid or Offer (NBBO). This means that institutional investors automatically adjust their limit orders when the stock price moves up or down. Predatory algorithms used by HFT firms are designed to artificially increase or decrease stock prices, which would force institutional investors to adjust their orders respectively, and trade at less favorable prices than they could trade in the absence of HFT (Arnuk & Saluzzi 2008). For example, assume that there is a buy limit order placed by an institutional investor s algorithm that is pegged to the NBBO. Assume further that currently the order stands at the best bid price of $10.00 and that the algorithm would adjust the order to remain at the top of the order book as long as the price does not exceed $ A predatory algorithm can relatively easily identify the strategy of the institutional algorithm using the techniques similar to automated market maker and liquidity rebate trader. When the institutional algorithm is spotted, an HFT firm places a buy limit order at a marginally higher price at $ The institutional investor automatically adjusts its price to $ When the predatory algorithm increases the bid price up to the institutional investor s limit of $10.10, it sells the stock short to the investor and expects the price would eventually return to its starting point. When it does, the HFT firm closes the short position and realizes the profit (Arnuk & Saluzzi 2008). Since all HFT firms observe the same signals from markets, their collective actions can artificially generate profit opportunities. This predatory strategy involves buying or selling small amounts of stock by many HFT algorithms simultaneously, which drives! 32!

37 the stock price up or down. As many institutional orders are related to the VWAP or the NBBO, they follow the trend and help the HFT to push the price. The HFT firms profit by acquiring options on particular assets and pushing them into the money by a greater amount than it costs to move the market (Arnuk & Saluzzi 2008). Alternatively, instead of pushing the prices by buying (or selling), a fast trader can simply ride the wave by purchasing (or selling) stock in the initiative phase of the price trend and profiting from a short-term momentum. Momentum traders also harm institutional investors as their trading drives prices away from fundamental values, and thereby decrease the price efficiency (Sornette & Von der Becke 2011). Momentum based trading has been implemented long time before the first HF traders have entered markets (Gomber et al. 2011). Other predatory strategies Quote Matching. Large investors and traders risk to be exploited by fast traders who apply the quote-matching strategy. According to Angel et al. (2010), this strategy inclines transaction costs for large market participants. Consider the following example provided by the authors: suppose that an institutional investor submits a buy limit order into the market at $ A fast HF trader observes it and attempts to purchase ahead of it, by bidding $10.01 or even $10.00 at a different exchange. If his order is executed, he obtains a long position in the stock. Later, if the stock price rises, the fast trader profits from the price increase. If the HF trader observes a price decline, it quickly turns around and sells to the institutional, who is bidding $ The large trader can update his quotes slower than the HFT who has a speed advantage. As a result, the fast trader profits from the price increase, while its loses are limited in a case of the price decline. Meanwhile, the institutional investor may experience a relative loss if it fails to purchase at $10.00 before the price rises (because the HFT might have absorbed the whole stock supply by positioning in front of the institutional). Moreover, he may buy at $10.00 if the price falls (because HFT can react to changes in a market faster and force the institutional to buy at its old quoted price of $10.00). Effectively, the situation implies that the fast trader possess a free of charge call option-like instrument, which gives return if prices rise and limits loss if prices fall. This call, on the other hand, is written for free by the institutional (Angel et al. 2010).! 33!

38 According to Angel et al. (2010) many buy-side traders support the creation of innovative trading systems that would help to solve quote-matching problem. Quote Stuffing. Quote stuffing is a practice of placing a large number of orders and canceling them instantaneously. This is done by high frequency traders in an attempt to flood the market with quotes, which would have to be processed by competitors. As a result, the competing high frequency traders would become slower in decision-making and lose the competitive edge (Egginton et al. 2012). Slower market participants and public authorities have heavily scrutinized these actions, as they create a false sense of the real demand and supply for a stock. Furthermore, this practice is widely blamed to have caused the Flash Crash on May 6, 2010 (Nanex 2010). Egginton et al. (2012) find that intense quote stuffing is comparable to the provision of fake liquidity. Furthermore, this practice is associated with worsened market quality, namely, decreased liquidity, higher trading costs, and increased short-term volatility. These findings contradict the arguments of HFT proponents, stating that HFT activity increases liquidity and enhances other market conditions (Egginton et al. 2012). According to Nanex (2011), in December 2011, NYSE proposed rules that would prohibit quote stuffing, which is an important step towards the end of this type of market abuse. Spoofing and Layering. According to AFM (2010), spoofing is defined as submitting a limit order (e.g. to sell) to the order book, without any intention of getting it executed. Instead, its size and ranking in the order book is meant to alter (in this example, lower) the existing bid-ask spread to a different level (AFM 2010). Layering is defined as an intensive spoofing, when a trader submits a significant number of limit orders (e.g. to sell) with different prices on one side of the order book. These orders are not meant to be executed but rather to create the impression of increased trading (in this case selling) pressure. The real intention of the trader is to trade on the opposite side of the book (in this example, to buy stock). For instance, an increase in selling pressure can lead to a cancelation of bids by market participants, which brings the best bid price down. In this case a trader who initiated the fake selling pressure can purchase stock cheaper and then cancel the initially placed sell limit orders (AFM 2010). There is a risk that the sell limit orders would be executed if the demand for the stock suddenly increases. However, high frequency traders can react instantaneously and cancel the orders, if the incline in demand is spotted. Similarly to quote stuffing, layering is a practice that fakes the! 34!

39 increase in liquidity supply and abuse markets. High frequency traders are considered to be significantly involved in these strategies (Gomber et al. 2011). All in all, high frequency traders can pursue a wide variety of different trading strategies, and some evidence suggests that their activity generally tends to improve the market quality. In particular automated market making is considered to provide liquidity and narrow bid-ask spreads, while market arbitrage eliminates price discrepancies among assets and different markets and contributes to the price discovery (Gomber et al. 2011). On the other hand, even the automated market making strategy could be possibly employed in a predatory manner if liquidity detection techniques are used by HFTs (Arnuk & Saluzzi 2008). What is more, tight competition among trading venues initiated a race: who will provide lower latency for its customers. As a result, in order to attract adequate liquidity, market centers introduced liquidity rebates and enabled colocation of HFT servers, which allows some market participants to observe market signals faster than others (SEC 2010c). These practices gave birth to some questionable HFT strategies discussed in this section. However, it is extremely difficult to quantify how detrimental they are to other investors (Papagiannis 2010). The second part of the following Litareture Review, named Empirical research on HFT strategies, discusses evidence from relevant academic studies.! 35!

40 4. Literature Review This section provides a thorough review of academic literature on high frequency trading. The section is divided into two major parts, namely HFT effects on market quality and Empirical research on HFT strategies. Moreover, the former of these parts is divided into two parts based on the articles presenting theoretical conjectures and empirical evidence. The Empirical research on HFT strategies sub-section is divided into Initiation of price trends and Front running HFT effects on market quality Theoretical models Due to new capital market characteristics that mainly arose because of technological advancement and emergence of electronic markets, Hoffman (2010) extended a classic Foucault s model of order placement decisions in a dynamic limit order market. Hoffman s trading model accommodates for the properties of a modern trading environment where investors, mainly due to growing presence of automated trading, differ greatly in speed with which they can react to the announcements of investmentrelated news and adjust their positions accordingly. Such an advantage allows algorithmic and high frequency traders to access the best quotes available for the trade, while institutional investors cannot do so. Consequently, this leads to a lower expected profit for institutional investors and their overall expected utility is lower than if they were competing with investors of identical opportunities. In essence, Hoffman argues that even though the use of HFT algorithms could have some positive externalities, such as improved market liquidity, this comes at the cost to be paid by institutional investors. As far as the growing speed in the stock markets in general is concerned, Pagnotta & Philippon (2012) have developed a theoretical model to investigate the impact of such a trend on the overall social welfare. They propose that the growing speed in the capital markets for a given number of exchanges, results in higher social welfare, and that investments in speed are generally desirable. Thus, as the most important feature of the HFT system is the speed with which it can react to the market events, it can be said that HFT increases social welfare. The presence of HFT in capital markets and its impact on the overall investors welfare! 36!

41 was also analyzed in theoretical model developed by Jovanovic & Menkveld (2011). However, the authors focused only on HFTs acting as middlemen in a limit-order market and presumably aiding the market by eliminating adverse selection costs between early and late investors. The costs are minimized due to HFTs ability to quickly react to new information and to update the quotes accordingly in a matter of seconds. Therefore, an early investor could pass off his security to a machinemiddleman, which quickly observes and updates the offer price. In such a way the investor can avoid being picked of by a late investor who might have an informational advantage. Consequently, the adverse selection costs for rational early investors are minimized and the likelihood of executing the trade is improved, leading to a higher social welfare. However, the model proposes that the HFTs presence has mixed consequences. In case HFTs have the informational advantage against late investors, the overall social welfare reduces. However, if late investors had additional information that guided their investment decisions and created an informational disadvantage to early investors, HFTs would eliminate adverse selection costs and thus increase the overall social welfare. (Jovanovic & Menkveld 2011) Quite contrary, Foucault et al. (2011) developed a theoretical model showing that HFTs actually create informational asymmetries and increase adverse selection costs. The authors argue that by being computationally more efficient than other investors, HFTs exploit the advantage that enables them to constantly outperform slower market participants. As a result, HFTs secure higher gains to themselves and at the same time increase adverse selection costs. To avoid informational asymmetries, other investors have to either leave the market or invest in costly technologies similar to ones used by HFTs. This induces investments in technology exceeding the utilitarian welfare maximizing level. It is also argued that in a rare and non-anticipated market event such as the Flash Crash of May 6, 2010, programmed algorithms are slower in adjusting than human investors. Therefore algorithms can potentially trigger further excessive price changes and harm the market. (Foucault et al. 2011) With respect to volatility in the stock markets, Cartea and Penalva (2011) developed a theoretical model showing that higher activity of HF traders increases volatility of stock prices and trading volume. However, they point out that the dispersion of prices arises not due to a higher amount of shares traded, but because of strategies HFTs apply. On the other hand, the authors show that high frequency trading increases liquidity in! 37!

42 markets and makes trading cheaper. Cartea and Penalva (2011) also investigates the factors influencing professional investors to implement HFT strategies themselves. The model developed by the authors shows that new entrants would consider implementing HFT technologies as long as their expected skills as HF traders are sufficient enough to cover the initial investments required. Otherwise, professional investors with low expected skills for successful HFT implementation will continue their traditional role and the equilibrium will be reached (Cartea & Penalva 2011) Empirical studies The recent evolution of the usage of HFT in capital markets worldwide induced quite a lot of empirical studies on the subject. Bearing in mind that availability and accessibility of the specific data needed for the research concerning HFT is limited, it is sensible to categorize papers according to the type of data used in each of the studies and then discuss the main insights and results found about HFTs impact on market quality. The ideal dataset for the exhaustive analysis should contain intraday information both from trade and order books at the millisecond level and an identifier of each trader, specifically pointing out the ones trading on high frequency basis. Such data would allow the researcher to observe each trader s strategy across stocks and over time, deduce the ones that trade on high frequency basis and then evaluate the impact on a market as a whole. Trade and quote data. HFTs identified To date only one research, carried out by Kirilenko et al. (2011), obtained such a dataset. However, it was limited to E-mini S&P 500 stock index of futures market around the period of the Flash Crash on May 6, Thus, it is one of the most cited papers on the subject. Kirilenko et al. (2011) analyze the behavior of various types of investors, including HFTs, during the period of extreme volatility in the U.S. financial markets. Particularly, the authors examined a change in investing behavior throughout the Flash Crash, a short period of 30 minutes on May 6 in 2010 during which stock market indices, stock-index futures, and exchange traded funds suffered a decline of unanticipated magnitude and, shortly after, almost equally-sized rebounded. Kirilenko et al. (2011) proposes that on the macro level the turmoil was caused by concerns related to the European sovereign! 38!

43 debt crisis. As the S&P volatility index increased and yields on long-term treasury bonds rose, fundamental investors started to sell their inventories. This highly increased the price pressure, as there were not enough investors willing to buy the assets. Initial sell orders were bought by both intermediaries and HFTs, and thus, the necessary liquidity was provided. However, minutes later HFTs started to liquidate their long positions and in fact competed for the liquidity with other investors. As there were not enough potential fundamental buyers, HFTs started to trade with each other creating a hot-potato effect and further depressing prices of assets. Only after the automatically activated short trading pause in the market, fundamental buyers started to aggressively buy depreciated stocks and the prices recovered rapidly. On the basis of empirical findings, the authors infer that HFTs contributed to the extensive volatility during the Flash Crash, mainly due to competition for liquidity in order to rebalance their positions. Hence, in events when price imbalances arise in the markets, HFTs are stubborn to adjust to the situation and continue their preprogrammed trading routines contributing to the even higher pressure on prices. Trade and quote data. Proxies used to identify HFT activity The other approach widely used in empirical research is to use standard intraday trade and quote data and develop fairly reasonable proxies to identify HFT activities. On the one hand, while using such a method it is not possible to precisely track HFT activity and the quality of each approximation used is debatable. On the other hand, the benefit of such an approach is the possibility to develop quite general models and make broad inferences from results. Hendershott et al. (2010) use a similar approach to investigate whether algorithmic trading improved liquidity in the New York Stock Exchange. This research is important because it was the first empirical study on the algorithmic trading s impact on market quality. Therefore, the conclusions the authors arrived at could be compared with the results deduced by examining purely high frequency activity. Hendershott et al. (2010) use a normalized measure of the NYSE electronic message traffic, which includes electronic order submissions, cancellations and trade reports, as a proxy for algorithmic trading. Since the measure is normalized by the overall trading volume, the variations in it are mainly due to deviations in the amount of limit order submissions and! 39!

44 cancelations. Thus, the measure mainly indicates the changes in the algorithmic liquidity supply. The data used is from the period covering during which, in 2003, the NYSE became a fully automated stock exchange. The results suggest that algorithmic trading improved the liquidity of stock markets. Furthermore, it appears that adverse selection costs shrank and price efficiency increased during the studied period. Finally, the costs of trading appeared to be lower. Hence, it seems that the implementation of automatic algorithmic trading, at least in the New York stock market, had numerous positive consequences (Hendershott et al. 2010). Hasbrouck and Saar (2011) investigate the low-latency trading activity and its impact on market quality measures. The dataset used contains order-level U.S. NASDAQ data, and the authors developed proxies to identify the low-latency trading activity using information on orders submissions, cancellations and executions. Hasbrouck et al. (2011) defined latency as the sum of three components: the time it takes for relevant information to reach the trader, the time it takes to process that information and the time it takes to generate and implement the response accordingly. Therefore, low-latency traders are basically investors trading on a high frequency basis. By examining the electronic messages in the millisecond environment, authors differentiate between the algorithms that operate according to a pre-defined schedule and the ones responding to the events affecting a market instantly. That is, the division between agency algorithmic traders and proprietary HFTs is achieved by using the periodicity assumption. The empirical findings suggest that low-latency activity improves market quality, more specifically short-term volatility is observed to be lower and market liquidity higher even in the times of greater economic uncertainty. Hasbrouck and Saar (2011) suggest that higher liquidity arises mostly because of algorithms trading with each other, while conventional liquidity measures are not worsened by their activity. Such findings somewhat contradict the results achieved by Kirilenko et al. (2011), however one has to take into account the different nature of the data used and, most importantly, the historical periods investigated. Even though Hasbrouck & Saar (2011) use the data from the periods of both normal and higher volatility in the market to make inferences about the HFTs impact under different market conditions, Kirilenko et al. (2011) focuses just on one event when unusually excessive volatility was witnessed. More results contradicting the positive impact of HFT on the market quality specifications were obtained by Zhang (2010). In his work the author concludes that! 40!

45 stock price volatility is positively related to the HFTs activity after controlling for other volatility drivers. Consistent with Kirilenko et al. (2011), he states that correlation is stronger within the periods of higher overall market uncertainty and more apparent when investigating the stocks of large companies. Furthermore, the results suggest that price discovery is less efficient when HFT is prevalent, mainly because stock prices react excessively to new information in the market. Such an effect on stock prices could be partly caused by feedback loops that are exaggerated by high frequency algorithms. Feedback loops, as defined in the report The future of computer trading in financial markets by Cliff (2011), are the small changes in a share price looping back on themselves and triggering even higher change on stock price and in such a way amplifying volatility. In particular, news feedback loop caused by HFT algorithms with news listener component implemented in them, could be the major factor causing the artificial overreaction of stock prices to market news. Some of the more sophisticated algorithms are pre-programmed to scan for particular headlines and tags in the news and act immediately if a pre-defined threshold is met. Thus, even if the small number of algorithms adjusts their positions accordingly to the new information in the market, their activity is observed by other algorithms, which are likely to adjust their own positions too. In this way the news feedback loop is created, which causes the stock prices to deviate from their fundamental value, and thus, amplify the overall volatility in the market. Even though the effects of feedback loops are most apparent in the times of high overall volatility in the market, it is argued that such effects, when HFT activity is prevalent, exist on a micro level even under the normal conditions in the market. Thus, HFT reduces market quality both by increasing volatility and distorting price discovery process (Cliff 2011). Trade and quote data. HFT partly identified The other type of data used by researchers contains information about the type of each trader. Particularly, some of HFT firms are identified by the stock exchange. However, the main drawback of working with such samples is inability to study all firms practicing high frequency strategies, rather just a part of them. Thus, such samples do not necessarily represent the actual market environment well and could be biased. In his exhaustive study of the HFT impact on market quality, Brogaard (2010) investigates the relationship between the strategies utilized by HF traders and the impact! 41!

