High Frequency Trading: a Big Data application a) High Frequency Trading reviewed High Frequency Trading (HFT) is currently available on forty electronic markets across six continents. HFTis used for trades and arbitrage in execution of a wide range of strategies, across a broad range of instruments including equities, foreign exchange and derivatives. A recent study by Morgan Stanley and Olive Wyman indicates that approximately 70% of global equities trading is executed by machines without human intervention.
The market share of HFT in the US was 36% for all types of instruments (55% for equities) in 2010/11 (FSOC 2012), Europe was estimated at 30-50% (FTSE Global Markets 2011), and in Australia as recently reported (ASIC 2013) it accounted for 22% of total equity market turnover. ASIC also reported that dark trading represented 25-30% of trading with 7% of that total in HFT form. HFT has different definitions dependent on market, instruments or strategy so care is needed when dealing with trend data.As a consequence ASIC in REP 331 (2013) defined in excess of 100 terms as applied to their research into this form of trading. There is often confusion between electronic trading, algorithmic trading and high frequency trading (as a combination of both). The most generally accepted description of this particular form of trading is given by IOSCO with the characteristics of: - HFT is a type of algorithmic trading but not all forms of algorithmic trading are high frequency - sophisticated technological tools are required - it is highly quantitative - high daily portfolio turnover and order to trade ratio usually involves flat positions at the end of the trading day (although long positions are not unknown) - it is latency sensitive (time taken from order placement to response with 34 microseconds reportedly the fastest), driving requirements such as direct electronic access and co-location.
It is claimed for current systems 100,000 orders can be processed per second, 500 million per day. The present target of advancing speeds to nanosecond (billionths of a second), it is estimated, could generate an extra $100million per year in earnings on global trades).In practice HFT strategies typically cluster around three specific objectives: low latency ie the critical speed factor; market making that exploits price discrepancies and is rewarded for posting liquidity and, statistical arbitrage (FTSE Global Markets 2011). Serious investments in various forms are required to achieve chosen strategies.
Stock Exchanges are required to make the investment in data storage, processing and communications to address the issues of latency and equality. Gore Hill cost the ASX $37 million with cabinet rentals costing client traders $2500 per month.There are already plans to extend the use of these facilities into Cloud Computing as part of a network of global data centres. Traders need to develop algorithms capable of not only executing commercially beneficial patterns of trade but also responding to changing patterns of information while minimising risks. The life span of an algorithm is reported to be as low as 14 days given the competitive countervailing measures.
Individual algorithms can cost from $10,000 to $1 million to develop. The forecast expenditure on algorithms for 2013 is $51 billion (Hughes-Liley, 2012).Costs will rise if further development results in the replacement of software by a hardware medium. The medium affects speeds and costs.
Algorithms are already being customised to both specific buyer and market conditions, including current difficulties of low volumes etc. The next generation of trading algorithms will be made up of: trading envelopes; execution constraints, and liquidity seeking tactics. Data for these advanced algorithms is already available (see below) Back office routines and systems will require upgrades.Issues arising from the data gathering and administrative aspects of the system include data volumes, speeds, standardisation, and security.
The costs arise due to both establishing the capability and its on-going maintenance and storage. The storage aspects link to cloud computing developments with the latter having it’s own energy issues. Associated with HFT patterns of trading are both positive and negative outcomes. Higher speed automatic trades have serviced smaller trading blocks, at lower prices and with reduced spreads.Greater latency supports a broader range of strategies with both advantages (more completions, reduced risk of signalling intentions, greater liquidity - but this is also contested) and disadvantages (iceberg style trading, anti-competitive advantage, system ‘noise’, fragmentation and decrease of market breadth, front running, stuffing).
Overall the most prominent risks involve ‘flash crashes due to the speed of the loss of control with attendant impacts. To date the most significant losses of value in global exchanges have all been in the era of face-to-face trading.The range of causal events leading to a sudden crash have increased and can be either endogenous (algorithm errors eg May 2010 flash crash, fat finger mistakes eg FKP Property Group, 2012) or exogenous (Twitter/Syrian Electronic Army, 2013) . The means of managing this form of risk inherently requires greater cooperation among the various institutions including regulators that constitute the full value chain.
In turn, these developments will undoubtedly lead to a requirement for more disclosure and even further regulations.The purpose of changes to the market regulations by ASIC since 2010 have followed the core theme of encouraging market developments that promote fair and transparent markets and facilitate HFT. The formal Australian exchanges face an environment of increased competition, significant capital investment, lower volumes, reduced spreads and lower liquidity and the emergence of linked pools that are currently unregulated. Whether the balance of outcomes from the major changes introduced in the last decade are judged positive or negative there is no doubt that the arket performance is under greater scrutiny than before even to the extent of challenging the basic focus on market transactions as opposed to fundamental services ( Kay Review, 2012). The reaction of the UK Government to the report was that it endorsed the 10 principles, and agreed with most of the 17 specific recommendations and fully supported the Statement of Good Practices. The Government is not, however, proposing any further regulation preferring instead to promote a change of culture through leadership by market participants.
One practical outcome is that negotiations have commenced with the European Commission to discontinue quarterly reporting. The general argument regarding the future role of Exchanges revolves around the inconsistency of high speed trading with patient investing (Clarke T, 2012). In other words long-term investment performance requires attention to more than short-term financial metrics. b) Big Data and pattern developments Just as HFT has no definitive definition neither has Big Data.Its characteristics, however, include, high volumes of data (sufficient data that conventional computers are inadequate, requiring parallel processing to cope).
