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 costs other investors, including long term investors such as superannuation funds, an estimated $1.6 to $1.9 billion per year. This estimate is for the opportunity cost created by HF traders interposing between natural buyers and sellers who may need to accept disadvantageous prices to obtain an execution. The opportunity cost is only one of three kinds of costs created by high frequency trading (HFT) in low latency markets and borne by other investors. Due to the absence of publicly available detailed audit trail market data, this estimate is based upon conservative aggregate market data on HFT activity, and relies on trading behaviour assumptions. The analysis provides further evidence that the changes to market systems that cater to HFT are undermining not only the public policy objectives underlying market regulation, but also the public policy objectives of the superannuation system. About Industry Super Network Industry Super Network (ISN) is an umbrella organisation for the industry super movement. ISN manages collective projects on behalf of a number of industry super funds with the objective of maximising the retirement savings of five million industry super members. Please direct questions and comments to: Zachary May Director of Policy Level 30, 2 Lonsdale Street, Melbourne, VIC 3000 03 9657 4369 zmay@industrysuper.com Anh Nguyen Policy Analyst Financial Markets Level 30, 2 Lonsdale Street, Melbourne, VIC 3000 03 9657 4350 anguyen@industrysuper.com Dr Christian Clark Research Analyst Level 30, 2 Lonsdale Street, Melbourne, VIC 3000 03 9657 4296 cclark@industrysuper.com The costs of high frequency trading June 2013 www.industrysupernetwork.com 1
Table of Contents KEY POINTS... 1 1. INTRODUCTION... 3 2. LITERATURE REVIEW... 5 2.1 Nanex Research (2012)... 5 2.2 Narang/Tradeworx (2010)... 7 2.3 Other studies... 7 3. DISCUSSION OF ISN QUANTIFICATION OF HFT COSTS TO NON-HFT INVESTORS... 8 3.1 Introduction... 8 3.2 ISN Approach to Estimate Investors Costs... 9 4. CONCLUSION... 12 The costs of high frequency trading June 2013 www.industrysupernetwork.com 2
1. Introduction With High Frequency Trading increasing its presence in Australia public equity markets, it is important to understand its effects both on market integrity and on other investors. While research to date has addressed market integrity and stability, the critical issue of what HFT costs long-term investors has largely been ignored. This analysis is the first attempt (to our knowledge) to estimate some of the costs of HFT to long-term investors in the Australian public markets. The substantial increase in the speed of markets in recent years is neither an inevitable or necessary application of technology, 1 but it has changed the playing field of investing. The capacity of HFT to achieve faster relative speeds compared to other traders largely depends on market centres having established facilities and systems to accommodate them. Many long term investors believe this change in market structure to cater to speed affects investors trading costs (and other costs) in a number of ways. These effects are centred on the issue of relative speed. 2 The growing presence of HFT and the increasing advantage and importance of relative speed, results in three potential costs to non-hf traders: Direct losses on trades. HF traders combine special access to market data and market infrastructure with sophisticated algorithms to functionally predict price movements, and advantageously interpose themselves between natural buyers and sellers in the market, obtain priority in the limit-order book queues as the market moves, and pick off stale orders before other investors can revise them. These result in systematic costs to non-hft traders, measurable on a trade-by-trade basis, due to participating in a low latency market with HFT. 3, 4 Opportunity costs. Second, when an HF trader utilises relative speed advantages to obtain priority in the limit order book, this results in the HF trader interposing between a natural buyer and seller; the HF trader acts as a substitute for the natural buyer. This has consequences, including that the natural buyer does not have its order filled. The natural buyer has to find another seller. This is costly, and could often involve crossing the spread and accepting an inferior price. This phenomenon occurs, and costs are generated, when HF traders substitute for the natural buyer, and when HF traders substitute for the natural seller. Broader economic costs. Third, the advantages of speed are always relative, and as a result, continuous trading in low latency markets is not a stable equilibrium. The incentive to be faster than other traders never dissipates. This incentive structure manifests in an arm s race in low latency technology and 1 See, Industry Super Network, Toward A Fairer And More Efficient Share Market: Frequent Sealed Bid Call Auctions with Random Durations (2013). 