46 they have on the main market quality characteristics such as liquidity, price efficiency and volatility. The author uses a dataset that includes information on trades and quotes on a group of 120 stocks, which are traded in the Nasdaq stock exchange. The data includes identifiers of HF traders provided by the stock exchange. The main characteristics used by the stock exchange to distinguish between ordinary and HF traders were the mix of widely accepted properties describing HFTs. These were the following. Firstly, the firm engages in proprietary trading using sophisticated trading tools such as complicated analytical software. Secondly, the firm co-locates its servers near the stock exchange to reduce latency. Thirdly, the firm frequently switches between long and short net positions throughout the day contradicting the strategy used by the ordinary investors. Finally, HFT firms normally have lower trades per orders ratio, the property that arises due to pinging strategies such firms might use. Main limitation of the dataset is that divisions of larger financial firms, such as Goldman Sachs or J.P Morgan, that likely uses HFT strategies were not included in the analysis even though they could have quite large share in total HFT activity (Brogaard 2010). The results of Brogaard s (2010) research state that the overall HFT s impact on the market is positive. The HFT firms improve the price discovery process by incorporating more useful information into the prices than non-hft investors. What concerns liquidity, Broogaard (2010) finds that on average the HFTs provides as much liquidity as they demand. Therefore, in general they do not affect this property of the market. Furthermore, the author concludes that these companies make quite large profits without affecting the overall market volatility in any negative way. To conclude, there are some discrepancies between the theoretical predictions arrived by academics and the results observed from empirical studies. Theoretical models, generally, highlight the negative HFT s impact on market quality and social welfare. Meanwhile, many empirical papers provide statistical inferences that HFT actually increases liquidity in the markets and improves price discovery process without negatively affecting volatility. However, even among empirical studies inconsistencies can be found, as some papers find negative effects of HFT on volatility, liquidity and overall welfare in the capital markets. Thus, further studies are required in order to get clearer insights.! 42!

47 4.2. Empirical research on HFT strategies Even though high frequency traders have already been active for a number of years, there is a lack of information available on how exactly HFT strategies are implemented and whether they help to gain unfair advantage against other market participants. Due to intense competition, most of the strategies that HFT firms employ are kept in secret. Moreover, only a few mathematicians and engineers understand how to create and operate HFT programs (McGowan 2010). Because of this opaqueness, governments have to rely on empirical studies while making decisions on policies concerning HFT (McGowan 2010). This section reviews academic literature that empirically investigates the feasibility of different HFT strategies Initiation of price trends Jarrow and Protter (2011) challenge the majority of previous empirical evidence, which find positive impact on overall market quality, and state that HFT may not decrease volatility and increase price efficiency. In fact, in their paper A Dysfunctional Role of High Frequency Trading in Electronic Markets they construct a no-arbitrage model of the price process that shows HFT s proneness to increase volatility, create mispricings and exploit them to the disadvantage of other traders. According to the authors, high frequency traders observe common signals in financial markets and create trends in prices through collective but independent actions. As a result, the fastest traders, who are ahead of price trends, profit from the slowest ones, who just follow the trends. The speed differential is the source of inequity, which makes HFT strategies optional processes, instead of predictable processes (Jarrow & Protter 2011). This could be compared to the idea of insider trading laws, which state that it is illegal to trade on the insider information until it is released to the market. In this case, the insider information is based on the order flow process, rather than fundamentals (Jarrow & Protter 2011). By using mathematical finance tools, Jarrow and Protter (2011) explicitly prove their key theorem: (1) There are no arbitrage opportunities for the ordinary traders; (2) High frequency traders earn abnormal trading profits; (3) If HFT strategies become predictable processes (the speed advantage is removed), then their abnormal trading profits are removed (Jarrow & Protter 2011, pp 11.).! 43!

48 These findings are consistent with Arnuk and Saluzzi (2008) statements that collective actions of predatory HFT algorithms drive asset prices and create abnormal profit opportunities to high frequency traders at the expense of other investors. Similarly to Jarrow and Protter, Cadogan (2012) in his paper Trading Rules over Fundamentals: a Stock Price Formula for High Frequency Trading, Bubbles and Crashes derives a model of the stock price process, which captures empirical regularities of high frequency trading (Cadogan 2012). The author uses mathematical finance techniques comparable to Jarrow and Protter s (2011), in order to derive a closed stock price formula. The results of Cadogan s work imply that high frequency traders observe particular volatility signals, which enable them to predict either an incline or a decline in an asset price and trade accordingly. Furthermore, the author states that volatility can be driven by collective actions of high frequency market makers, who try to create bid-ask bounce through quote stuffing and profit on it (Cadogan 2012). Another recent academic work written by Cartea and Penalva (2011) derives a different theoretical model from the two discussed above. They use an equilibrium model introduced by Grossman and Miller (1988), which describes the interactions between institutional investors and professional traders. The latter provide liquidity to the former and are perceived as intermediaries among institutional investors over time. Cartea and Penalva (2011) expand the model with high frequency traders, who are pictured as instantaneous intermediaries between institutional investors and professional traders, since their speed enables them to position in the top of limit order books. Their model implies that the intermediate role of HTFs generate economic profits at the expense of investors and professional traders. Specifically, findings suggest that HFT strategies generate price dispersion by creating and exploiting opportunities to purchase stock from investors at a lower price and to sell to professional traders at a higher price. Although Cartea and Penalva (2011) do not specify the HFT strategies in their work, they give a numerical example of how HFTs can extract rents from other market participants. Suppose an institutional investor decides to sell a few blocks of shares, and, thus, needs liquidity. The current best bid price is at $ High frequency traders can identify the increased selling pressure, caused by the institutional investor, and cancel all their buy! 44!

49 offers. Additionally, they post sell offers in order to increase the selling pressure further and make other market participants cancel their buy offers. When the market around $10.50 dries out, HFTs post buy offers at the new best bid price of $10.47 and use speed advantage in order to remain at the top of the order book. HFTs buy shares from the institutional investor at the new best bid price and, when the selling pressure eases, post sell offers at higher prices ($10.49 and $10.48). Even though other professional traders are sorry to see the price rebound, those who were willing to buy at $10.50 in the beginning are still willing to buy at $ As a result HFT profits from purchasing shares at $10.47 and selling at $10.49 and $ The institutional investor receives a couple of cents per share less, while professional traders are able to buy shares marginally cheaper than in the absence of HFT. However, HFTs are the biggest winners in the situation, because the speed advantage allows them to cancel and repost at different levels of the order book. As a result, collectively they can influence the selling pressure and position themselves at the top of the order book to be the first to benefit from the temporary decrease in price (Cartea & Penalva 2011). However some research fails to find evidence that collective actions by high frequency traders can artificially create profit opportunities. Aldridge (2011) in her paper Can High Frequency Traders Game Futures? uses tick data from the period and examines whether the Eurex Eurobund futures market possesses particular conditions that would enable High Frequency Traders to create and exploit mispricings at the expense of other investors. Particularly, the author tests the feasibility of the pumpand-dump arbitrage, which involves the artificial creation of price trends. In order for the arbitrage strategy to be feasible, buys and sells should have an asymmetric impact on the asset prices in the market (Gatheral 2010). The author s test provides implications towards the applicability of predatory algorithms, discussed by Arnuk and Saluzzi (2008), which collectively drive prices and profit from the created mispricing (Aldrige 2011). Aldrige (2011) does not find evidence that HFTs could employ predatory strategies, which would hurt other traders. In fact, the results suggest that prices in Eurobund futures market react symmetrically. It means that traders cannot lock in profits by artificially driving the prices upwards (or downwards), as the prices move back adequately while they attempt to close positions (Aldrige 2011).! 45!

50 Front running Instead of focusing on price trend creation by high frequency traders, some researchers investigate the ability of high frequency traders to individually front run institutional investors and retail traders. Front running includes those strategies that use liquidity detection techniques in order to discern the trading patterns of other market participants and trade ahead of them. The strategies discussed in the previous section ( HFT strategies ), namely (1) automated market making, (2) latency arbitrage, (3) liquidity rebate trading and (4) individual actions of some predatory algorithms, share a common feature: learning about large orders coming from institutional investors (e.g. through pinging) and trading ahead of them accordingly. Thus, these strategies fall under the definition of front running strategies (Brogaard 2010). Garvey and Wu (2009) examine the effects of the High Frequency Traders co-location within stock exchanges on other market participants. In the paper Speed, distance, and electronic trading: New evidence on why location matters they study orders submitted by 2000 electronic stock traders, who are widely dispersed in the US. Their findings reveal that traders who co-locate their servers in the exchanges are in fact involved in high frequency trading strategies. Moreover, co-location creates a significant execution speed advantage, which is translated into economic profits by HFTs. According to the authors, traders that are located closest to the market centers execute their orders on average 2.8% faster than others. This results in HFTs ability to act before price changes occur publicly and exploit latency arbitrage (Garvey & Wu 2009). In the meantime, traders who are located further away from the exchanges incur additional costs created by the HFTs speed advantage. Firstly, slower traders cannot trade based on new information in the market as faster market forces ensure the information is incorporated into market prices instantaneously. Secondly, distant traders can be exploited, as they cannot react to information changes fast enough. Thirdly, in a case of information asymmetries, bid-ask spreads increase as limit order traders attempt to protect themselves from better informed HFTs. This partly explains why HFTs usually position themselves in the top of limit order books. These empirical findings are in line with Arnuk and Saluzzi s (2009) conjectures concerning latency arbitrage and front running slow institutional investors (Garvey & Wu 2009).! 46!

51 McInish and Upson (2012) in their recent study Strategic Liquidity Supply in a Market with Fast and Slow Traders test whether current regulations in US create opportunities for high frequency traders to capture arbitrage profits at the expense of slow traders. Order Protection Rule (Rule 611), released by the Securities and Exchange Commision (SEC), states that all trades across different markets have to be executed at a National Best Bid and Offer (NBBO). However, in computer driven markets prices within one exchange can update faster than they can be compared across exchanges to evaluate an NBBO. As a result there is an exception to the rule, which states that a trade can be executed at the least aggressive quote present over the previous 1 second at the exchange where the trade is being executed 5 (McInish & Upson 2012). As high frequency traders observe prices across exchanges in real time, they can take advantage of this exception and strategically supply liquidity at prices inferior to the NBBO in order to exploit mispricing across the markets (McInish & Upson 2012). For example, assume the current NBBO Ask price stands at $10.05 and updates to $ Assume further that a fast trader has placed a sell limit order in one of the exchanges at $ The trader sees the mispricing, however he can choose not to update his quote to $10.03 for 1 second. A slow trader, who cannot see the new NBBO yet, can submit a marketable buy order to the exchange at $10.05, and if it is done within one second, which is allowed by the Flicker Quote Exception, the trade is executed. The fast trader can quickly buy back stock in another exchange at $10.03 and lock in an arbitrage profit of $0.02 (McInish & Upson 2012). The authors of the paper find empirical evidence that this strategy is extensively employed and highly profitable. They use Daily Trade and Quote Data (DTAQ) from 2008 of 100 largest NYSE listed stocks to estimate that this high frequency trading strategy in US earns approximately $233 million per year to the detriment of institutional investors and retail traders. It constitutes more than 8% of the total annual profits earned by HFTs (McInish & Upson 2012). As the advantage for HFTs originates from being the first to observe mispricings in different markets, this strategy is similar to the Latency Arbitrage strategy discussed by Abbink (2011).!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 5 The least aggressive quote present over the previous 1 second in an exchange is called a Flicker price! 47!

52 An empirical study conducted by QSG LLC (2009) suggests that trading algorithms pegged to VWAP incur higher costs due to high frequency trading activities. The Group employed its innovative tick-based platform called T-Cost Pro, which measures transaction costs. According to the authors of the study, HFT strategies anticipate the order flow from other market participants and, thus, game passive order techniques employed by institutional investors and other traders (QSG LLC 2009). VWAP algorithms are especially vulnerable, as they are easy to predict. (QSG LLC 2009) These finding are consistent with Arnuk and Saluzzi s (2008) statements concerning the front running using predatory algorithms. Brogaard (2010) in his paper High frequency trading and its impact on market quality investigates whether HFTs tend to engage in the front running strategies in general. HFTs are claimed to be able to detect large orders to be executed through pinging. For example, if a large order placed by a non-hft firm is detected, an HFT firm can place a corresponding limit order offering a marginally better price to market participants, and then resell the stock to the non-hft firm at an increased price. This practice allows HFT firms to profit by receiving liquidity rebates while hurting non- HFT traders. Brogaard performs a test, which is based on a testable implication that HFTs, who are involved in front running, should become more active right before large trades are initiated by institutional investors. Consequently, the author compares the extent of HFT activity right before the execution of trades of different sizes initiated by non-hft firms. He finds that larger trades initiated by non-hft firms tend to be preceded by fewer HFT initiated trades, thus he does not find empirical evidence that the front running strategies are employed by high frequency traders in Nasdaq stock exchange (Brogaard 2010). In conclusion, academic literature empirically investigating the impact of HFT strategies on other market participants is scarce. Furthermore, the existing research fails to provide common conclusions regarding the applicability of these strategies and the extent of potential harm done to other traders. While different theoretical models conjecture towards exploitation through price trend creation, empirical evidence suggests differently (Aldrige 2011). While some academics find evidence of market abuse through latency arbitrage, Brogaard (2010) fails to find evidence of any type of front running. In order to shed some light on the subject, our thesis attempts to study HFT strategies further.! 48!

53 5. Data 5.1. Why Danish stock market? Our study focuses on Danish stocks listed on Nasdaq OMX Copenhagen stock exchange. There are a number of reasons why we chose this market. First of all, we see Danish stock exchange as one of the healthiest markets as it is not threatened by economic turmoil. Ever since Euro Zone started experiencing severe economic problems related to Greece default risk, Denmark and other Nordic countries became a safe shelter for risk-averse investors (Reuters 2012). As the Danish stock market is not affected by financial distress the results of our study reveal HFT s role under normal market conditions. Secondly, high frequency traders have been increasingly penetrating the Danish stock market over the last few years. According to Nasdaq OMX Nordic, the combined share of algorithmic and high frequency trading turnover in Copenhagen stock exchange has grown from 4% in 2006 to 32% in Similar incline can be seen in the share of number of trades. The HFT share of turnover in Copenhagen stock exchange has tripled from 2.5% to 7.5% in one year from January, 2011 to January, 2012 (Nasdaq OMX 2012a). These numbers might seem relatively modest compared to other European and US stock markets. This can be explained by high percentage (~30%) of trades being executed over the counter, and the fact that HFTs prefer trading on electronic communication networks rather than traditional exchanges (Patterson & Eaglesham 2012). However, the latter fact might imply that HFTs, who decide to penetrate traditional trading venues extensively populated by institutional investors, have stronger incentives to exploit them. Thirdly, the majority of academic literature focuses on US stock markets. Even though the extent of HFT activities in Europe is smaller than in US, it is rapidly growing (Sybase 2012). Furthermore, authorities in Europe are pressured by institutional investors to introduce particular regulations (Liquidnet 2011). The European and US markets are different with respect to current regulations, investor protection, involvement of HFTs etc., therefore we identify the need to further study HFT effect in Europe (Gomber et al. 2012).! 49!

54 Finally, the Copenhagen Stock Exchange belongs to the Nasdaq Group, which means that the results of our research could partly be generalized among the other members of Nasdaq. This is especially applicable to the Nasdaq OMX Nordic and Nasdaq OMX Baltic markets. Countries belonging to these groups also belong to the European Economic Area and, thus, their exchanges are regulated by the same authorities Dataset Our dataset was provided by the Copenhagen Stock Exchange and contains information about 24 large capitalization companies listed on Nasdaq OMX Copenhagen. The small and middle capitalization firms are excluded from the sample since the majority of HF trades tend to be executed on more liquid and less volatile stocks, which are often the properties of larger companies (Hasbrouck & Saar 2011). Furthermore, the majority of trades on CSE are executed on the large capitalization companies 6. The high number of trades executed is considered to be one of the main features defining HFT (Brogaard 2010). Thus, the sample consisting only of the large cap firms should be sufficient for our study. The dataset consists of the data on trades that were executed during the first two months of the year The trades are time stamped to the accuracy of one second. Since we are investigating every trade in the time period under question, the sample consists of more than 2.25 million observations and should be representative enough for our research purposes. There are several main characteristics listed about every trade executed, such as execution date and time, volume, execution price and participants. Furthermore, there is some more specific information provided. In particular, it is stated which participant in the transaction added and which removed liquidity as well as whether an investor traded on his own account, client account or acted as a market maker. It is also exhibited what kind of order was used and whether the transaction was executed on the exchange or over the counter. However, the most important feature of the dataset for our research purposes is the possibility to distinguish which transactions were made by high frequency traders. Nasdaq OMX Nordic is among just several exchanges worldwide,!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 6 According to our estimation, around 95% of total turnover was attributable to large capitalization stocks in January, Initial data was provided by CSE.! 50!