A terabyte of data per day ie 1012 would qualify but smaller data streams involving high speeds or standard & non-standard varieties of data may also qualify dependent on strategic impact and the value created. Big Data and associated Pattern Analysis (BDPA) are two components of an interactive string, or value chain that at its simplest can be described as: 1)Initial data – composite data – pattern analysis – application (eg HFT or whatever) Just as with HFT so with BDPA where the associated pattern analysis is a type of algorithmic modelling but not all forms of algorithmic modelling result in either pattern analysis or are dependent on information resources that correspond to big data. In the future ‘Big Data’ will be increasingly associated with the processing of non-standard information as for example footage taken from security cameras. In the realms of zettebytes (1021) it is believed that upto 90% of data will be non-standard.BDPA has many existing applications ranging through, health, security, meteorology, election forecasts, cosmology and military applications. The information that comprises the initial data whether standardised or non-standardised needs to be converted to a form that makes it accessible to machine languages ie can be represented in ‘0’ & ‘1’.
The existing sources of this data include objective mathematical, scientific, financial information and data streams from new sources such as the ‘internet of things’.Non-standard data is initially in other formats for example subjective as in words, photos, sounds etc. Data assembly including issues of quality, security, provenance, replicability, and auditability are major issues. Timely, standardised data of a given quality is both an expensive asset to acquire and then maintain with a danger of toxicity from uncleansed data.
Manipulation of data into meta-metrics (Stefani A,Xenos M, 2009), suitable for pattern analysis provides the basis for mathematical, sociological or some combination of the two types of analysis (patterned).The application objectives will be specific but the duplication of human intuition type processes at high speed is challenging irrespective of the nature of the data. Even then the requirement to explain the results and allow for critical analysis of the deliverables is a further challenge as was proven by the experience with models for the securitisation of debt involving secondary mortgages. It is early days but all of the above data related issues and more are being addressed by participants such as IBM, Oracle, SAP, HP(Autonomy), GE, McKinsey, Intel and others including the universities and CSIRO.
Accompanying these commercial developments other research is also making a contribution drawing on advances in algorithm design and application of Artificial Intelligence, cloud computing and other forms of basic investments such as NBN. c) HFT as a BDPA Given the nature of the trends in financial markets and associated developments of the underlying technology, the current pattern of shrinking volumes, increasing participation of electronic processing with its orresponding reduction in spreads, reduced latency, high and continuing levels of investment and shortages of experienced and skilled personnel will exert pressure on the status quo. In this environment the fundamental role of equity trading markets will be examined. Given that intermediation is not an end in itself the objectives of any such review will include the changes required to make the market more responsive to its fundamental purposes of providing capital for investment by companies generating value added by their output of products and services, both short and long-term.
The Kay report suggests that ‘stewardship’ will become an increasingly important aspect of the future direction. For that to become a reality there will need to be a much stronger and regular flow of information between Corporations and asset managers or similar. Some of that information will be objective, other aspects more subjective as in the assessment of the quality of Boards. The number and types of sources of information to be scoured and correlated will be increased.
The evolution of the acquisition practices and application patterns of information is consistent with the way all of us, in our use of the internet, are already being assessed through our patterns of key strokes, sites visited, information sought, purchases made. As the mobile communications industry says, “when something on line is free you’re not the customer, you’re the product”. At UTS the University is collecting information about student’s online activites and backgrounds, helping to predict those most likely to fail their courses (PreissB, April 2013). Perhaps that is not the way that it will all work out.Maybe the gaming of the market will become the primary activity supporting trading as a secondary feature ( Tobais P, et al, 2013‘).
Their research was based on an analysis of Google Trends data, based on the frequency of the mention of a number of terms by comparison to the performance of the DJIA index. They identified acceptable correlations originating with subjective and objective data. Reference was made earlier to the evolving structure of algorithms accompanied by the comment that appropriate data was available. Just as in the case of market gaming, data availability is a critical variable.
In this regard the Thomson-Reuters service of machine-readable news feeds that is based on the search for usage of terms such as ‘buy’, ‘earnings’, or ‘position’ from 50,000 new sites and 4million social media sites is as significant as the other stages in the process. Aite Group estimate that 35% of quantitative firms use some form of machine readable news feeds, for example, Vinva’s Quant Funds. It is now very clear that the full value resulting from any future direction that the financial markets take will be as a member of a value chain rather than within strategic boundaries of its own choosing.There remains the choice and prioritisation of the epistemology but not the ontology (as defined for information science), which is clearly grounded in computer technology. It may be stretching the argument too far too soon regarding the epistemology but in a value chain with a focus on efficient long-term investment, it could be that the role of exchanges may be likened to a gambling house, providing the chips and the tables while the serious work of allocating scarce “at risk” capital is conducted elsewhere machine to machine with minimal regulation consistent with a free capital market philosophy.
) References ASIC, 2013, ‘ASIC reports on dark liquidity and high frequency trading’, ASIC REP331 Clarke T, 2012,’Could high frequency trading lead to our own ‘flash crash’? , The Conversation, 17/09/12 FSOC, 2012 Financial Stability Oversight Council, Annual Report 2012, pp86-89 FTSE, Global Markets’, 2012,High Frequency Trading, when to call time on HFT’, The 2012 Global Trading Handbook,p51. Hughes-Liley R,2012,’Can algorithms really think like humans’, FTSE GlobalMarkets’, July/August 2012, p63. IOSCO, International Organisation of Securities Commission Kay J, 2012, ‘The Kay Review of UK Equity Markets and Long-term Decision Making’, Interim Report, Department for Business, Innovation and Skills. Preiss B,2013, ‘Students online habits tracked to predict success’, SMH 14/04/13 Stefani A, Xenos M, 2009, ‘Meta-metric Evaluation of E-Commerce – related Metrics’, Electronic notes in theoretical computer science 233, 2009, pp’s 59-72.