2 Some academic literature classifies these costs as latency costs, i.e. the costs associated with the time delay between trading decision and trade execution. For discussion, see Moallemi, C. C., & Saglam, M. (2012). The Cost of Latency, Graduate School of Business, Columbia University, Working Paper. 3 See, e.g., Kirilenko, Kyle, Samadi, and Tuzun, The Flash Crash: The Impact of High Frequency Trading on an Electronic Market (2011) (finding that HFTs are able to buy right as the prices are about to increase. HFTs then turn around and begin selling 10 to 20 seconds after a price increase. ) 4 See, Baron, Brogaard, and Kirilenko, The Trading Profits of High Frequency Traders (2012) (analysing transaction-level data with user identifications (i.e., audit trail data) and finding HFT is systematically profitable when it interacts with other kinds of traders. See, Panel C, Table 3.) The costs of high frequency trading June 2013 www.industrysupernetwork.com 3
algorithmic game playing. This creates a broader economic cost of HFT as more and more resources are devoted to developing, executing and regulating the ensuing arm s race. 5 We do not attempt to estimate the first or the third kinds of costs, in large part because audit trail data is not publicly available. This analysis provides an estimate for the second cost the opportunity cost to other investors. We estimate the potential costs to long-term investors due to HFT activities to be approximately $1.6 billion per year, with a range up to $1.9 billion per year. This is likely to rise as HFT activities increase in Australia. We note that this estimate rests upon conservative assumptions and the real cost facing long-term investors could be larger if other costs, including opportunity costs that we have not sought to quantify, are considered. We stress that there is a significant lack of transparency on HFT activities, and that audit trail market data is not publicly available, which significantly affects the ability to estimate the subject opportunity costs with precision. Typically, high frequency traders earn small spreads for their trades. The HFT gains per share are measured in cents (or fractions of a cent in some markets). This may seem small, but the aggregate costs for longterm investors over time are sizeable, according to our calculations. We also note that, in a market where HFT is increasingly prevalent, there is a risk that limit orders of non- HFT investors will increasingly be filled only when high frequency traders do not want to execute at that price. This raises the fundamental question of what incentives exist for non-hft investors to continue to display interest when high frequency traders can interpose themselves. Many super funds are long-term investors. Industry super funds focus on deploying capital that supports sustainable economic growth and generates superior returns for over five million beneficiaries. Super funds rely on capital markets to help meet the retirement needs of working Australians. ISN has previously observed that HFT advantages arising from market structure and access are inconsistent with basic principles of fairness in securities regulation, namely the establishment of a level playing field in which sound capital allocation decisions and investment acumen are rewarded. This study provides additional support for our concerns. This study provides evidence that HFT activities, and the low latency market environments that to cater to HFT, are undermining not only the public policy objectives underlying market regulation, but also the public policy objectives of the superannuation system. 5 See, e.g., Biais, B., & Woolley, P. (2011). High Frequency Trading, (March). (finding that there is a significant risk that [HF traders] could, on balance, be detrimental to slow traders, in particular due to adverse selection. In that case, final investors and their investment managers will need to seek protection from HFT. To do so, buy-side firms will continue to develop their own trading algorithms, and thus participate in the arms race. (Note omitted).) The costs of high frequency trading June 2013 www.industrysupernetwork.com 4