55 which requires algorithmic traders to obtain a special type of account in order to carry their trading practices (Nasdaq OMX 2012d). Thus, algorithmic activity can be tracked with a high level of accuracy. Furthermore, we were provided with a separate dataset, identifying which trading accounts exactly could be regarded as high frequency ones. The measure used is the ratio of total orders placed to the orders that were actually executed. The ratio is calculated monthly, and if it passes the threshold of 10, an algorithmic trader s order flow is considered to be of high frequency nature for that month. Not surprisingly, in the majority of the cases, a particular market participant is either always or never acting as a HF trader, proving the viability of this measure. By implementing custom-made SAS code, we combined the information from both datasets and identified exactly, which trades were executed by HF traders. The main features of the dataset are summarized in a Table 2, with particular attention drawn on metrics regarding HFT. Table 2. Main statistics on the trades executed on the large capitalization companies in Nasdaq OMX Copenhagen. Average turnover is expressed in Danish Kroner. Average HFT turnover represents the value of transactions in which HF trader participated. Average HFT activity level is defined as a ratio of trades executed by HFTs to the number of all trades in that period, expressed in percentages. Average # of Trades Average # of Trades Average Turnover Average Turnover per minute per day per trade per day ,850 53,192 2,758,023,255 Average # of Average # of Average HFT Average HFT Average HFT HFT trades per HFT trades per Turnover per Turnover per activity level minute day trade day 13 6,231 37, ,500,000 9% As it can be observed from the Table 2, the HFT plays an important role in CSE. On average they participated in transactions worth 230 million Danish Kroner per day during January and February, Furthermore, HFT activity level for January and February was around 9%, meaning that HF trader was a participant in almost every tenth transaction executed in CSE. Our further analysis showed that there was a! 51!

56 significant variation of HFT activity level among individual stocks with the average values ranging from 4% to 21%. Finally, additional dataset consisting of information on the historical price level of OMX Copenhagen 20 (OMXC20, formerly known as KFX), was used for the analysis. The OMXC20 index portfolio consists of the 20 most traded shares of the 25 largest shares in terms of market capitalization listed on NASDAQ OMX Copenhagen (Nasdaq OMX 2012f). Thus, the index could be regarded as quite accurate average measure for the performance of the whole Danish stock market. We used historical values for OMXC dating back to the October of 1996 up to the latest values available for the month of July, The prices observed were reported daily and totaled to 3956 observations.! 52!

57 6. Methodology 6.1. Testing price efficiency! Efficient market hypothesis and martingale The efficient market hypothesis was developed independently by P.A. Samuelson and E.F. Fama in1965 and was widely applied to both theoretical and empirical models. In essence, the EMH states that prices of securities in any efficient capital market fully reflect all the relevant information available about securities traded. Therefore, investors cannot achieve abnormal profits from trading on information contained in security s price history. The idea is historically closely related to the martingale concept. A stochastic process in discrete time X t with E[X t ]< is called a discrete-time martingale, if E[X t+1 X 0,,X t ]=X t. In words, the stochastic process, conditional on all related information up to and including the time t, is expected to have the same value in a subsequent period. Thus, the martingale hypothesis generally means that the best forecast for tomorrow s asset price, if all information from its previous innovations is included, is the asset s price today. Even though the martingale concept does not hold for long-term financial returns, as investors have to be rewarded for the investmentassociated risk, it can be applied to investigate short-term returns. Martingale is closely related to the random walk model, which will form the basis of our empirical design (Arlt & Arltová 2000). Random Walk model There are three types of the random walk model distinguished in financial literature depending on assumptions regarding the distribution of error term. Generally, process of the random walk is expressed in the following manner:!! =! +!!!! +!!!,!!!!!! ~!!" 0,!!,!! 0!!! where! is the price of a security at time t,! is the expected drift and!! is an independently and identically distributed error term with the mean zero and the variance equal to!!. The independence of increments indicates that error terms are not only uncorrelated, but also their any non-linear functions are uncorrelated. Such model is defined as RW1. It follows, that the model has a conditional mean and a variance of the! 53!

58 forms: E[P t P 0 ] = P 0 +!! t; Var t [P t P 0 ]=!!!!. Therefore, it can be stated that random walk is a non-stationary process and its conditional mean and variance are linear functions of time the property, which is a cornerstone for several tests of market efficiency. The assumption of the independent and identically distributed error terms can be relaxed a bit. The RW2 model assumes independent, but not identically distributed innovations, which allows for the unconditional heteroscedasticity in increments!! and enables the model to better resemble the true evolution of financial assets prices over a longer time period. The weakest form of the RW model is called RW3, which assumes that increments are neither identically distributed nor independent. However, error terms are considered to be uncorrelated. This model allows for the conditional heteroscedasticity providing an even more realistic representation of the true behavior of asset prices (Arlt & Arltová 2000). Variance ratios In order to investigate our first research question regarding HFT and the price efficiency, we follow the methodology of variance ratio test, first developed by Lo and MacKinlay in Their methodology was used by Litzenberger et al. (2010) in their empirical study of the evolution of variance ratios over time in the U.S. stock market. However, the authors sampled observations at higher-frequency and used shorter estimation periods than conventional empirical studies that used the variance ratio test before them. In such way, Litzenberger et al. (2010) tried to capture HFT effects on the price efficiency, and draw some valuable inferences from the results. Our method, though a bit different in several aspects, is quite similar. First of all, we use continuously compounded returns described as:!! = ln!!!!!!, where P t denotes the price of security at time t. Continuously compounded multi-period return is the sum of single-period continuously compounded returns, mathematically:! 54!

59 !!! = ln![!!!!!!!!!!!!!!!!!!!!!!!! ]=!!!!! +!!!! + +!!!!!!. As stated earlier, a very important property of the random walk model is that its innovations are a linear function of time. Due to the relaxed assumptions regarding the distribution of increments, it is much harder to state this feature for RW2 and RW3 than for RW1, because the variances for individual innovations are allowed to change over time. However, it still has to hold that the variance of the sum of returns has to be equal to the sum of the variances of individual returns. Since the linearity property is valid, the variance of!!! has to be!!!. As a result, it is possible to check whether the stock price is a random walk process by comparing the observed variances from two periods, namely the!!! variance has to be! times the variance of!!. One can consider the ratio of the two periods return!! 2 to twice the return of one period return!! 1. Under stationarity condition, the variance ratio can be expressed mathematically as:!" 2 =!"#!! 2 2!"#!! 1 =!"#!! +!!!! 2!"#!! + 2!"#!!,!!!! = =1+! 1, 2!"#!! 2!"#!! where! 1 is the first-order autocorrelation coefficient of the process!!. As it can be seen, when there is no autocorrelation prevalent, the variance ratio should be equal to one. The discrepancies from the norm indicate that the variance of the process is growing either faster or slower than linearly and thus it could be stated that it does not follow the random walk. The mathematical expression for the! holding periods variance ratio is as follows:!"! = 1 + 2!!!!!!(1!!)!!(!), It has to be equal to one in order to conclude that the process is RW, even under assumptions describing RW2 and RW3, if the variances of!!!are finite. For practical implementation purposes, several estimators have to be calculated in order to arrive at the VR for specific period. In particular, assuming nq+1 observations:!! =!!"!!"!!!!!,!!!! =!!"!"!!!!(!!!!),!!!!! =!!"!!!!!(!!"!!!). where! is the estimated mean,!!! is the estimated variance for the whole estimation period and!!!! represents the variance estimated using q holding periods. Then it! 55!

60 follows that the estimated variance ratio for the period is!!" =!!!!! (Lo & MacKinlay 1988; Arlt & Arltova 2000). Test for the market efficiency Using the property of linearly changing variance over time for the random walk process, Lo & MacKinlay (1988) developed a test statistic in order to determine whether a price of a security behaves as the RW process. If averaged market index is investigated, the performance of the whole market can be evaluated and statistical conclusions can be drawn upon its efficiency. The test s null hypothesis, adapted to the properties of the RW1 process, can be formulated as follows:!! :!!! =! +!!,!!!!!! ~!!!"!! 0,!!,!!! It means that the returns! follow the random walk with the drift! and it s increments!!, are normal, independent and identically distributed with the variance!! and the mean of zero. The test statistic, adapted to accommodate for heteroscedastic increments, takes form of:!"!"! 1 ~!!! 0,2! 1, where "~! " denotes an asymptotic distribution. Test can be simply standardized to have asymptotically standard normal distribution by dividing the statistic by 2(! 1) (Lo & MacKinlay 1988). Throughout the remaining parts of our paper we refer to this test as the Lo and MacKinlay s test. Empirical design In order to evaluate the relationship between HFT and the stock price efficiency, we apply the core ideas regarding the random walk and the variance ratio test. Moreover, we closely follow the research carried out by Litzenberger et al (2010). Our research is divided into three major blocks based upon the length of the study periods. The first and the second parts intend to investigate daily data from the time period of almost 16 years. During this time algorithmic trading has developed and spread worldwide. As algorithmic trading is the root from which HFT has emerged, the first two blocks of this research are designed to evaluate the relationship between AT and the stock price! 56!

61 efficiency. The third block covers the data from the first two months of 2012 at the high frequency level. This block intends to investigate the relationship between the stock price efficiency and HFT, explicitly. 1. Historical CSE Price Efficiency Analysis The purpose of this part is to shed some light upon the overall changes in the price efficiency in the CSE during the last couple of decades. As an aggregator of the average performance of large market capitalization stocks in the CSE, we use the daily OMXC 20 index. The data obtained is divided into two blocks, depending on the time period. The first block consist of the prices on OMXC 20 index from October 1996 until the end of 2002, while the second block covers the data from the beginning of 2003 until July As argued in the section of Background Information, the algorithmic trading became prevalent worldwide in the beginning of this century (Cliff 2011). Consistently with empirical evidence found by previous research we expect to observe a positive relationship between algorithmic trading and the stock price efficiency (Hendershott et al. 2010). Therefore, we expect that prices in the second period are relatively more efficient than in the first one. We calculate the variance ratios and the test statistics proposed by Lo & MacKinlay (1988) for both periods. For comparison purposes, we calculate three variance ratios with different lags q for each period. We use variances calculated for q=2; 4; 8. For example, in the case of the two lags variance ratio, the average 1-day variance for the whole period under investigation is divided by the half of average 2-day variance for the same period. The result is then compared to one. The method is repeated using lags of four and eight trading days giving us three ratios and test statistics for each of the two periods. Our first null hypothesis is formulated as follows: there is a positive relationship between algorithmic trading and the stock price efficiency. Technically it means that the random walk hypothesis should hold more robustly for the later period of time than for the earlier period. 2. CSE Price Efficiency Analysis. Shorter time spans We carry out our analysis by investigating the same dataset of OMXC 20 historical prices, but now dividing it into several shorter periods, in order to examine if any extreme discrepancies from the norm in the variance ratios during the particular periods! 57!

62 can be found. In this way we could indicate more specific time intervals, during which the values of variance ratios are unusually higher or lower, and then investigate how these irregularities relate to the changes in the trading environment during the period. We divide the whole dataset into the blocks of yearly and two-year periods of observations. Then the procedure is the same as in the first part of our research with variance ratios calculated for q=2; 4; 8. As the algorithmic trading was booming in 2008 and from 2010 onwards 7, we expect to find a corresponding pattern in the price efficiency measures. This part of the research gives additional insights on our first hypothesis, formulated in the previous paragraph. 3. Random Walk Analysis on a Micro Level Finally, we continue our analysis on the data sampled at even shorter time horizons. For this we use the dataset provided by Nasdaq OMX Copenhagen. It consists of the tradelevel information on equity, time stamped to seconds. The purpose is to evaluate the evolution of variance ratios using high-frequency sampling for the period of two months, starting on January 2, We compare the evolution of variance ratios to the levels of HFT activity throughout the period. In order to perform our analysis, we select three most frequently traded stocks for the period. In this step we deviate a bit from the empirical design used by Litzenberger et al. (2010). The authors performed the analysis using the trade data combined of all shares together. The reason why they did not differentiate among stocks is that their data set did not allow them to identify which stocks attracted most of HFT attention. In contract, we have this possibility. Moreover, they investigated the U.S. stock markets, which are abundant of frequently traded stocks (Litzenberger et al. 2010). However, Copenhagen Stock Exchange is much smaller and there is a relatively large gap between the most and the least frequently traded companies. 8 Also, high frequency traders tend to participate in stocks that are most liquid (Aldrige 2011). Therefore, in order to capture as pure effect of high frequency traders on the stock price efficiency as possible, we chose to investigate only the three most frequently traded stocks. Unfortunately, our dataset does not include the order book. Therefore, we cannot!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 7!See Appendix D! 8 The table demonstrating the number of trades for each stock can be found in the Appendix A.! 58!

63 determine the quote midpoint, which would be helpful in investigating the price discovery process (Brogaard 2010). This is the main reason why we chose to apply the methodology consistent with Litzenberger et al. (2010) who find the way around this issue and estimate midpoints by taking the average trading prices throughout a particular time interval (e.g. 1 minute). Having an exact quote midpoint would have enabled us to apply methodologies used by Brogaard (2010), Hendershot and Riodan (2011) and Hasbrouck and Saar (2011), who investigate and quantify the direct impact of HFT on the price efficiency. Having the data only on the trade-level we can only follow Litzenberger et al. (2010), who attempt to draw conclusions based on the observed patterns of the evolution of variance ratios and the corresponding high frequency trading levels. As a result, we are also restricted to an investigation and comparison between general trends of price efficiency and HFT. Since we cannot quantify the exact impact, we formulate our second null hypothesis as follows: there is a positive relationship between high frequency trading and the stock price efficiency. In order to determine the level of the overall HFT activity during the first two months of 2012 we perform the following procedure. We start by eliminating the trades from our dataset, which were executed by investors trading from accounts other than exclusively algorithmic ones. Then we proceed by selecting the trades, which were executed by the market participants identified by Nasdaq OMX Copenhagen as operating on high frequency basis. We include all trades in which HFTs participated on either side of transactions. The same method is applied to all three individual stocks. The HFT level is measured as the number of trades, in which at least one high frequency trader participated, divided by the total number of trades executed during a particular period of time. Finally, the variance ratio analysis is performed for individual stocks on high sampling frequency. At this sampling rate the micro structural effects, such as bid-ask bounce, are expected to be prevalent, and possibly distorting results (Litzenberger et al. 2010). These effects are anticipated to be less noticeable at lower sampling frequency. As mentioned earlier, Litzenberg et al. (2010) suggests using midpoint prices for periods the data is to be partitioned to, rather than the last trading price. This should adequately minimize the bid-ask bounce effect. In our analysis, we partition the data into the bins of 1 minute and then calculate the midpoint price for each bin. In case there were no trades executed during the minute, the last calculated midpoint price is assigned to the! 59!

64 bin. Then, using the calculated values, we apply the variance ratio analysis, for the periods of one trading week. Thus, it leaves us with 8 separate periods for which variance ratios and test statistics are calculated. The variance ratios are calculated using lags of 2, 4 and 8 minutes. Following this, the corresponding Lo, MacKinlay (1988) test statistics are estimated. Finally, by examining the results, we can compare the weekly evolution of variance ratios to the evolution of the high frequency trading level Testing predatory trading strategies In order to study our second research question (Do high frequency traders employ predatory trading strategies that harm other market participants?) we chose to apply the research methods employed by Brogaard (2010) and Aldrige (2011). For us their methodology is appropriate to use due to three main reasons. Firstly, the two tests employed by these two authors can provide some insight into almost all the predatory strategies discussed in the literature review section. Aldrige s test implies the feasibility of collective price trend creation by predatory algorithms, while Brogaard s test covers exploitation possibilities using front running strategies such as (1) automated market making, (2) latency arbitrage, (3) liquidity rebate trading and (4) individual actions of some predatory algorithms. Secondly, our results become directly comparable to the evidence from the previous research. Neither Brogaard (2010) nor Aldrige (2011) find evidence of exploitative practices pursued by high frequency traders. However, our market of interest (Copenhagen Stock Market) is different from theirs (Brogaard focused on US stock market, while Aldrige investigated European futures market). Therefore, if our results are consistent with theirs, they could be generalized across other financial markets and have stronger external validity. Contradictions in findings could be attributed to the different characteristics of the financial markets, or the periods of study. Thirdly, our dataset is sufficient to apply the methodologies employed by the two authors. Without having any insider information it is very difficult to quantify the harm done by HFT to other investors (Gomber et al. 2011). Therefore, consistently with Brogaard and Aldrige, we limit ourselves to studying and answering an overall question of whether high frequency traders tend to employ strategies that harm other market participants.! 60!

65 Price trend creation In order to determine whether HFT firms tend to collectively create price trends, we perform a test used by Aldrige (2011) in her paper Can High-Frequency Traders Game Futures?. This test helps to evaluate whether HFT firms are able to drive asset prices up or down and profit from false momentum, while hurting slower investors. Consistently with the author, we refer to this strategy as the high frequency pump-anddump arbitrage. According to Gatheral (2010), the high frequency pump-and-dump arbitrage is feasible only under certain conditions related to market impact. Market impact is defined as a change in price caused by the execution of the trade (Aldrige 2011). Specifically, the pump-and-dump opportunity exists if the permanent market impact resulting from a buyer-initiated trade is not symmetric in size to the market impact resulting from a seller-initiated trade (Aldrige 2011). For instance, if the price impact following buyer-initiated trades is on average higher than the price impact caused by seller-initiated trades, traders can push the prices up by repeatedly purchasing ( pumping ) a security. After a particular price level is reached, traders can turn around and sell ( dump ) all the newly bought securities, incurring lower market impact and realizing profits. In contrast, if the permanent price impact is symmetric in size, then the increase in price gained through purchasing is offset by the decline in price incurred through selling (Aldrige 2011). In order to formally test the feasibility of the pump-and-dump arbitrage in Copenhagen Stock Exchange, we define the following model, which is consistent with Aldrige (2011).!"#$%&! denotes the volume of a trade executed at time t.!"#$%&! > 0 indicates that a trade is initiated by a buyer, while!"#$%&! < 0 stands for a seller-initiated trade (Aldrige 2011).!!"#$%&! is a permanent market impact caused by a trade of size!"#$%&!, and is measured as a change in price following the trade (Aldrige 2011). This means that if!!"#$%& >!(!"#$%&) (the absolute value of market impact caused by a buyer-initiated trade is higher than the absolute value of market impact caused by the seller-initiated trade), a trader can profit from pumping the price! 61!

66 through purchasing a security, and then dumping it, because selling affects the price less than buying. In contrast, if!!"#$%& <!(!"#$%&), one can game the market by selling securities short and then purchasing them back. (Aldrige 2011) In order to examine the market impact for different periods of time, we test our model using different durations of lags. They are measured by a number of ticks τ following the execution of a trade at time t. Figure 2. Sequence of trade executions over time. Source: Aldrige (2011) Time t -1! t! t +1 t +2 t +3 Figure 2 presents an example of executions of trades, which are distributed by unequal time intervals. For instance, when we study the market impact of a trade executed at time t using τ = 2, we look at the difference between prices at time t + 2 and at time t 1. As a result, we use the following specification of market impact!:!!!! (!"#$%&! ) = ln[!!!! ] ln![!!!! ] (1)!!!! (!"#$%&! ) = ln[!!!! ] ln![!!!! ]! It means that the market impact caused by a trade executed at time t, is equal to the logarithm of price, at which the subsequent trade was executed, minus the logarithm of price, at which the preceding trade was executed. The index τ determines the length of the lag, as it indicates the number of the subsequent trade. Consistently with Aldrige (2011) and Dufour and Engle (2000), we choose to draw the main conclusions from results given by τ = 5. Five ticks following the trade of interest has been used as a cutoff point by these authors, and has been proven to be adequate in order to properly capture market dynamics. (Aldrige 2011)! 62!

67 Consistently with a number of academics that studied market impact, e.g. Aldrige (2011), Gatheral (2010), Huberman and Stanzl (2001), Lillo, Farmer and Mantegna (2003), and Kissell and Glantz (2002), we use a linear regression specification:!!!!!"#$%&! = α! + β!!"#$%&! + ε!!! (2) Where!"#$%&! denotes the volume of a trade executed at time t (volumes of seller initiated trades are assigned a minus sign);!!!!!"#$%&! is the market impact specified in the equation (1); β! is the coefficient representing the market impact dependent on trade volume. Technically, it measures the influence of trade volume on the market impact felt after τ subsequent trade ticks. α! represents the market impact independent on trade volume. If trade volumes of buyer-initiated trades and seller-initiated trades have identical influence in absolute terms on the market impact (i.e. β!!"#$%!!"!#!$#%&!!"#$%& = β!!"##"$!!"!"#"$%!!"#$%& ), then high frequency pump-and dump arbitrage is not feasible (Aldrige 2011). As a result, our third null hypothesis is formulated as follows:!! : High frequency pump-and-dump arbitrage is not feasible and, thus, HFT firms are not able to profit by creating artificial price trends in Copenhagen Stock Exchange.!! : High frequency pump-and-dump arbitrage is feasible and, thus, HFT firms are able to profit by creating artificial price trends in Copenhagen Stock Exchange. Technically,!! :!β!!"#$%!!"!#!$#%&!!"#$%& = β!!"##"$!!"!#$#%&!!"#$%&!! :!β!!"#$%!!"!#!$#%&!!"#$%& β!!"##"$!!"!#$#%&!!"#$%& If we fail to reject the null hypothesis, it would suggest that HFT firms are not employing artificial price trend creation strategies, as they are not feasible. If we reject the null hypothesis, it would mean that HFT firms might be able to exploit pump-and-! 63!

68 dump arbitrage. However, we would not be able to provide evidence that it is actually practiced. It would remain a question for further research. We estimate the model formulated in the equation (2) using OLS, consistently with Aldrige (2011) and other authors mentioned above. Finally, we perform the difference tests between β!!"#$%!!"!#!$#%&!!"#$%& and β!!"##"$!!"!#$#%&!!"#$%&. We estimate the significance of the difference between the coefficients using another regression specification and combined observations for both buyer and seller-initiated trades:!!!!!"#$%&! = α! + β!,!!"#$%&! + β!,!!"#! + β!,!!"#$%&!"#! + ε!!! (3) The variables in the regression include trade volume and a dummy variable named Buy. Buy is equal to 1 if a trade is initiated by a buyer, and 0 if a trade is initiated by a seller. Furthermore we include the product of the trade volume and the dummy variable. The coefficient on the latter variable is the difference between the slope for buyer-initiated trades and the slope for seller-initiated trades. This is a common method to compare the regression coefficients between two groups (UCLA 2012) Front running strategies In order to test whether HFT firms tend to be involved in front running strategies, we use the methodology consistent with Brogaard s (2010). As discussed in the Literature Review section, front running strategies consist of two steps, namely, liquidity detection (e.g. pinging ) and trading ahead of institutional investors and slower retail traders. Strategies such as (1) automated market making, (2) latency arbitrage, (3) liquidity rebate trading and (4) actions of predatory algorithms rest upon both of these steps, as discussed by Arnuk and Saluzzi (2008). Therefore, the test suggested by Brogaard (2010) provides some insight into the practical application of all these strategies. In the remainder of the paper we refer to this methodology as the Brogaard s test. According to Brogaard (2010), if HFT firms are involved in front running strategies, they should increase their trading activity, when they detect a large order to be submitted by institutional investors. Therefore, we should observe a boost in a number of trades implemented by HFT firms right before a large trade is executed by an institutional investor (Brogaard 2010). Our dataset distinguishes between orders! 64!

69 submitted by HFTs and other traders; therefore we are able to test whether such practice is implemented on a systematic basis. Consistently with Brogaard (2010), for each non-hft initiated trade in our dataset we calculate the percentage of preceding trades that were initiated by HFT firms. In line with the author, we determine the percentages using 10 trades prior the trades initiated by non-hft firms. We assign every non-hft initiated trade to one of twenty bins, based on the trade size. The bins are created in such a way that every bin would contain approximately the same number of trades. Then, for each bin we calculate the average percentage of trades that were initiated by HFTs prior to non-hft initiated trades. Eventually, we compare the results among the bins using analysis of variance (ANOVA), expecting that the larger trades initiated by non-hft firms would be preceded by the higher proportion of HFT-initiated trades (Brogaard 2010). As a result, our fourth null hypothesis is formulated as follows:!! : HFT firms do not employ front running strategies.!! : HFT firms tend to systematically employ front running strategies. Technically,!! : The average percentages of trades initiated by HFTs prior to non-hft initiated trades are equal among the bins.!! : The average percentages of trades initiated by HFTs prior to non-hft initiated trades are not equal among the bins. We agree with Brogaard s (2010) proposition stating that HFTs should increase their trading activity, when they detect a large order to be submitted by institutional investors, if they are involved in front running strategies. We also agree with the assumption that only the trades that are initiated by non-hfts should be studied. However, we do not agree with the assumption that only the trades in which HFTs removed liquidity should be taken into account. Instead, we suggest that both passive and active high frequency trading should be considered. Our reasoning is based on Arnuk and Saluzzi (2008) explanations of predatory strategies discussed in our Literature Review section. According to the authors, HFT firms ping stocks in order to learn whether there are institutional investors who are willing to trade stocks at prices! 65!

70 in between the bid-ask spread (e.g. an institutional investor s algorithm might be programmed to immediately purchase stock if it is offered at some particular price). If such an institutional investor is detected, an HFT can start building up its stock position either by positioning themselves in the top of order books (passive trading) or executing marketable orders (active trading). Afterwards, the HFT firm might place an order at the price, which the institutional is willing to pay, and make him execute the order. In this example of front-running the HFT forces the non-hft is to execute a marketable order, while the HFT increases his activity in both passive and active trading in order to build up the stock position (Arnuk & Saluzzi 2008). As a result, we repeat the Brogaard s test with one distinction. We test whether an overall HFT activity increases before a large order is executed by a non-hft.! 66!

71 7. Results 7.1. Price efficiency Historical CSE price efficiency analysis In the first part of our empirical analysis, we performed Lo and Mackinley variance ratios tests for the Nasdaq OMXC 20 index. The index represents the weighted market value of 20 most traded stocks on CSE and thus can be considered as quite accurate estimate for the performance of the whole market. The analysis was performed for the periods and , with the first null hypothesis that the random walk would hold more robustly for the later period of time than for the earlier one. The main results are summarized in the Table 3. Table 3. The Lo and Mackinlay test statistics and corresponding p-values for OMXC 20 market index calculated for variance ratios with q holding periods (number of days). Critical t-statistic s absolute value = 1.96 with 5% level of significance T-statistic P-value T-statistic P-value q= q= q= As can be seen from the Table 3, the random walk hypothesis is rejected for the period while using the lag q = 2 to calculate the variance ratio with the 5% level of significance. The result is highly statistically significant with the p-value close to zero. However, while using longer holding periods, namely of 4 and 8 days, the random walk hypothesis cannot be rejected. Nevertheless, since the random walk hypothesis should hold for all the lags q chosen, one can conclude that the OMXC 20 index did not follow the random walk process during the period (Litzenberger et al. 2010). As for the period , the random walk hypothesis cannot be rejected using all three different holding periods for the variance ratio calculations. These findings imply that there was an improvement in price efficiency in CSE during the last decade. This! 67!

72 period coincides with the rise of algorithmic and, later on, high frequency trading. As a result, we find some support for our null hypothesis stating that there is a positive relationship between algorithmic trading and the stock price efficiency CSE Price efficiency analysis. Shorter time spans In order to better understand the effects automated trading may have on the price efficiency, we continue our research with the examination of shorter time intervals. We group the observations of the OMXC 20 index prices and calculate the variance ratios for every year and every two years from October 1996 to July This leaves us with 8 and 16 time intervals to be investigated. For each of the periods we apply the same methodology to calculate variance ratios and Lo & MacKinlay (1988) test statistics. The results for the test statistics 9 are quite consistent across the time periods investigated. As for observations grouped into the bins of two years, the RW1 hypothesis is not rejected across all the eight periods 10. This implies that the market was efficient during every two-year period. In order to better understand the evolution of variance ratios during the periods examined, we apply the visual approach extensively used by Litzenberg et al. (2010). Figure 3. Average variance ratio of OMXC 20, calculated every 2 years.! 1.4! 1.2! 1! 0.8! 0.6! 0.4! 0.2! 0! ! ! ! ! 2006! ! ! ! Average! variance! ratio!(2! years)! In Figure 3 the variance ratios for q holding periods of 2, 4 and 8 days are averaged. The average of the variance ratios remained relatively stable during the study period, slightly fluctuating between 1 and 1.2. Although we find that the market was efficient in every two-year period, the evolution of variance ratios, depicted in Figure 3, points out!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9 The tables with exact values can be found in Appendix B and C. 10 The results: variance ratios, t-statistics and p-values can be found in Appendix C! 68!

73 that the level of efficiency was higher during the periods and In these periods the average variance ratio reached 1, which is exactly what the RW1 predicts. Consistently with this pattern, the extent of the algorithmic trading boosted in the beginning of 2008, plummeted in 2009, and rebounded in In 2011 and 2012 the number of trades executed by algorithmic traders increased rapidly from 25% to 40%. During the same time interval the number of trades executed by broker and routing accounts decreased gradually 11. In addition, from the beginning of 2010 until 2012, average number of trades executed by HF traders increased significantly from 2.5% to 10% in the CSE 12. This suggests that the improvement in OMXC 20 price efficiency during the last two years can be related to the increase in AT and HFT activity level. On the more conservative side, it can be said that even with high rise of the AT and HFT prevalence in the CSE, the market efficiency was not worsened. Based on this evidence, we cannot reject our first null hypothesis that there is a positive relationship between algorithmic trading and the stock price efficiency. We further analyzed the random walk hypothesis for index price using even shorter time periods of one year. The calculated results 13 are highly consistent with the ones obtained for the two-year time span. They indicate that we cannot reject the random walk hypothesis for either of the periods examined except one. Surprisingly, the only period when the RW hypothesis can be rejected at 5% level of significance is the last one for the period 07/ /2012. The preceding period exhibited unusually high deviations from the variance ratio of 1 and was very close to the rejection region too. This somewhat contradicts our previous findings about the increased price efficiency during the last several years. However, after investigating the periods more carefully, we found that in the month of August, 2011, the OMXC 20 index experienced unusually high level of volatility, with the drop in value of more than 14% during the month 14. After this drop the index fully recovered by the start of Therefore, while using two years period for investigation, the Lo & MacKinlay (1988) test reports that the index price follows random walk, since both the decrease and recovery in price are captured in this timeframe. This is contrary for the one-year estimation period case, since the cut-off!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 11 The graphic illustration can be found in Appendix D 12 The graphic illustration can be found in Appendix E. 13 The results: variance ratios, t-statistics and p-values can be found in Appendix B. 14 For the graph representing changes in volatility and price of OMXC 20 during the period of interest, please see Appendix F.!! 69!

74 point coincides with the period of unusually high disturbance. Thus, the test captures one-way trend and reports that the RW hypothesis can be rejected. Since it is not a regular market event, we believe that the results suggesting the rejection of RW hypothesis for the last period investigated should be ignored. From the first part of our price efficiency analysis we may summarize that Nasdaq OMX Copenhagen cannot be described as an inefficient market. We used OMXC 20 index price, which could be considered as a proper proxy to illustrate the price fluctuations in the whole market. We consistently failed to reject the random walk hypothesis practically for every period investigated so far. Moreover, we were able to identify a possible positive relationship between the increase in the automated trading activity and the index price efficiency, which is in line with our first null hypothesis Random Walk analysis on a micro level. Relation with HFT We continued our empirical study with the price efficiency analysis for individual stocks in the high-frequency environment. First of all, we chose three stocks traded on CSE, namely Novo Nordisk B, Maersk B and Carlsberg B. The selection was made using three criterion: 1. The most traded stock; 2. The most traded stock by HFT; 3. The stock that attracted moderate HFT level. The average percentage of trades HFT were involved in was calculated to be 13.3%. It is the arithmetic mean of six most heavily traded stocks by HFT in CSE. The stock that was traded the most was VWS (Vestas Wind Systems) with 330,995 trades executed. However, we chose not to include this stock in our analysis due to important news concerning the company, which were reported during the period of our analysis. In particular, VWS announced about the significant job cuts in January and the company s chief financial officer resigned later in February. We believe that this may have triggered an unusual pattern of the stock price movements and thus it is not a valid option for our analysis. Instead, we chose Novo Nordisk B as the stock traded the most during the period there were 207,815 trades executed.! 70!

75 Maersk B was the stock, which was traded by HFT s most actively, with HF traders participating in 19.5% of all trades. Finally, Carlsberg B exhibited nearly average level of HFT in 14.9% of all transactions at least one of the counterparties was a HF trader. We believe that stocks, selected in this way, should construct a sample representative enough to capture possible relation between the HFT activity level and the stock price efficiency. The analysis was carried out by applying the variance ratio tests for each of the three stocks individually. We divided our dataset into 8 full trading weeks, and calculated variance ratios for each of them. The estimated test statistics 15 are highly statistically significant, rejecting the random walk hypothesis for all three stocks in every week. This is no surprise because at this sampling frequency, prices tend to exhibit serial autocorrelation and have extreme outliers, which make standard variance ratio tests unreliable (Andersen et al. 2001). Hence, in this part of our analysis, consistently with Litzenberger et al. (2010), we concentrated specifically on the evolution of variance ratios. We then compared it with the corresponding evolution of the HFT activity. The results for all three stocks are presented in the Figure 4. As it can be seen from the graphs, there is no general pattern of how actively HF traders engage in trading across the different periods or stocks. In some periods HFT activity increases quite drastically implying that some specific strategies are used. We will elaborate our analysis on this subject in the next section of our thesis. As for the average variance ratio, the general downward trend towards 1 can be observed in the later periods suggesting the improvement in price efficiency in February. Nonetheless, the most important finding of our analysis is the relationship between the HFT level and the variance ratio evolution during time. As it can be observed from all the three graphs in Figure 4, the average variance ratio exhibits a negative relationship with the HFT activity level for the majority of the periods. There are several intervals when the %HFT line moves to the same direction or remains neutral relatively to the Average Variance Ratio line, but the general asymmetric trend is apparent. Thus, the results imply that the higher intensity of HFT improves the stock price efficiency. These findings are in line with the broader insights concluded from the first two sections of this price efficiency analysis. However, in this case the relation is captured more!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 15 The results: variance ratios, t-statistics and p-values can be found in Appendix G.! 71!

76 precisely. Consequently, we find support for our null hypothesis, stating that there is a positive relationship between high frequency trading and the stock price efficiency. Figure 4. The average variance ratio evolution compared with the HFT activity level as a percentage of all trades executed over the period of 8 weeks.! Maersk B Novo Nordisk B % 20% 15% 10% 5% 0% % 16% 14% 12% 10% 8% 6% 4% 2% 0% Average Variance Ratio %HFT Average Variance Ratio %HFT Carlsberg B % % 15% 10% 5% Average Variance Ratio %HFT 0% 7.2. Price trend creation Main results In order to investigate the feasibility of the high frequency pump-and-dump arbitrage in the Copenhagen Stock Exchange we applied the methodology consistent with Aldrige (2011), who implemented an adequate study in the Eurex Eurobund futures market. We ran a regression (2) described in Methodology section, Price Trend Creation subsection. Our main results are presented in Table 4. The table presents the results obtained by using trade data for January and February 2012, on the 6 largest and most liquid stocks listed on Copenhagen Stock Exchange (CSE). We choose to investigate! 72!