2. Literature review There are few studies to date attempting to quantify the costs of HFT to long-term investors. Two relevant studies are outlined below. 2.1 Nanex Research (2012) 6 Data Sample characteristics are outlined below: Sample includes all stocks priced above $2 listed on NYSE, ARCA, AMEX, and Nasdaq (i.e. lit markets only). Period: 03 January 2006 25 July 2012. The study analyses all sub-penny trades that have a price expressed in more than two decimal places and that are not priced exactly at ½ cents. Market Structure Unique to the US equity markets is the Sub-Penny Rule, which allows traders and brokers to step ahead of the best bid-ask existing in the limit order book by quoting sub-penny pricing. Retail investors are unable to use this practice. Methodology Nanex collects all sub-penny transactions of all listed equities. Sub-penny trades are divided into bins, according to the last two digits, such that each bin represents potential price improvement relative to a decimal quote: o Eg: Bin 1 represents prices 0.0001 away from the nearest cent and so on. This represents the potential price improvements for non-hft investors. o There are 49 bins in total (because each direction from the nearest cent can represent that same price improvement, e.g.,.0099 and.0001 each are a.0001 potential price improvement). Nanex categorises the parties affected by each trade into one of three groups: o Group A: Non-HFT Investors whose orders are filled and thus receive price improvements from HFT. o Group B: High Frequency Traders who can quote up to four decimal points. o Group C: Non-HFT Investors, whose orders were stepped ahead of by HFT. This group is left with unfilled orders. Costs and benefits for each group of investors are calculated as follows: o Group A: price improvements for each bin. Price improvement * number of shares in each bin. Total price improvements are the total benefits across all bins. o Group C, whose order is unfilled, must cross the spread and the minimum loss for them is the minimum bid-ask spread (1 cent): Cost for C = 0.01 x no. of shares traded in each bin. 6 Nanex Research (2012). Sub-penny Price Anomaly, http://www.nanex.net/aqck2/3517.html. Accessed 01 May 2013. The costs of high frequency trading June 2013 www.industrysupernetwork.com 5
o o Total costs are sum across all bins. Group B: HFT gross profit: It is assumed that HFT profits are Investor C s loss, minus price improvements provided to Group A investors. Total costs and benefits are summed across all bins. Results The benefits for Group A investors (i.e., the price improvement) are very low compared to the profits of HFT and the losses to Group C investors. The opportunity costs for Group C investors are large since they have to cross the spread to get their trades executed. Figure 1 Profits and losses from sub-penny trades $8,000,000.00 $6,000,000.00 $4,000,000.00 $2,000,000.00 $- $(2,000,000.00) $(4,000,000.00) Investor A Investor B HFT $(6,000,000.00) $(8,000,000.00) $(10,000,000.00) Source: Nanex Research Comments The research provides an estimation of the transfer effects between investors and HFT. Focusing on only sub-penny trades, the analysis provides an estimation of HFT profits and costs to retail investors. The estimates are conservative, in our view, since sub-penny trades are only a subset of total trades. It is expected that the costs for retail investors due to HFT activities are far larger. The costs of high frequency trading June 2013 www.industrysupernetwork.com 6
2.2 Narang/Tradeworx (2010) 7 Background Tradeworx is an HFT firm. Their submission to the US Securities and Exchange Commission (SEC) in 2010 provided an insight into the profitability of HFT. Methodology Tradeworx estimated that HFT expected net profit is 0.1 cents per share traded. The before-cost expected return is estimated at 0.2 cents per share traded. Average HFT proportion of trading volume at the time of the submission was estimated at around 30-60% (Tradeworx s own estimation was 40%). Calculation to derive annual HFT profit is: (daily share turnover) x (HFT share of turnover) x 2 x (estimated cents-per-share returns) x (250 days) Results HFT net profits are estimated at US$2 billion per year (2010 estimates). 