77 only the six most frequently traded stocks, which on average are traded at least 7 times per minute, because we study the feasibility of pump-and-dump arbitrage at the high frequency level. The six stocks constitute roughly half of the total number of trades among the large cap companies listed on CSE. In line with Aldrige (2011), we report and draw the conclusions from the results obtained by using five ticks following the trade of interest (τ = 5). In relevant literature five ticks have been proven to be a sufficient cut-off point in order to properly capture market dynamics (Aldrige 2011). The regression analysis was also performed using other numbers of ticks following the trade of interest (τ equal from 1 to 10). However, it provided very similar results in terms of the sign and the significance of the regression coefficients, thus, they are not reported. Table 4 The results of the OLS regression (2)!!!!!"#$%&! = α! + β!!"#$%&! + ε!!!, described in the Methodology section, Price Trend Creation sub-section. The reported numbers are the intercept and the coefficients on trade volume, estimated using τ = 5, and trade data of stocks listed on CSE for January and February N denotes the number of observations used for regression analyses. Dependent variable: Market Impact Buyer-initiated trades!!!!!"#$%&! Intercept 1.19!!10-4 *** ( ) Volume 3.04!!10-8 *** ( ) Seller-initiated trades -1.22!!10-4 *** ( 0.000) 1.74!!10-8 *** ( ) N 476, ,471 *** denotes significance at 1 percent level. P-values are reported in parentheses. As the Table 4 indicates, the regression was run using 476,653 observations for buyerinitiated trades and 478,471 observations for seller-initiated trades. The estimated coefficients representing the market impact dependent on trade volume ( β!!"#$%!!"!#!$#%&!!"#$%& and β!!"##"$!!"!#$#%&!!"#$%& ) are equal to !! and !!, respectively. It means that an increase in trading volume by one stock in a buyer-initiated trade at time t is associated with !! % = !! % increase in the relative stock price after five ticks (!!!! /!!!! ). For instance, consider a buyer who decides to initiate a trade bidding 10,000 shares more than he initially intended. Assume that the stock was last sold at price!!!! =10 kr. The buyer s decision! 73!

78 to buy 10,000 shares more is expected to influence the share price to go up by kr per share (or almost one third of a cent per share) after five ticks, all other factors influencing the stock price holding constant. The explicit calculation can be found in Figure 5 below. Figure 5. The example of the interpretation of the coefficient representing the market impact dependent on trade volume. The buyer-initiated trade case.! Recall that our dependent variable is!"(!!!! /!!!! ), while the independent variable is Volume. An increase in volume by 1 share is associated with the !! %! or !! % increase in!!!!. Therefore 10,000 shares would result in the increase!!!!! of the ratio by %. (!!!!!" ) =>!!!! The corresponding coefficient in the case of seller-initiated trades can be interpreted similarly. An increase in trading volume by one stock in a seller-initiated trade at time t is associated with !! % = !! % decrease in the relative price of the stock after five ticks (!!!! /!!!! ). Recall that the coefficient is positive, since we assigned a minus sign to the trade volume observations in the seller-initiated case in our study. Applying the above example in Figure 5 for the seller-initiated trade case, we would expect the price to decrease by kr. to kr. per share after five ticks, assuming that no other factors influence the price during that period and that the price can obtain values expressed with 5 digits after the decimal point. We find that both coefficients for buyer-initiated and seller-initiated trades are statistically significant at 1% significance level. T-statistic for β!!"#$%!!"!#!$#%&!!"#$%& is (pvalue=0.000) while t-statistic for β!!"##"$!!"!#!$#%&!!"#$%& is (p-value=0.000). We test the significance of the difference between β!!"#$%!!"!"!#"$%!!"#$%& and β!!"##"$!!"!#!$#%&!!"#$%& using a regression specification (3), which is also described in Methodology section, Price Trend Creation sub-section. The regression results are presented in Table 5. We find that the coefficient on the variable Volume Buy, which corresponds to the difference between coefficients β!!"#$%!!"!#!$#%&!!"#$%& and β!!"##"$!!"!#!$#%&!!"#$%&, amounts to !!. However it is not statistically significant even at 10% significance level as the associated t-statistic = and p! 74!

79 value = As a result, we fail reject our null hypothesis no. 3 stating that!! :!β!!"#$%!!"!#!$#%&!!"!"#$ = β!!"##"$!!"!#$#%&!!"#$%&. In words, we fail to reject the hypothesis stating that high frequency pump-and-dump arbitrage is not feasible and, thus, HFT firms are not able to profit by creating artificial price trends in Copenhagen Stock Exchange. It means that we do not find evidence supporting the existence of high frequency pump-and-dump arbitrage opportunities. These findings are consistent with Aldrige (2011), who fails to find evidence of high frequency pump-and-dump arbitrage opportunities in Eurex Eurobund futures market. Table 5 The results of the OLS regression (3)!!!!!"#$%&! = α! + β!,!!"#$%&! + β!,!!"#! + β!,!!"#$%&!"#! + ε!!!, described in the Methodology section, Price Trend Creation sub-section. The reported numbers are the intercept and the coefficients on the independent variables, estimated using τ = 5, and trade data of stocks listed on CSE for January and February, N denotes the number of observations used for regression estimation. Dependent variable: Market Impact!!!!!"#$%&! Buyer and seller-initiated trades Intercept -1.22!!10-4 *** ( 0.000) Volume 1.74!!10-8 *** ( ) Buy 2.41!!10-4 *** ( 0.000) Volume Buy 1.30!!10-8 (0.134) N 955,124 *** denote significance at 1 percent level. P-values are reported in parentheses Weekly results Following Aldrige (2011), we divide our period of study into smaller time intervals in order to check whether the results are consistent over time. The first two months in the year 2012 had 8 full weeks, thus we divide the observations into 8 weeklong periods. We run the same regression as presented by formula (2) in the Methodology section. The results are explicitly presented in Appendix H, for the buyer-initiated trades and Appendix I, for the seller-initiated trades. The weekly dynamics of the estimated! 75!

80 volume coefficients on market impact are also depicted in Figure 6. Consistently with the main results, obtained by using all observations aggregately, we find that the volume coefficients on the market impact are positive and statistically significant for buyer-initiated trades every week. The size of the coefficients range from !! to !! and, except for week 6, all coefficients are statistically significant at 1% significance level. In week 6 the slope is significantly different from zero at 10% significance level with p-value = As for seller-initiated trades, all coefficients are positive and most of them (excluding week 1 and week 7) are statistically significant at 1% significance level. In week 1 and week 7 the slopes are significant at 5% level (pvalue = and p-value = 0.013, respectively). The coefficients range from !! to !!. Although we find trading volume to have a significant effect on stock price, we still do not find evidence that buyer-initiated trades have significantly different impact on price in absolute terms compared to seller-initiated trades. As can be seen in the Figure 6, the estimates of volume coefficients fluctuate together for buyer and seller-initiated trades during the 8 weeks. Therefore, during the weeks when prices become more sensitive to trading volume of buyer-initiated trades, the same happens for seller-initiated trades. It can be seen in the Table 6 that the differences between the coefficients for buyer and seller-initiated trades throughout the 8 weeks are not statistically different from zero: the p-values range from to Furthermore, the difference between the coefficients changes rapidly every week and sometimes changes the sign from positive to negative. These results support our conclusion that there are no pump-and-dump arbitrage opportunities in the Copenhagen Stock Exchange. Figure 6. Weekly coefficients on Volume, representing the market impact dependent on trade volume for buyer and seller-initiated trades. Jan-Feb, E>08! 5.00E>08! 4.00E>08! 3.00E>08! 2.00E>08! β!coefaicient!on!buyer> initiated!trades! β!coefaicient!on!seller> initiated!trades! 1.00E>08! 0.00E+00! 1! 2! 3! 4! 5! 6! 7! 8!! 76!

81 Table 6. Weekly differences between the coefficients on Volume for buyer and seller-initiated trades. The significance of the differences is estimated by using the OLS regression (3) described in the Methodology section, Price Trend Creation sub-section. Consistently with the remainder of the paper, they were estimated using τ = 5, and weekly trade data of stocks listed on CSE for January and February P-values are reported in parentheses. β!!"#$%!!"!#!$!"#!!"#$%& β!!"##"$!!"!#!$#%&!!"#$%&, Week !!10-9 ( ) Week !!10-9 ( ) Week !!10-9 ( ) Week !!10-9 ( ) Week !!10-9 ( ) Week !!10-9 ( ) Week !!10-9 ( ) Week !!10-9 ( ) Robustness check Finally, in order to check the robustness of our results, we extend our regression by adding two more explanatory variables. According to previous literature, liquidity and volatility help explain market impact (Aldrige 2011). For instance, Burghardt, Hanweck and Lei (2006) found that the extent of market impact depends on market depth. Almgren, Thum, Hauptmann, and Li (2005) investigated permanent market impact in equities market and concluded that the impact is affected by liquidity characteristics such as bid-ask spread. Dufour and Engle (2000) suggested that shorter inter-trade duration (time interval between trades) results in a larger market impact. According to Ferraris (2008), businesses use bid-ask spread and volatility in their models in order to forecast market impact caused by executions of trades. Our data set enables us to construct the measures of volatility and inter-trade duration. Following Gerhard and Hautsch (2000) we construct the variable of inter-trade duration! 77!

82 by measuring the waiting time between the trades executed at times! 1 and! + τ, where! equals the time of the trade of interest, and τ indicates the number of the following trade, which is used to measure the market impact (again, we draw the main conclusions using τ = 5). The stock price volatility variable is constructed in line with Boehmer et al. (2012), who investigate the high frequency trading effect on market quality. They subtract the smallest stock price recorded throughout the trading date from the highest one and standardize the difference by dividing it by the closing price. We use a corresponding measure of volatility in our study. The results of our extended regression analysis are provided in Appendix J. It can be seen that both volatility and inter-trade duration help explaining the market impact. Coefficients on both variables are statistically significant at 1% significance level in both buyer and seller-initiated cases. It can also be seen in the table that the coefficients on volume for both buyer and seller-initiated trades remain positive and statistically significant at 1% significance level. The volume coefficient for buyer-initiated trades equals !! (p-value = 0.000) and the corresponding coefficient for sellerinitiated trades equals !! (p-value = 0.000). We found similar patterns in the results of the weekly analysis, which are not reported but can be provided upon request. It appears that the inclusion of the two explanatory variables does not affect the overall results of the initial regression specification. Aldrige (2011) uses the same two variables in order to check the robustness of her results. Additionally, she uses one more variable, namely the bid-ask spread. Unfortunately, our dataset is limited to trade data and we could not get access to order books, thus we cannot include the spread variable in our analysis. However, according to Gerhard and Hautsch (2000), inter-trade duration is a measure of market intensity, and thus liquidity. As the bid-ask spread is also a liquidity characteristic, we expect the inter-trade duration variable to capture the variation in market impact caused by the variation in liquidity. Furthermore, Aldrige (2011) does not find any of the three explanatory variables to affect the results of the initial regression specification. As a result, we conclude that the initial regression specification provides us with robust results.! 78!

83 7.3. Front running Replicating the Brogaard s test In order to test whether HFTs tend to be involved in front running strategies we applied the methodology suggested by Brogaard (2010). Firstly, we present the results of the replicated Brogaard s test. Secondly, we provide the results of the same test, which was performed considering the overall HFT activity rather then merely active high frequency trading. In line with the author, we used 10 ticks prior each non-hft initiated trade to calculate the percentage of trades that were initiated by HFTs. Therefore, we omitted first 10 ticks of each trading day for each stock when looking for non-hft initiated trades. After filtering data for non-hft initiated trades we were left with 1,546,998 observations, which we assigned to 20 bins according to trade size. The bin no. 1 contains the smallest while the bin no. 20 consists of the largest trades. Each bin consists of 77,350 trades, except for the bin no. 20, which includes 77,348 trades. For each bin we calculated the average percentage of trades initiated by HFTs prior each non-hft initiated trade lying in a bin. In total our dataset contains 212,299 trades initiated by HFTs The estimated average percentages of HFT participation together with standard deviations and average trade sizes of the bins are depicted in Figure 7. Explicit numbers used to form the graph can be found in Appendix K. As we can see from the graph, HFTs tend to trade more actively compared to non-hfts, when there are larger trades to be executed by non-hfts. Recall that the activity rises only a few moments before the execution of non-hft initiated trades. Therefore, it suggests that HFTs have some superior information, which could be acquired by pinging stocks. According to Brogaard (2010) and Arnuk and Saluzzi (2008), pinging is a part of front running strategy and the increased activity can be explained as effort to build up a necessary position in stocks, which could later be offered to the non-hft.! 79!

84 Figure 7. The results of the Brogaard s test using HFT initiated trades as the measure of HFT activity %! 12.00%! 10.00%! 8.00%! 6.00%! 4.00%! 2.00%! ! ! ! ! ! 50000! Average!%!of!HFT! initiated!trades!prior!the! non>hft!initiated!trades! Standard!deviation!of! percentages!within!a!bin! Average!trade!size! initiated!by!non>hft! 0.00%! 1! 3! 5! 7! 9! 11! 13! 15! 17! 19! 0! Before drawing the conclusions we implement the analysis of variance (ANOVA) in order to check whether the differences among the average percentages are statistically significant among the bins. As can be seen in Table 7, we find the mean of squared deviation among bins to be 402 times larger than within bins. The F-statistic is equal to with p-value = Table 7 ANOVA The table presents the variation of the percentage of HFT initiated trades that occurred prior each non-hft initiated trade. The variation is decomposed into between bins and within bins variation. Source Sum of squares Discount factor Mean of squares F-statistic P-value Between Bins Within Bins Total As a result, we reject the null hypothesis no. 4 stating that the average percentages in different bins are equal. We observe increasing HFT activity when larger trades are to! 80!

85 be executed, thus we conclude that we find evidence of front running strategies being implemented by high frequency traders Taking the whole HFT activity into account We repeat the same analysis using a different way of estimating HFT activity. Instead of only picking trades initiated by HFTs, we take every trade that HFTs participated in either as liquidity providers or liquidity takers. Again our dataset consists of 1,546,998 observations, which corresponds to the number of non-hft initiated trades. Each of these observations are assigned a number which indicates the proportion of preceding trades in which HFTs participated either as liquidity providers or liquidity takers. In total, HFTs participated in 535,949 trades in our dataset. The results are presented in Figure 8. Explicit numbers used to form the graph can be found in Appendix L. We can see that the extent of HFT activity prior to the execution of larger trades is even larger when we consider not only HFT initiated trades but all trades where HFT participates. While in the previous case the average percentage of HFT participation ranged from 8% for the smallest trades to 12% for the largest trades, in this case the range is even wider: 9% for the smallest trades to 13%-16% for the largest trades. It can be seen in Table 8 that the difference in the average percentages among the bins is also statistically significant at 1% significance level. These results support our initial findings and provide us with further confidence to reject the null hypothesis of no HFTs involvement in the front running strategies. Table 8 ANOVA The table presents the variation of the percentage of HFT participation in trading that occurred prior each non-hft initiated trade. The variation is decomposed into between bins and within bins variation. Source Sum of squares Discount factor Mean of squares F-statistic P-value Between Bins Within Bins Total ! 81!

86 Figure 8. The results of the Brogaard s test using the following measure of HFT activity: HFT participation in trading either as liquidity providers or liquidity takers %! 16.00%! 14.00%! 12.00%! 10.00%! 8.00%! 6.00%! 4.00%! 2.00%! ! ! ! ! ! 50000! Average!%!of!HFT! participation!in!trading,! prior!non>hft!initiated! trades! Standard!deviation!of! percentages!within!a!bin! Average!trade!size! initiated!by!non>hft! 0.00%! 1! 3! 5! 7! 9! 11! 13! 15! 17! 19! 0!! 82!

87 8. Discussion Our study does not attempt to provide a definite answer to a question whether HFT activity has a positive or negative effect on other market participants. This is not surprising as a number of former studies also fail to reach a common conclusion regarding this issue. Generally HFT can affect investors by influencing the market quality and through predatory trading, however the overall impact might be impossible to quantify (Gomber et al. 2011). The matter in question is not whether HFTs are good or bad but rather which HFT activities are contributing to the common welfare and which of them are unfairly redistributing wealth. Authorities in US and Europe keep procrastinating the introduction of laws that would regulate HFT activities, as it is necessary to carefully examine different aspects of high frequency trading (Gomber et al. 2011). The results of our study shed some light on some of those aspects, namely the HFT relationship with the stock price efficiency, HFTs involvement in price trend creation, and the practice of front running Implications of findings Consistently with a majority of previous research, our evidence suggests a positive relationship between the extent of algorithmic trading and the stock price efficiency. The same was found for the high frequency trading in particular. It is in line with our first and second null hypotheses, respectively. Our dataset (or the absence of the order book data) does not allow us to investigate the issue more in depth in order to determine and quantify the direction of causality. However we find it hard to believe that high frequency traders would be attracted by and increase their activities in more efficient markets, instruments or particular stocks. Instead, they are expected to trade in venues or instruments where price discrepancies are observed often. Their speed advantage enables them to be the first to exploit mispricing and push the prices towards the more efficient level (McGowan 2010). As a result, we conclude that high frequency trading tends to improve the stock price efficiency. In fact, our findings also imply that HFTs involvement in market arbitrage strategies tend to overweight the possible HFTs participation in price trend creation, since these two strategies have exactly the opposite affects on price efficiency (Hendershott & Riodan 2011).! 83!

88 These findings support the evidence obtained from our study on price trend creation. We saw that the market impact dependent on trade volume of buyer-initiated trades is not statistically different from the corresponding impact of seller-initiated trades. It means that our third null hypothesis (no feasibility of pump-and-dump arbitrage in CSE) remains viable. It implies that the effort of pumping up the prices by purchasing stock is likely to be offset by the subsequent negative market impact caused while attempting to realize profits and liquidate the position. Generally speaking, the results of the tests showed that the initiation of price trends does not pay off at high frequency level, thus HFTs are not likely to get involved in such practice (at least on a regular bases). These findings are consistent with Aldrige (2011), which gives us additional confidence of their robustness. The third test we made provided controversial results compared to what has been found earlier. To our knowledge, Brogaard s (2010) test is the only attempt to study HFTs involvement in front running. The author investigated U.S. stock markets and did not find any evidence that high frequency traders would tend to front run other traders. In fact, his test showed the opposite: larger trades initiated by non-hfts were preceded by less intensive HFT activity. After replicating Brogaard s test we found that high frequency traders systematically tend to become more active compared to other traders before large trades are executed by non-hfts. It implies that they acquire some superior information about upcoming trades, which could be explained by stock pinging. Consequently, we rejected our fourth null hypotheses of no front running by HFTs. The difference between Brogaard s (2010) results and ours might be caused by the differences in the markets and the periods of study. Brogaard used data from 2008, 2009 and only 5 days from It is possible that 4-3 years ago, HFT software or hardware had not been developed to a level, which would enable them to implement front running strategies to the same extent as today. Furthermore, Brogaard identifies particular high frequency trading firms based on HFT characteristics, which means that branches of large investment banks, which could be involved in HFT are omitted. In contrast, we identify HFT activities of separate accounts, which can belong to any company, thus our measure of HFT is more precise.! 84!

89 8.2. Answering the research questions Based on the overall findings, we are finally able to answer our research questions. As far as the first question is concerned ((1) What effect does high frequency trading have on the stock price efficiency?), we conclude that the activity of high frequency traders tends to enhance the stock price efficiency. As for the second question ((2) Do high frequency traders employ predatory trading strategies that harm other market participants?), our answer is as follows: high frequency traders do not seem to get involved in price trend creation, however they tend to ping stocks and practice front running against other traders External validity Although we investigated data from a single market, namely the Copenhagen Stock Exchange (CSE), we have reasons to believe that our findings to some extent can be generalized among other markets as well. The fact that the Copenhagen Stock Exchange belongs to a world s largest trading company Nasdaq OMX Group, provides confidence that our results can have external validity. Nasdaq OMX Group operates 18 equity markets in USA and Europe. It constantly attempts to provide the newest trading technology to the members of the Group, which makes the characteristics of these markets comparable (Nasdaq OMX 2012g). This is especially applicable to other markets in Europe, particularly the other Nordic markets, which are regulated by the same law as the CSE Regulatory discussions Both USA and Europe are still searching for the best answers how to regulate high frequency traders. USA, on the other hand, has been already forced to establish some regulations regarding the issue due to the Flash Crash (Gomber et al. 2011). As discussed in the Background section, Regulations in US and Europe sub-section, the European Commission is still considering the amendments that relate to (1) controlling the risk of trading system errors, (2) market making obligations, (3) safeguard mechanisms (e.g. circuit breakers), and (4) the prevention of predatory trading. (Gomber et al. 2011)! 85!

90 The results of our research provide some insight into the last of those aspects. Our evidence suggests that HFTs tend to be engaged in predatory trading, namely pinging and front running. It is extremely difficult to quantify the redistribution of wealth caused by front running, as it would require identifying each order that was front run by HFTs. However, considering the overall HFTs investments in hardware, software, human resources and co-location, their businesses are likely to be very profitable (Friederich & Payne 2011). According to our results, at least a part of those profits come through unfair practice of front running. This is done at the expense of institutional investors, who trade on behalf of small savers, pensioners and other ordinary people, thus we believe it is a public concern to prevent such practice (Friederich & Payne 2011). The European Commission is currently considering two amendments in MiFID meant for fighting pinging and front running. The first is the introduction of a minimum time period between the submission and the cancelation of orders (Gomber et al. 2011). This regulation would impede pinging, as this technique consists of placing orders and immediately canceling them if they are not executed. Orders being forced to stay in the order book at least for a second would be exposed and other traders would be provided with an option to execute them (Gomber et al. 2011). The second amendment under consideration is the introduction of a maximum ratio of orders submitted to transactions executed by a single trader account. This regulation could also help fighting front running, as pinging requires excessive number of placements and cancelations of orders. (Gomber et al. 2011) Unfortunately, both of these amendments have significant drawbacks. Firstly, if a trader reaches the maximum trade ratio level or the minimum order lifetime is not expired, he might be restricted from reacting to important news. Secondly, minimum lifetimes could incentivize the development of algorithms that would track and exploit the trapped orders. Thirdly, the maximum trade ratios could make trading in foreign securities problematic, because such trading depends on exchange rates, which adjust at high frequency. Finally, the trade ratios might incentivize traders to extensively execute extremely small orders (e.g. one share) in order to increase the number of trades executed and, thus, reduce the overall ratio. (Gomber et al. 2011)! 86!

91 In order not to cause all these problems, the two amendments have to be carefully designed and tested. In addition to these two solutions, we propose our own modification in MiFID that would target front running activities. It consists of specifying a definition of pinging and enforcing all trading venues to track this activity. Based on Brogaard (2010), Arnuk and Saluzzi (2008) and our findings, we propose to define pinging as excessive placing and canceling of small visible and hidden limit orders within the bid-ask spread. The words excessive and small should be specified based on further research. If pinging activity is identified, the exchanges could suspend the high frequency trader s account. In case of repeated violations, some monetary punishment could be imposed. According to Liquidnet (2010), exchanges are able to track pinging activities, thus technically it should already be feasible Data limitations and suggestions for future research The main limitation of our research was the absence of the order book data. Even though we acquired an extensive data on trades, which enabled us to identify high frequency traders, liquidity providers and liquidity takers, an access to the order books would have allowed us to investigate the activities of high frequency traders even deeper. For instance, being in possession of the order book data would have enabled us to construct a variable for the bid-ask spread, which could be used in the robustness check of the Aldrige s test. Also, having an exact midpoint of the bid-ask spread would be helpful in examining price efficiency using different methodologies applied by previous research. Finally, the order book could be used to run a supplementary analysis for the Brogaard s test. We find that larger non-hft initiated trades are preceded by more intensive HFT activity, however, it could be useful to check whether those larger orders were visible in the order book. If they were hidden orders, it would provide additional confidence that HFTs learned about them by pinging stocks. We suggest these improvements to be implemented by further research. Our main suggestion for further research is to focus on studying efficient ways to prevent front running. For instance, one could attempt to develop the methodology that would efficiently track stock pinging activity. It could be used by regulators in order to! 87!

92 identify predatory traders. In order to introduce proper laws, it is necessary to establish a specific definition of pinging, which could also be a task for further research.! 88!

93 9. Conclusion The extent of high frequency trading has increased rapidly in financial markets during the last several years. Today it has already become a widespread practice, however, its contribution to the overall trading environment is still a puzzle. Particularly, mixed empirical evidence found by scholars provokes extensive debates regarding the impact of HFT on stock price efficiency, and the involvement of HFTs in predatory trading. The purpose of our thesis is to shed some more light into the topic. Based on our results we attempt to provide recommendations for regulatory bodies controlling trading activities in Europe. Our topic focuses on both the relation between HFT and the stock price efficiency, and the HFT s involvement in predatory trading strategies, which may provide significant, yet unfair gains. We approach the issue by providing the extensive review of the relevant academic literature and by performing empirical analysis on the Copenhagen Stock Exchange. The main reasons why we choose to investigate this market are the absence of relevant empirical research in Nordic markets, the steep growth curve of HFT prevalence and the unique possibility to obtain the data precisely tracking HFT activities. Our empirical results suggest that there is a positive relationship between the level of HFT and the stock price efficiency. This finding is in line with the majority of results from previous empirical studies. It implies that HFTs tend to get involved more in efficiency-improving strategies, such as market arbitrage, instead of creating mispricings by initiation of price trends. Consequently, we answer our first research question as follows. The activity of high frequency traders tends to enhance the stock price efficiency. As far as predatory trading practices are concerned, we investigate two types of strategies, namely price trend creation and front running. Consistently with the findings regarding the price efficiency, we do not find the high frequency pump-and-dump arbitrage to be feasible. It implies that price trend creation strategies do not pay off, and thus, HFTs are not likely to pursue them. It also implies that HFTs do not influence stock prices synthetically, and hence, it can be concluded that HFTs at least do not worsen the efficiency of stock prices.! 89!

94 Finally, we do find evidence that HFTs tend to use front running strategies and in this way are able to benefit at the expense of institutional investors. This conclusion contradicts the results from the previous research carried out by Brogaard (2010) who used the same methodology. However, Brogaard s (2010) and our datasets differ with respect to the study periods and the measures of HFT activities. The author used data from , whereas we investigate data from We believe that HFT algorithms and the computational power could have improved significantly during the two years period enabling HFTs to successfully apply front running strategies. As a result, the answer to our second research question is: HFTs do not seem to get involved in price trend creation, however they tend to ping stocks and practice front running against other traders. As we find evidence that HFTs tend to get involved in front running, we would like to provide several recommendations for the regulatory bodies that could prevent the unfair exploitation of ordinary investors. Firstly, as suggested by Gomber et al. (2011), a minimum time requirement between a submission and cancelation of the orders can be introduced. Secondly, a maximum ratio of submitted orders to executed trades could be imposed for a single trading account. Both of these measures would impede the pinging activity, which is closely related to the front running strategies. Thirdly, the regulatory bodies could enforce trading venues to track pinging and punish violators. This legal requirement would effectively stop pinging practice and in turn prevent HFTs from using front running strategies. Unfortunately, these propositions have their own drawbacks too, which have to be examined carefully. As a result, there is still a need for further research to find an optimal solution.! 90!

95 10. Reference List Abbink, J 2011, A Solution to Predatory High Frequency Trading?, Seeking Alpha, viewed 7 May 2012, < AFM 2010, High frequency trading: the application of advanced trading technology in the European marketplace Authority for the Financial Markets, viewed 10 May 2012, < -reportengels.ashx>. Aite Group 2009, High Frequency Trading: A Critical Ingredient in Today s Trading Market, Aite Group, viewed 13 April 2012, < Aldrige, I 2011, Can High-Frequency Traders Game Futures?, Journal of Trading, viewed 25 April 2012 < Almgren, R 2009, Quantitative Challenges in Algorithmic Execution, Quantitative Brokers, viewed 7 June 2012, < Almgren, R, Thum, C, Hauptmann, E & Li, H 2005, Direct estimation of equity market impact, Risk, vol. 18, pp , viewed 13 July 2012, < Andersen, T G, Bollerslev, T & Das, A 2001, Variance-ratio Statistics and Highfrequency Data: Testing for Changes in Intraday Volatility Patterns, viewed 04 July, 2012, < 70&uid=4&sid= >. Angel, J, Harris, L & Spatt, CS 2010, Equity Trading in the 21 st Century, Marshall Research Paper Series, viewed 17 May 2012, < Arlt, J & Arltova, M 2000, Variance ratios, viewed July 20, 2012, < Arnuk, S & Saluzzi, J 2008, Toxic equity trading order flow on Wall Street. The real force behind the explosion in volume and volatility, Themis Trading LLC, viewed 5 May 2012, < ll_street_ pdf>. Arnuk, S & Saluzzi, J 2009, Latency arbitrage: The real power behind predatory HFT, Themis Trading LLC, viewed 5 May 2012, < aper_--_latency_arbitrage_--_december_ pdf>.! 91!

96 Biais, B, Hillion, P & Spatt, C 1995, An empirical analysis of the limit order book and the order flow in the Paris Bourse, Journal of Finance, vol. 50, no. 5, pp , viewed 14 April 2012, < n_spatt_1995.pdf>. BMO Capital Markets 2009, The Impact of High Frequency Trading on the Canadian Market, BMO Capital Markets, viewed 5 May 2012, < Brogaard, J A 2010, High Frequency Trading and Its Impact on Market Quality, viewed 03 March, 2012, < Bunge, J 2010, Direct Edge Aims to Restart Pricing Fight in U.S. Stock Trade, The Wall Street Journal, viewed 7 May 2012 < html>. Burghardt, G, Hanweck, J & Lei, L 2006, Measuring Market Impact and Liquidity, Journal of Trading, vol. 1, no. 4, pp , viewed 15 July 2012, < Cadogan, G 2012, Trading Rules Over Fundamentals: A Stock Price Formula for High Frequency Trading, Bubbles and Crashes, SSRN, viewed 20 May 2012, < Cartea, A & Penalva, J 2011, Where is the Value in High Frequency Trading?, SSRN, viewed 12 March, 2012, < CESR 2010a, Call for Evidence. Microstructural issues of the European equity markets, Committee of European Securities Regulators, viewed 09 June 2012, < CESR 2010b, CESR Technical Advice to the European Commission in the Context of the MiFID Review and Responses to the European Commission Request for Additional Information, Committee of European Securities Regulators, viewed 09 June 2012, < ditional_information_mifid_review.pdf>. CFTC 2010, Co-Location/Proximity Hosting Services. Proposed rule (Federal Register (Vol. 75, No.112)), Commodity and Futures Trading Commission, viewed 09 June 2012, < Cliff, D 2011, The future of computer trading in financial markets: the impact of technology developments, Government Office for Science, viewed 17 April 2012, < Cliff, D, Brown, D & Treleaven, P 2010, Technology trends in the financial markets: a 2020 vision, Government Office for Science, viewed 17 April 2012,! 92!

97 < technology-trends-in-financial-markets>. Dufour, A & Engle, RF 2000, Time and the price impact of a trade, Journal of Finance, vol. 55, no. 6, pp , viewed 11 June 2012, < Duhigg, C 2009, Stock Traders Find Speed Pays, in Milliseconds, New York Times, viewed 8 June 2012, < Egginton, J, Ness, BFV, Ness, RAV 2012, Quote Stuffing, SSRN, viewed 17 May 2012, < European Commission 2004, Markets in Financial Instruments Directive (MiFID), Directive 2004/39/EC, viewed June , < European Commission 2008, Markets in Financial Instruments Directive (MiFID), Directive 2008/10/EC, viewed June , < European Parliament 2010, Report on regulation of trading in financial instruments dark pools etc. (2010/2075(INI)), Committee on Economic and Monetary Affairs, viewed 24 June 2012, < //EP//NONSGML+REPORT+A DOC+PDF+V0//EN>. Ferraris, A 2008, Equity Market Impact Models: Mathematics at the interface between business and research, Stifterverband fur die Deutsche Wissenschaft, Berlin, viewed 15 July 2012, < Financial Times Lexicon 2012, viewed 4 June 2012, < Foucault, T, Biais, B & Moinas, S 2011, Equilibrium High Frequency Trading, SSRN, viewed 11 March 2012, < Friederich, S & Payne, R 2011, Computer based trading, liquidity and trading costs, Government Office for Science, viewed 22 April 2012, < Garvey, R & Wu, F 2010, Speed, distance, and electronic trading: New evidence on why location matters, Journal of Financial Markets, vol. 13. Gatheral, J 2010, No-Dynamic-Arbitrage and Market Impact, Quantitative Finance, vol. 10, no. 7, pp , viewed 11 Junes 2012, < Gerhard, F & Hautsch, N 2000, Determinants of inter-trade durations using proportional hazard ARMA models, viewed 15 July 2012, < 93!

98 Gomber, P, Arndt, B, Lutat, M & Uhle, T 2011, High Frequency Trading, SSRN, viewed 5 May 2012, < Gyurko, LG 2011, The evolution of algorithmic classes, Government office for science, viewed 17 April 2012, < Hasbrouck, J & Saar, G 2011, Low-Latency Trading, SSRN, viewed 19 March 2012, < Hatrick K & Deliya D, Seasonality, Microstructure and Market Evolution: An Algorithmic Perspective, Deutsche Bank, viewed 18 May, 2012, < Hendershott, T, Jones, C M & Menkveld, A J 2010, Does Algorithmic Trading Improve Liquidity?, Journal of Finance, Vol. 66, pp. 1-33, viewed 15 March 2012, < Hendershott, T & Riodan, R 2011, Algorithmic trading and information, pp. 2,viewed 20 March 2012, < Hoffmann, P 2011, A dynamic limit order market with fast and slow traders, SSRN, viewed 03 March 2012, < Huberman, G & Stanzl W 2001, Quasi-Arbitrage and Price Manipulation, Econometrica, vol. 74, pp , viewed 11 June 2012, < Iati, R 2009, The real story of software trading espionage, Advanced Trading, viewed 7 June 2012, < Interactive Brokers 2012, viewed 15 May 2012, < Jarrow, RA & Protter, P 2011, A Dysfunctional Role of High Requency Trading in Electronic Markets, EconPapers, viewed 20 May 2012, < htm> Jovanovic, B & Menkveld, A J 2011, Middlemen in Limit-Order Markets, SSRN, viewed 05 March 2012, < Kearns, M, Kulesza, A & Nevmyvaka, Y 2010, Empirical Limitations on High Frequency Trading Profitability, Journal of Trading, vol. 5, viewed 21 May 2012, < Kirilenko, A A, Kyle, A S, Samadi, M & Tuzun 2011, The Flash Crash: The Impact of High Frequency Trading on an Electronic Market, SSRN, viewed 05 March 2012, < 94!