2.3 Other studies Moallemi, C. C., & Saglam, M. (2012) attempt to quantify latency costs incurred on a human trading timeframe using data from the New York Stock Exchange (NYSE) over time. They estimate that in very liquid stocks with a minimum bid-ask spread of 1 cent, the cost of latency is $0.0015 $0.0025 per share traded. This is comparable with the HFT profits estimated by Tradeworx. 8 Pragma studies describe the liquidity costs that long-term investors face in the presence of HFT. 9 Pragma s research found that: Even though there are more orders with attractive quotes posted by HFT, their order size is much smaller and likely to be inefficient to fill long-term investors orders. This potentially creates a queue in the limit-order book. Long-term investors either have to cross the spread to finalise their trades or they have to queue longer in the limit-order book, facing higher execution risk. Pragma s examples are hypothetical but they provide an insight into the problems HFT can cause long-term investors. 7 Narang, M. (2010). Tradeworx, Inc. Public Comment on SEC Market Structure Concept Release. http://www.sec.gov/comments/s7-02-10/s70210-129.pdf. Accessed 30 September 2012. 8 See, Moallemi, C. C., & Saglam, M. (2012). The Cost of Latency, Graduate School of Business, Columbia University, Working Paper. 9 See, Pragma (2012), HFT and the Hidden Cost of Deep Liquidity (pp. 1 6) and Pragma (2012), The Difficulty of Trading (pp. 1 4). The costs of high frequency trading June 2013 www.industrysupernetwork.com 7
3. Discussion of ISN quantification of HFT costs to non-hft investors 3.1 Introduction The Australian equity market is different in structure from the US or European equity markets, including that it is currently less fragmented. The emergence of low latency trading platforms is a fairly recent phenomenon, with Chi-X and Purematch (an ASX trading platform) entering in late 2011. Trading volume in ASX-listed names is concentrated on the top stocks in the S&P ASX 200 index. 10 Noting the sub penny rule is unique to the US, the methodology of the Nanex research is applicable in the Australian context. To do so, we consider the case where long-term investors take positions in the market by posting a limit order. The relative speed and latency advantages enjoyed by high frequency traders, together with an market system accommodative to fast algorithmic trading, helps them to (i) see and process the incoming order flow before other traders and (ii) post orders in the book, with a small price improvement, before other investors respond at the same price. Thus high frequency traders obtain priority over non-hft quotes lower in the book and attract market orders and liquidity while notionally providing it. This characterisation of certain HFT strategies is consistent with the empirical data released on HFT by ASIC in March 2013. 11 The non-hft investors, including long-term investors, facing execution risks, must cross the bid-ask spread and 12, 13, 14 transact at an inferior price. The diagram below illustrates this intuition. In the scenario without HFT activities, long-term investor orders are matched with natural buyers and sellers in the market (whether or not they cross the spread to execute an order). In the second scenario, HFT traders are introduced and force long-term investors to cross the spreads to get their orders filled. In doing so, they interpose between the natural buyers and sellers, create an additional cost for long-term investors and capture almost all of this cost as gross profit. 10 The S&P/ASX 200 index accounts for around 80% of Australian equity market capitalization and comprises of the most liquid stocks in the market. 11 ASIC found that 88% of HFT orders were at the best price, while HFT only provided 20% of the depth at the best price in the ASX200. Australian Securities and Investments Commission, Report 331: Dark liquidity and high-frequency trading (pp. 75, 77). 12 In their hypothetical example, Pragma also suggest that with HFT activities competing at the top of the order book, other traders have to cross the spreads more often to mitigate the risks that their orders are unfilled, See, Pragma. (2012) HFT and the Hidden Cost of Deep Liquidity (pp. 1 6). 13 We acknowledge that crossing the spread is only part of the potential costs that long-term investors face. With the time delay, it is possible that the market has moved against long-term investors. We have not sought to quantify this type of opportunity cost in this study. 14 HFT can provide price improvement in connection with obtaining order book priority, but as shown in our analysis and Nanex research (2012), such price improvements often are small in magnitude and are less than opportunity costs. The costs of high frequency trading June 2013 www.industrysupernetwork.com 8
Figure 2 Diagram of HFT interposition Limit Book Orders (Buy) Market Orders Limit Book Orders (Sell) Without HFT Buy Order at $10.30 $10.25 $10.20 Sell Orders at Market Price Buy Orders At Market Price Order crossing spread Sell Orders at $10.50 $10.55 $11.00 Commentary without HFT Buyers place buy orders in Limit Book and the best priced orders at $10.30 are filled by incoming Market Sell Orders. Sellers place sell orders in Limit Book and the best priced orders at $10.50 are filled by incoming Market Buy Orders. If a Buyer (Seller) crosses spread (shaded arrow), it is still matched to a natural counterparty posting a stable order in the book. With HFT With HFT HFT Order at $10.31 HFT Order at $10.49 $10.30 Orders no longer being filled from market orders cross to the other $10.50 side of the order book and incur $10.25 worse prices $10.55 HFT Order at $10.31 $10.25 HFT Order at $10.49 $10.30 $10.50 $10.55 Commentary with HFT HFT post orders fractionally faster than other market participants with small price improvement (1c=minimum tic size on ASX). Orders matching this price will be behind HFT order due to time priority rules. To execute orders, non-hft participants cross the spread buying and selling at worse prices. Buy Order Flow with HFT HFT sells at $10.49, 1c less than best price in without HFT. Market Orders save 1c per share. Orders crossing the spread incur a cost of 19c per share they buy at $10.49 not $10.30. With HFT HFT Order at $10.31 HFT Order at $10.49 $10.30 $10.50 $10.25 $10.55 Sell Order Flow with HFT HFT buys at $10.31, 1c more than best price without HFT. Market Orders save 1c per share. Orders crossing the spread incur a cost of 19c per share they sell at $10.31 not $10.50. With HFT HFT Order at $10.31 HFT Order at $10.49 $10.30 $10.50 $10.25 $10.55 HFT Gross Profit HFT buy at $10.49 from orders crossing spread HFT sell at $10.31 to orders crossing spread HFT gross profit is equal to the spread of 18c Legend: Buy Order Sell Order Order Crossing Spread 3.2 ISN Approach to Estimate Investors Costs Assumptions The market comprises of two categories of participants: HFT and non-hft (including long-term institutional and retail investors) who take position in the market by participating in the limit order book. The costs of high frequency trading June 2013 www.industrysupernetwork.com 9
We assume that HFT trading activities are 25% of total market volume, as a base case, recognising that the range of estimates vary. This assumption is consistent with ASIC descriptions. 15 Methodology Following Narang (2010), the annual HFT cost for non-hft investors is calculated as: (average daily share turnover) x (HFT share of turnover) x (estimated per share costs) x (250 days) We estimate per-share costs using historical closing bid-ask spreads, following the intuition that cost for non-hft investors is the magnitude of the spread (i.e. each market order filled by an HFT is an opportunity cost for the long-term investor whose limit order is not filled, which cost is equal to the spread). Our sample of shares is the constituents of S&P/ASX 200 index, and our sample period is from 1 September 2010 to 27 September 2012. Trading statistics Table 1 Sample statistics. Average daily volume all ASX200 stocks 703,176,467 Average daily volume < $2 stocks (more than 10c) 345,778,574 Average daily volume > $2 stocks 357,397,893 % $2 stocks volume/total 200 stocks volume 49% Source: ISN, Iress Estimated HFT costs using historical bid-ask spreads We look at the historical daily closing bid-ask spread 16 during the sample period to arrive at a limited, but realistic, cost estimation for long-term investors due to HFT activities. 17 15 There have been different estimates of HFT activity as a portion of market activity in Australia. According to TABB, HFT is estimated at 20% of market volume. Frino, Lepone and Mistry (2011) are reported to provide a much higher range from 30-80%. See, Psomadelis, W., & Powell, S. B., 2011, High frequency trading credible research tells the real story, Schroders Special Report. The Hon. Belinda Gibson, Deputy Chairman, Australian Securities and Investments Commission stated that HFT activities are around 25% of total trading. We use this as our base case. See http://asic.gov.au/asic/pdflib.nsf/lookupbyfilename/asics-focus-relatingto-markets--speech-to-fisd.pdf/$file/asics-focus-relating-to-markets--speech-to-fisd.pdf Australian Securities and Investments Commission s recent research included in Report 331: Dark liquidity and high-frequency trading, found that HFT constituted 27% of trading by value and 32% of traders. The report provides no figure for volume, i.e., total shares transacted. 