99 Kissel, R & Glantz M 2002, Optimal Trading Strategies: Quantitative Approaches for Managing Market Impact and Trading Risk, Amacom, New York. Lillo, F, Farmer, JD, & Mantegna RN 2003, Master curve for price impact function, Nature, vol. 421, pp , viewed 11 June 2012, < Liquidnet 2010, Concept Release on Equity Market Structure, Liquidnet, viewed 10 August 2012, < Liquidnet 2011, Institutional traders around the world concerned by high frequency trading, global survey shows, Liquidnet, viewed 3 June 2012, < L.pdf> Litzenberger, R, Castura, J, Gorelick, R & Dwivedi, Y 2010, Market Efficiency and Microstructure Evolution in U.S. Equity Markets: A High-Frequency Perspective, viewed 05 May, 2012, < Lo, A W & MacKinlay, A C, 1988 Stock Prices do not Follow Random Walks: Evidence from a Simple Specification Test, The Review of Financial Studies, Volume 1, Issue 1 (Spring, 1988), 41-66, viewed 07 April, 2012, < LSEG (London Stock Exchange Group) 2010, London Stock Exchange Group response to CESR call for evidence on micro-structural issues of the European Equity Markets, viewed 15 May 2012, < Mackenzie, M 2009, High frequency trading dominates the debate, Financial Times, viewed 7 June 2012, < Madhavan, AN 2000, Market microstructure: A survey, Journal of Financial Markets, vol. 3, no. 3, pp , viewed 24 May 2012, < Markham, JQ & Harty, DJ 2008, For Whom the Bell Tolls: The Demise of Exchange Trading Floors and the Growth of ECNs, Journal of Corporation Law, vol. 33, no. 4, viewed 14 April 2012, < McAndrews, J & Stefandis, C 2000, The emergence of electronic communications networks in the U.S. equity markets, Current Issues in Economics and Finance, Federal Reserve banks of New York, vol. 6, no. 12, pp. 1, viewed 29 May 2012, < McGowan, MJ 2010, The rise of computerized high frequency trading: use and controversy, Duke Law and Technology Review, no. 16, viewed 7 May 2012, < 95!

100 McInish, T & Upson, J 2012, Strategic Liquidity Supply in a Market with Fast and Slow Traders, SSRN, viewed 18 May 2012, < Mehta, N 2009, High Frequency Trading is a Tough Game, Traders Magazine, viewed 10 May 2012, < Menkveld, AJ 2012, High Frequency Trading and the New-Market Makers, SSRN, viewed 5 May 2012, < Mittal, H 2008, Are you playing in a Toxic Dark Pool? A Guide to preventing information leakage, The Journal of Trading, vol. 3, no. 3, viewed 15 May 2012, < Moyer, L & Lambert, E 2009, The new masters of Wall Street, Forbes, viewed 3 June 2012, < Nanex 2010, Analysis of the Flash Crash, viewed 27 April 2012, < Nanex 2011, Quote Stuffing Banned!, viewed 27 April 2012, < Nasdaq OMX 2010, NASDAQ OMX Launches INET Trading System across Its Seven Markets in the Nordics and Baltics, viewed July , < Nasdaq OMX 2012a, Computer Trading in the Nordics, presentation given during the seminar in Nasdaq OMX, Copenhagen, 8th of February Nasdaq OMX 2012b, The Nasdaq OMX Group, Inc, viewed at July , < Nasdaq OMX 2012c, Monthly Report - Equity Trading by Company and Instrument January 2012, viewed at July , < 02&lang=en>. Nasdaq OMX 2012d, Market Model NASDAQ OMX Nordic INET Nordic, viewed at June , < del_2.7.pdf>. Nasdaq OMX 2012e, NASDAQ OMX Nordic Co-Location Services, viewed at June , < Nasdaq OMX 2012f, NASDAQ OMX Copenhagen weekly turnover report, viewed at June , < 96!

101 Nasdaq OMX 2012g, What is Nasdaq?, viewed 14 August 2012, < Pagnotta, E & Philippon, T 2012, Competing on Speed, viewed 03 March 2012, < 531.pdf>. Papagiannis, N 2010, Market structure arbitrage. Fast trading techniques that are making some investors furious, Morningstar Alternative Investments Observer, vol. 2, no. 4, viewed 6 May 2012, < NonACC.PDF>. Patterson, S & Eaglesham, J 2012, SEC Probes Rapid Trading, The Wall Street Journal, viewed 29 May 2012, < ml> Patterson, S 2010, Fast traders new edge: investment firms grab stock data first, and use it seconds before others, The Wall Street Journal, viewed 7 May 2012, < ml>. Pragma Securities 2012, HFT and the hidden cost of deep liquidity, Pragma Securities, viewed 14 August 2012, < of_deep_liquidity.pdf>. QSG LLC 2009, QSG Study Proves Higher Trading Costs Incurred for VWAP Algorithms vs Arrival Price Algorithms, High Frequency Trading Contributing Factor, Quantitative Services Group LLC, viewed 20 May 2012, < Rath, J 2010, NYSE Euronext Plans Global Network of Trading Hubs, Data Center Knowledge, viewed 7 May 2012, < Reuters 2012, Danish forex reserves jump in May on intervention, viewed 8 June 2012, < SEC 2005, 17 CFR Parts 200, 201, et al. Regulation NMS; Final Rule, Securities and Exchange Commission Federal Register, viewed May , < SEC 2009, Proposed Rule: Elimination of Flash Order Exception from Rule 602 of Regulation NMS. Release No , Securities and Exchange Commission Federal Register, viewed June , SEC 2010a, SEC Approves New Rules Prohibiting Market Maker Stub Quotes,! 97!

102 Securities and Exchange Commission, viewed 06 June, 2012, < SEC 2010b, Speech by SEC Chairman: Opening Statement at the SEC Open Meeting Consolidated Audit Trail, Securities and Exchange Commission, viewed 06 June, 2012, < SEC 2010c, 17 CFR Part 242 Concept Release on Equity Market Structure; Proposed Rule Securities and Exchange Commission Federal Register, viewed May < SEC 2012, Investor Bulletin: New Measures to Address Market Volatility, viewed June , < Sornette, D & Von der Becke, S 2011, Crashes and High Frequency Trading, Swiss Finance Institute Research Paper, no , viewed 25 April 2012, < Stoll, HR 2006, Electronic Trading in Stock Markets, Journal of Economic Perspective, vol. 20, no. 1, viewed 29 May 2012, < Sybase 2012, Staying on top of algos, Sybase, viewed 3 June 2012, < Tabb Group 2009, US Equity High Frequency Trading: Strategies, Sizing and Market Structure, Tabb Group, New York, viewed 7 May 2012, < UCLA 2012, Academic Technology Services, viewed 13 July 2012, < Wahba, P & Chasan, E 2009, Geeks trump alpha-males as algos dominate Wall St., Reuters, viewed 7 June 2012, < Zhang, F X 2010, High-Frequency Trading, Stock Volatility, and Price Discovery, SSRN, viewed 21 March, 2012, < 98!

ELECTRONIC TRADING GLOSSARY

ELECTRONIC TRADING GLOSSARY ELECTRONIC TRADING GLOSSARY Algorithms: A series of specific steps used to complete a task. Many firms use them to execute trades with computers. Algorithmic Trading: The practice of using computer software

More information

High Frequency Trading Volumes Continue to Increase Throughout the World

High Frequency Trading Volumes Continue to Increase Throughout the World High Frequency Trading Volumes Continue to Increase Throughout the World High Frequency Trading (HFT) can be defined as any automated trading strategy where investment decisions are driven by quantitative

More information

From Traditional Floor Trading to Electronic High Frequency Trading (HFT) Market Implications and Regulatory Aspects Prof. Dr. Hans Peter Burghof

From Traditional Floor Trading to Electronic High Frequency Trading (HFT) Market Implications and Regulatory Aspects Prof. Dr. Hans Peter Burghof From Traditional Floor Trading to Electronic High Frequency Trading (HFT) Market Implications and Regulatory Aspects Prof. Dr. Hans Peter Burghof Universität Hohenheim Institut für Financial Management

More information

High frequency trading

High frequency trading High frequency trading Bruno Biais (Toulouse School of Economics) Presentation prepared for the European Institute of Financial Regulation Paris, Sept 2011 Outline 1) Description 2) Motivation for HFT

More information

FI report. Investigation into high frequency and algorithmic trading

FI report. Investigation into high frequency and algorithmic trading FI report Investigation into high frequency and algorithmic trading FEBRUARY 2012 February 2012 Ref. 11-10857 Contents FI's conclusions from its investigation into high frequency trading in Sweden 3 Background

More information

High Frequency Trading Background and Current Regulatory Discussion

High Frequency Trading Background and Current Regulatory Discussion 2. DVFA Banken Forum Frankfurt 20. Juni 2012 High Frequency Trading Background and Current Regulatory Discussion Prof. Dr. Peter Gomber Chair of Business Administration, especially e-finance E-Finance

More information

G100 VIEWS HIGH FREQUENCY TRADING. Group of 100

G100 VIEWS HIGH FREQUENCY TRADING. Group of 100 G100 VIEWS ON HIGH FREQUENCY TRADING DECEMBER 2012 -1- Over the last few years there has been a marked increase in media and regulatory scrutiny of high frequency trading ("HFT") in Australia. HFT, a subset

More information

Toxic Equity Trading Order Flow on Wall Street

Toxic Equity Trading Order Flow on Wall Street Toxic Equity Trading Order Flow on Wall Street INTRODUCTION The Real Force Behind the Explosion in Volume and Volatility By Sal L. Arnuk and Joseph Saluzzi A Themis Trading LLC White Paper Retail and institutional

More information

Fast Trading and Prop Trading

Fast Trading and Prop Trading Fast Trading and Prop Trading B. Biais, F. Declerck, S. Moinas (Toulouse School of Economics) December 11, 2014 Market Microstructure Confronting many viewpoints #3 New market organization, new financial

More information

Algorithmic and advanced orders in SaxoTrader

Algorithmic and advanced orders in SaxoTrader Algorithmic and advanced orders in SaxoTrader Summary This document describes the algorithmic and advanced orders types functionality in the new Trade Ticket in SaxoTrader. This functionality allows the

More information

Algorithmic trading Equilibrium, efficiency & stability

Algorithmic trading Equilibrium, efficiency & stability Algorithmic trading Equilibrium, efficiency & stability Presentation prepared for the conference Market Microstructure: Confronting many viewpoints Institut Louis Bachelier Décembre 2010 Bruno Biais Toulouse

More information

White Paper Electronic Trading- Algorithmic & High Frequency Trading. PENINSULA STRATEGY, Namir Hamid

White Paper Electronic Trading- Algorithmic & High Frequency Trading. PENINSULA STRATEGY, Namir Hamid White Paper Electronic Trading- Algorithmic & High Frequency Trading PENINSULA STRATEGY, Namir Hamid AUG 2011 Table Of Contents EXECUTIVE SUMMARY...3 Overview... 3 Background... 3 HIGH FREQUENCY ALGORITHMIC

More information

Interactive Brokers Quarterly Order Routing Report Quarter Ending March 31, 2013

Interactive Brokers Quarterly Order Routing Report Quarter Ending March 31, 2013 I. Introduction Interactive Brokers Quarterly Order Routing Report Quarter Ending March 31, 2013 Interactive Brokers ( IB ) has prepared this report pursuant to a U.S. Securities and Exchange Commission

More information

Setting the Scene. FIX the Enabler & Electronic Trading

Setting the Scene. FIX the Enabler & Electronic Trading Setting the Scene FIX the Enabler & Electronic Trading Topics Overview of FIX and connectivity Direct Market Access Algorithmic Trading Dark Pools and Smart Order Routing 2 10,000+ firms use FIX globally

More information

Financial Markets and Institutions Abridged 10 th Edition

Financial Markets and Institutions Abridged 10 th Edition Financial Markets and Institutions Abridged 10 th Edition by Jeff Madura 1 12 Market Microstructure and Strategies Chapter Objectives describe the common types of stock transactions explain how stock transactions

More information

Interactive Brokers Order Routing and Payment for Orders Disclosure

Interactive Brokers Order Routing and Payment for Orders Disclosure Interactive Brokers Order Routing and Payment for Orders Disclosure 1. IB's Order Routing System: IB does not sell its order flow to another broker to handle and route. Instead, IB has built a real-time,

More information

Analysis of High-frequency Trading at Tokyo Stock Exchange

Analysis of High-frequency Trading at Tokyo Stock Exchange This article was translated by the author and reprinted from the June 2014 issue of the Securities Analysts Journal with the permission of the Securities Analysts Association of Japan (SAAJ). Analysis

More information

High-frequency trading, flash crashes & regulation Prof. Philip Treleaven

High-frequency trading, flash crashes & regulation Prof. Philip Treleaven High-frequency trading, flash crashes & regulation Prof. Philip Treleaven Director, UCL Centre for Financial Computing UCL Professor of Computing www.financialcomputing.org [email protected] Normal

More information

This paper sets out the challenges faced to maintain efficient markets, and the actions that the WFE and its member exchanges support.

This paper sets out the challenges faced to maintain efficient markets, and the actions that the WFE and its member exchanges support. Understanding High Frequency Trading (HFT) Executive Summary This paper is designed to cover the definitions of HFT set by regulators, the impact HFT has made on markets, the actions taken by exchange

More information

High-frequency trading and execution costs

High-frequency trading and execution costs High-frequency trading and execution costs Amy Kwan Richard Philip* Current version: January 13 2015 Abstract We examine whether high-frequency traders (HFT) increase the transaction costs of slower institutional

More information

The Need for Speed: It s Important, Even for VWAP Strategies

The Need for Speed: It s Important, Even for VWAP Strategies Market Insights The Need for Speed: It s Important, Even for VWAP Strategies November 201 by Phil Mackintosh CONTENTS Speed benefits passive investors too 2 Speed helps a market maker 3 Speed improves

More information

What is High Frequency Trading?

What is High Frequency Trading? What is High Frequency Trading? Released December 29, 2014 The impact of high frequency trading or HFT on U.S. equity markets has generated significant attention in recent years and increasingly in the

More information

Dated January 2015 Advanced Execution Services. Crossfinder User Guidelines Asia Pacific

Dated January 2015 Advanced Execution Services. Crossfinder User Guidelines Asia Pacific Dated January 2015 Advanced Execution Services Crossfinder User Guidelines Asia Pacific Important Matters Relating to Orders Routed to Crossfinder Credit Suisse s alternative execution platform Crossfinder

More information

- JPX Working Paper - Analysis of High-Frequency Trading at Tokyo Stock Exchange. March 2014, Go Hosaka, Tokyo Stock Exchange, Inc

- JPX Working Paper - Analysis of High-Frequency Trading at Tokyo Stock Exchange. March 2014, Go Hosaka, Tokyo Stock Exchange, Inc - JPX Working Paper - Analysis of High-Frequency Trading at Tokyo Stock Exchange March 2014, Go Hosaka, Tokyo Stock Exchange, Inc 1. Background 2. Earlier Studies 3. Data Sources and Estimates 4. Empirical

More information

WORKING WORKING PAPER PAPER

WORKING WORKING PAPER PAPER Japan Exchange Group, Inc. Visual Identity Design System Manual Japan Exchange Group, Inc. Japan Exchange Group, Inc. Visual Identity Design System Manual Visual Identity Design System Manual JPX JPX 17

More information

An objective look at high-frequency trading and dark pools May 6, 2015

An objective look at high-frequency trading and dark pools May 6, 2015 www.pwc.com/us/investorresourceinstitute An objective look at high-frequency trading and dark pools May 6, 2015 An objective look at high-frequency trading and dark pools High-frequency trading has been

More information

FINANCIER. An apparent paradox may have emerged in market making: bid-ask spreads. Equity market microstructure and the challenges of regulating HFT

FINANCIER. An apparent paradox may have emerged in market making: bid-ask spreads. Equity market microstructure and the challenges of regulating HFT REPRINT FINANCIER WORLDWIDE JANUARY 2015 FINANCIER BANKING & FINANCE Equity market microstructure and the challenges of regulating HFT PAUL HINTON AND MICHAEL I. CRAGG THE BRATTLE GROUP An apparent paradox

More information

Market Making and Liquidity Provision in Modern Markets

Market Making and Liquidity Provision in Modern Markets Canada STA 2015 Market Making and Liquidity Provision in Modern Markets Phil Mackintosh 2 What am I going to talk about? Why are Modern Markets Important? Trading is now physics at the speed of light Jan

More information

ORDER EXECUTION POLICY

ORDER EXECUTION POLICY ORDER EXECUTION POLICY Saxo Capital Markets UK Limited is authorised and regulated by the Financial Conduct Authority, Firm Reference Number 551422. Registered address: 26th Floor, 40 Bank Street, Canary

More information

CFTC Technology Advisory Committee Sub-Committee on Automated and High Frequency Trading Working Group 1 Participants

CFTC Technology Advisory Committee Sub-Committee on Automated and High Frequency Trading Working Group 1 Participants CFTC Technology Advisory Committee Sub-Committee on Automated and High Frequency Trading Working Group 1 Participants Joan Manley, George Pullen CFTC Sean Castette Getco LLC Colin Clark NYSE Euronext Chris

More information

Machine Learning and Algorithmic Trading

Machine Learning and Algorithmic Trading Machine Learning and Algorithmic Trading In Fixed Income Markets Algorithmic Trading, computerized trading controlled by algorithms, is natural evolution of security markets. This area has evolved both

More information

High-frequency trading: towards capital market efficiency, or a step too far?