16 Dollar bid-ask spread is calculated as closing ask price minus closing bid price as reported by Iress. 17 This analysis uses closing daily bid-ask spreads reported by Iress. We believe this approach is reasonable given the level of precision is limited by the lack of publicly available audit trail data. However, a complete analysis of bid-ask spreads throughout the day, or a number of snapshots of bid-ask spreads, each would result in a more accurate calculation. The costs of high frequency trading June 2013 www.industrysupernetwork.com 10
We calculate the daily dollar volume-weighted bid-ask spread for two groups of stocks (below and above $2) to ensure the measurements are not skewed toward less liquid stocks which typically have larger spreads. As depicted in the figures 3 and 4, and as one would expect, the level of historical bid-ask spread in the market exceeds the minimum tick size of 1 basis point for stocks worth more than $2, and 0.5 cents for stocks with prices from 10 cents to $2. Figure 3 Average daily closing dollar bid-ask spreads for shares priced above $2 0.1600 0.1400 0.1200 0.1000 0.0800 0.0600 0.0400 0.0200-01/09/2010 01/03/2011 01/09/2011 01/03/2012 01/09/2012 Min Case > $2 Spread > $2 stock Figure 4 Average daily closing dollar bid-ask spreads for shares priced at or below $2 0.0600 0.0500 0.0400 0.0300 0.0200 0.0100-01/09/2010 01/03/2011 01/09/2011 01/03/2012 01/09/2012 Spread < $2 stock Min Case < $2 Volume Weighted spread < $2 Source: ISN, Iress The costs of high frequency trading June 2013 www.industrysupernetwork.com 11
In this scenario, we assume that the cost per share for investors is the daily average bid-ask spread. Essentially: 18 CPS for $2 stocks $ 0.014 CPS for > $2 stocks $ 0.059 Applying the above costs to the cost formula produces the results set out in Table 2: 19 Table 2 Range of HFT costs to non-hft investors Market Share of HFT 25% 30% < $2 stocks $310,525,940 $372,631,128 > $2 stocks $1,315,837,167 $1,579,004,600 Total estimated costs $1,626,363,107 $1,951,635,728 Source: ISN estimates 4. Conclusion ISN estimates that HFT activities cost non-hft market participants, including Australian long-term investors such as super funds, up to $1.9 billion per year, with a best estimate of over $1.6 billion per year. This estimate would enlarge if HFT activities increase. ISN notes that these estimates are based on conservative assumptions. We believe that the costs for long-term investors could be much larger if, e.g., opportunity costs such as adverse price movement were included. 20 Costs arising from delays in execution due to a lower position in order book queues also are not included. Less direct costs, such as those arising from (i) developing and implementing enhanced regulatory surveillance, (ii) cost passed on by brokers for compliance and other systems, and (iii) data management, storage and analysis resulting from the increased trading and quotation activity common to low latency markets with HFT also are not considered. 18 These numbers are higher than the latency costs of $0.0015-$0.0025 estimated in Moallemi, C. C., & Saglam, M. (2012). However, this is to be expected given the differences between the NYSE and the ASX, and the nature and degree of HFT activity in each market. We believe that our CPS numbers are reasonable in light of these differences. 19 (average daily share turnover) x (HFT share of turnover) x (estimated per share costs) x (250 days) 20 In addition, HFT creates a potential lemon problem in that HFT strategies could enjoy structural informational advantages (and the incentives to exploit them), resulting in uncertainty for non-hft counterparties. In the face of asymmetric information risks, non-hft counterparties may need to assume the information that they do not have is adverse. This can result in market prices necessarily deviating from fundamental value. In theory and at the extreme, it can even result in a broken market, in which equilibrium involves no exchange. Cf., Akerlof, G. (1970). The costs of high frequency trading June 2013 www.industrysupernetwork.com 12