High-frequency trading: towards capital market efficiency, or a step too far? Agenda Advancing economics in business High-frequency trading High-frequency trading: towards capital market efficiency, or a step too far? The growth in high-frequency trading has been a significant development

More information

High-Frequency Trading

High-Frequency Trading High-Frequency Trading Peter Gomber, Björn Arndt, Marco Lutat, Tim Uhle Chair of Business Administration, especially e-finance E-Finance Lab Prof. Dr. Peter Gomber Campus Westend RuW P.O. Box 69 D-60629

More information

Toxic Arbitrage. Abstract

Toxic Arbitrage. Abstract Toxic Arbitrage Thierry Foucault Roman Kozhan Wing Wah Tham Abstract Arbitrage opportunities arise when new information affects the price of one security because dealers in other related securities are

More information

Robert Bartlett UC Berkeley School of Law. Justin McCrary UC Berkeley School of Law. for internal use only

Robert Bartlett UC Berkeley School of Law. Justin McCrary UC Berkeley School of Law. for internal use only Shall We Haggle in Pennies at the Speed of Light or in Nickels in the Dark? How Minimum Price Variation Regulates High Frequency Trading and Dark Liquidity Robert Bartlett UC Berkeley School of Law Justin

More information

FAIR GAME OR FATALLY FLAWED?

FAIR GAME OR FATALLY FLAWED? ISN RESEARCH FAIR GAME OR FATALLY FLAWED? SOME COSTS OF HIGH FREQUENCY TRADING IN LOW LATENCY MARKETS June 2013 KEY POINTS The activity of High Frequency Traders (HF traders) in Australia s equity markets

More information

Bank of Canada Workshop on Regulation, Transparency, and the Quality of Fixed- Income Markets

Bank of Canada Workshop on Regulation, Transparency, and the Quality of Fixed- Income Markets Financial System Review Bank of Canada Workshop on Regulation, Transparency, and the Quality of Fixed- Income Markets Lorie Zorn In February 2004, the Bank of Canada hosted a two-day workshop, Regulation,

More information

Chinese University of Hong Kong Conference on HKEx and the Market Structure Revolution

Chinese University of Hong Kong Conference on HKEx and the Market Structure Revolution Chinese University of Hong Kong Conference on HKEx and the Market Structure Revolution The Impact of Market Structure Changes on Securities Exchanges Regulation 31 March 2012 Keith Lui Executive Director,

More information

Towards an Automated Trading Ecosystem

Towards an Automated Trading Ecosystem Towards an Automated Trading Ecosystem Charles-Albert LEHALLE May 16, 2014 Outline 1 The need for Automated Trading Suppliers Users More technically... 2 Implied Changes New practices New (infrastructure)

More information

TRADING STRATEGIES 2011. Urs Rutschmann, COO Tbricks

TRADING STRATEGIES 2011. Urs Rutschmann, COO Tbricks TRADING STRATEGIES 2011 Urs Rutschmann, COO Tbricks SEMANTIC CHALLENGE High Frequency Trading Low latency trading Algorithmic trading Strategy trading Systematic trading Speed Arbitrage Statistical Arbitrage

More information

SPDR S&P Software & Services ETF

SPDR S&P Software & Services ETF SPDR S&P Software & Services ETF Summary Prospectus-October 31, 2015 XSW (NYSE Ticker) Before you invest in the SPDR S&P Software & Services ETF (the Fund ), you may want to review the Fund's prospectus

More information

SAXO BANK S BEST EXECUTION POLICY

SAXO BANK S BEST EXECUTION POLICY SAXO BANK S BEST EXECUTION POLICY THE SPECIALIST IN TRADING AND INVESTMENT Page 1 of 8 Page 1 of 8 1 INTRODUCTION 1.1 This policy is issued pursuant to, and in compliance with, EU Directive 2004/39/EC

More information

Statement of Kevin Cronin Global Head of Equity Trading, Invesco Ltd. Joint CFTC-SEC Advisory Committee on Emerging Regulatory Issues August 11, 2010

Statement of Kevin Cronin Global Head of Equity Trading, Invesco Ltd. Joint CFTC-SEC Advisory Committee on Emerging Regulatory Issues August 11, 2010 Statement of Kevin Cronin Global Head of Equity Trading, Invesco Ltd. Joint CFTC-SEC Advisory Committee on Emerging Regulatory Issues August 11, 2010 Thank you, Chairman Schapiro, Chairman Gensler and

More information

a. CME Has Conducted an Initial Review of Detailed Trading Records

a. CME Has Conducted an Initial Review of Detailed Trading Records TESTIMONY OF TERRENCE A. DUFFY EXECUTIVE CHAIRMAN CME GROUP INC. BEFORE THE Subcommittee on Capital Markets, Insurance and Government Sponsored Enterprises of the HOUSE COMMITTEE ON FINANCIAL SERVICES

More information

Valdi for Equity Trading High performance trading solutions for global markets

Valdi for Equity Trading High performance trading solutions for global markets Valdi for Equity Trading High performance trading solutions for global markets EDA Orders SunGard s VALDI: SOLUTIONS FOR Equity Trading Traders on electronic markets face enormous challenges in maintaining

More information

Increased Scrutiny of High-Frequency Trading

Increased Scrutiny of High-Frequency Trading Increased Scrutiny of High-Frequency Trading Posted by Noam Noked, co-editor, HLS Forum on Corporate Governance and Financial Regulation, on Friday May 23, 2014 Editor s Note: The following post comes

More information

INTERACTIVE BROKERS LLC A Member of the Interactive Brokers Group

INTERACTIVE BROKERS LLC A Member of the Interactive Brokers Group David M. Battan Executive Vice President and General Counsel INTERACTIVE BROKERS LLC A Member of the Interactive Brokers Group 1725 EYE STREET, N.W. SUITE 300 WASHINGTON, DC 20006 TEL (202) 530-3205 July

More information

General Forex Glossary

General Forex Glossary General Forex Glossary A ADR American Depository Receipt Arbitrage The simultaneous buying and selling of a security at two different prices in two different markets, with the aim of creating profits without

More information

Valdi. Equity Trading

Valdi. Equity Trading Valdi Equity Trading Valdi EDA Orders VALDI SOLUTIONS FOR EQUITY TRADING Traders on electronic markets face enormous challenges in maintaining and growing their profitability. The drive for greater efficiency

More information

Investors Exchange (IEX) Building A Market that Works for Investors. September 9, 2015

Investors Exchange (IEX) Building A Market that Works for Investors. September 9, 2015 Investors Exchange (IEX) Building A Market that Works for Investors September 9, 2015 FLASH BOYS: A WALL STREET REVOLT 2015 IEX Group, Inc. and its subsidiaries, including Investors Exchange LLC and IEX

More information

How aggressive are high frequency traders?

How aggressive are high frequency traders? How aggressive are high frequency traders? Björn Hagströmer, Lars Nordén and Dong Zhang Stockholm University School of Business, S 106 91 Stockholm July 30, 2013 Abstract We study order aggressiveness

More information

High Frequency Trading + Stochastic Latency and Regulation 2.0. Andrei Kirilenko MIT Sloan

High Frequency Trading + Stochastic Latency and Regulation 2.0. Andrei Kirilenko MIT Sloan High Frequency Trading + Stochastic Latency and Regulation 2.0 Andrei Kirilenko MIT Sloan High Frequency Trading: Good or Evil? Good Bryan Durkin, Chief Operating Officer, CME Group: "There is considerable

More information

The Future of Algorithmic Trading

The Future of Algorithmic Trading The Future of Algorithmic Trading Apama Track Dan Hubscher, Principal Product Marketing Manager, Progress Apama Progress Exchange Brasil 2009 Progress Exchange Brasil 2009 21 de Outubro São Paulo Brasil

More information

Effective Trade Execution

Effective Trade Execution Effective Trade Execution Riccardo Cesari Massimiliano Marzo Paolo Zagaglia Quaderni - Working Paper DSE N 836 Effective Trade Execution 1 RICCARDO CESARI Department of Statistics, University of Bologna

More information

Stock Market -Trading and market participants

Stock Market -Trading and market participants Stock Market -Trading and market participants Ruichang LU ( 卢 瑞 昌 ) Department of Finance Guanghua School of Management Peking University Overview Trading Stock Understand trading order Trading cost Margin

More information

Market Structure Overview. Goldman Sachs September, 2009

Market Structure Overview. Goldman Sachs September, 2009 Market Structure Overview Goldman Sachs September, 2009 Summary The US equities market is increasingly efficient and is broadly regarded as the best in the world. Spreads are reduced, execution costs are

More information

Nine Questions Every ETF Investor Should Ask Before Investing

Nine Questions Every ETF Investor Should Ask Before Investing Nine Questions Every ETF Investor Should Ask Before Investing UnderstandETFs.org Copyright 2012 by the Investment Company Institute. All rights reserved. ICI permits use of this publication in any way,

More information

ESMA MiFID II / MiFIR Consultation and Discussion Papers General Comments on Market Structure Issues

ESMA MiFID II / MiFIR Consultation and Discussion Papers General Comments on Market Structure Issues 1 August 2014 ESMA MiFID II / MiFIR Consultation and Discussion Papers General Comments on Market Structure Issues Financial markets have changed and technology has evolved meaningfully since 2007 when

More information

What do we know about high-frequency trading? Charles M. Jones* Columbia Business School Version 3.4: March 20, 2013 ABSTRACT

What do we know about high-frequency trading? Charles M. Jones* Columbia Business School Version 3.4: March 20, 2013 ABSTRACT What do we know about high-frequency trading? Charles M. Jones* Columbia Business School Version 3.4: March 20, 2013 ABSTRACT This paper reviews recent theoretical and empirical research on high-frequency

More information

Optimal trading? In what sense?

Optimal trading? In what sense? Optimal trading? In what sense? Market Microstructure in Practice 3/3 Charles-Albert Lehalle Senior Research Advisor, Capital Fund Management, Paris April 2015, Printed the April 13, 2015 CA Lehalle 1

More information

A practical guide to FX Arbitrage

A practical guide to FX Arbitrage A practical guide to FX Arbitrage FX Arbitrage is a highly debated topic in the FX community with many unknowns, as successful arbitrageurs may not be incentivized to disclose their methodology until after

More information

About Hedge Funds. What is a Hedge Fund?

About Hedge Funds. What is a Hedge Fund? About Hedge Funds What is a Hedge Fund? A hedge fund is a fund that can take both long and short positions, use arbitrage, buy and sell undervalued securities, trade options or bonds, and invest in almost

More information

Algorithmic Trading, High-Frequency Trading and Colocation: What does it mean to Emerging Market?

Algorithmic Trading, High-Frequency Trading and Colocation: What does it mean to Emerging Market? Algorithmic Trading, High-Frequency Trading and Colocation: What does it mean to Emerging Market? Ashok Jhunjhunwala, IIT Madras [email protected] HFTs are being pushed out of the more established markets,

More information

Empower Mobile Algorithmic Trading Services with Cloud Computing

Empower Mobile Algorithmic Trading Services with Cloud Computing Empower Mobile Algorithmic Trading Services with Cloud Computing Junwei Ma Southwestern University of Finance and Economics Abstract: Mobile devices can offer real time connection to stock exchange market

More information

A New Episode in the Stock Exchange Mergers Saga: Intercontinental Exchange (ICE)/New York Stock Exchange (NYSE)

A New Episode in the Stock Exchange Mergers Saga: Intercontinental Exchange (ICE)/New York Stock Exchange (NYSE) Competition Policy International A New Episode in the Stock Exchange Mergers Saga: Intercontinental Exchange (ICE)/New York Stock Exchange (NYSE) Professor Ioannis Kokkoris (Center for Commercial Law and

More information

ELECTRONIC TRADING AND FINANCIAL MARKETS

ELECTRONIC TRADING AND FINANCIAL MARKETS November 29, 2010 Bank of Japan ELECTRONIC TRADING AND FINANCIAL MARKETS Speech at the Paris EUROPLACE International Financial Forum in Tokyo Kiyohiko G. Nishimura Deputy Governor of the Bank of Japan

More information

Millennium Exchange. Fast, flexible, multi-asset trading engine

Millennium Exchange. Fast, flexible, multi-asset trading engine Millennium Exchange Fast, flexible, multi-asset trading engine To thrive in the current environment, you need a trading platform that moves as quickly as you do. Innovation is power Millennium Exchange

More information

NDD execution: NDD can help remove the conflict of interest >>> providing a confl ict free environment for Retail FX traders CLIENT.

NDD execution: NDD can help remove the conflict of interest >>> providing a confl ict free environment for Retail FX traders CLIENT. The Broker team NDD execution: providing a confl ict free environment for Retail FX traders In forex trading, the electronic execution engine used by Non Dealing Desk (NDD) brokers provides traders with

More information

Testimony on H.R. 1053: The Common Cents Stock Pricing Act of 1997

Testimony on H.R. 1053: The Common Cents Stock Pricing Act of 1997 Testimony on H.R. 1053: The Common Cents Stock Pricing Act of 1997 Lawrence Harris Marshall School of Business University of Southern California Presented to U.S. House of Representatives Committee on

More information

The Impact of High Frequency Trading on the Canadian Market

The Impact of High Frequency Trading on the Canadian Market The Impact of High Frequency Trading on the Canadian Market July 22, 2009 Quantitative Execution Services (416) 359-5743 [email protected] Doug Clark (416) 359-4151 [email protected] Rizwan Awan, CFA (416)

More information

Conditional and complex orders

Conditional and complex orders Conditional and complex orders Securities Trading: Principles and Procedures Chapter 12 Algorithms (Algos) Less complex More complex Qualified orders IOC, FOK, etc. Conditional orders Stop, pegged, discretionary,

More information

Algorithmic trading - Overview. Views expressed herein are personal views of the author

Algorithmic trading - Overview. Views expressed herein are personal views of the author Algorithmic trading - Overview Views expressed herein are personal views of the author Scenario 1 You are a fund manager and have Rs 500 Crores in hand (USD 83 million) to be invested. You have a highly

More information

Trade Execution Analysis Generated by Markit

Trade Execution Analysis Generated by Markit Trade Execution Analysis Generated by Markit Global Liquidity Partners Execution Peer Review 2nd Quarter 2015 Contents S VT POV Report Summary Summarizes the trade execution document and illustrates the

More information

Digitization of Financial Markets: Impact and Future

Digitization of Financial Markets: Impact and Future Digitization of Financial Markets: Impact and Future Prateek Rani 1, Adithya Srinivasan 2 Abstract Financial instruments were traditionally traded when stockbrokers and traders met at trading floors and

More information

Xetra. The market. Xetra: Europe s largest trading platform for ETFs. ETF. One transaction is all you need.

Xetra. The market. Xetra: Europe s largest trading platform for ETFs. ETF. One transaction is all you need. Xetra. The market. Xetra: Europe s largest trading platform for ETFs ETF. One transaction is all you need. Deutsche Börse Group is the leading global service provider to the securities industry. Its cutting-edge

More information

Bernard S. Donefer Distinguished Lecturer Baruch College, CUNY [email protected]

Bernard S. Donefer Distinguished Lecturer Baruch College, CUNY bernard.donefer@baruch.cuny.edu Bernard S. Donefer Distinguished Lecturer Baruch College, CUNY [email protected] Principal, Conatum Consulting LLC www.conatum.com [email protected] 2008 Bernard S. Donefer. All rights

More information

High Frequency Trading An Asset Manager s Perspective

High Frequency Trading An Asset Manager s Perspective NBIM Discussion NOTE #1-2013 High Frequency Trading An Asset Manager s Perspective In this note we review the rapidly expanding literature in the area of market microstructure, high frequency and computer-based

More information

February 22, 2015 MEMORANDUM

February 22, 2015 MEMORANDUM February 22, 2015 MEMORANDUM Re: Due Diligence Information for Advisors, Brokers, Hedge Funds and Other Financial Institutions and Intermediaries Using or Considering Interactive Brokers LLC as Prime Broker/Custodian

More information

Smart or Out Smarted?

Smart or Out Smarted? Smart or Out Smarted? By Michael O Conor, Jordan & Jordan The phrase Smart Order Routing (SOR) has become standard terminology only in the past few years with dramatic changes in equity trading caused

More information

News Trading and Speed

News Trading and Speed News Trading and Speed Thierry Foucault, Johan Hombert, and Ioanid Rosu (HEC) High Frequency Trading Conference Plan Plan 1. Introduction - Research questions 2. Model 3. Is news trading different? 4.

More information

Delivering NIST Time to Financial Markets Via Common-View GPS Measurements

Delivering NIST Time to Financial Markets Via Common-View GPS Measurements Delivering NIST Time to Financial Markets Via Common-View GPS Measurements Michael Lombardi NIST Time and Frequency Division [email protected] 55 th CGSIC Meeting Timing Subcommittee Tampa, Florida September

More information