I Introduction
High frequency trading (HFT) mobilises special hardware and software to send orders at extremely low latencies. Being first in line is inherent to finance, but in the last decade the confluence of trends in computer hardware, programming, mathematical modelling, and financial innovation pushed the limits of trading speed to unprecedented levels and widened the gap between humans and algorithms. In the blink of an eye, the fastest matching engine can process around 10,000,000 orders. Such speed and massive data management allow HFT to operate with a few hundred financial instruments on several trading venues at the same time. Consequently, a small blip of one algorithm can lead to a systemic crash in a matter of minutes. For example, on 6 May 2010, a single algorithm precipitated the loss of $1 trillion of the US market capitalisation in a 10-minute-long cascade of falling prices. More astoundingly, without any human intervention the markets recovered to the previous levels in the following 15 minutes. The event earned the name FlashCrash. HFT was quickly in the spotlight. Some blamed it as the root cause of the crash and called for its prohibition. Others expressed a more mitigated view that HFT played a key role only insofar as it spread the crash. All in all, nobody was indifferent about algorithms wiping out an individual’s savings; HFT and algorithmic trading at large was ripe for regulatory intervention.
Soon after, the G20 Seoul Global Summit urged national and regional regulators to “mitigate the risks posed to the financial system by the latest technological developments”. The International Organization of Securities Commissions (IOSCO) issued a report in 2011 with guidelines on how to treat the new risks, including pre-deployment testing, real-time monitoring of algorithms and more extensive use of automated norms such as circuit breakers.Footnote 1
By November 2010, the US financial regulatory agencies the Securities and Exchange Commission (SEC) and the Commodity Futures Trading Commission (CFTC) published a joint report describing the sequence of events in the FlashCrash 2010. But given the lack of, or difficult access to, ex post market data, it took five years to identify the algorithm, directed by a solo trader based in the UK and unregistered with the authorities. Interestingly, the trading algorithm’s manipulative activity was detected and proved using in large part market data.Footnote 2 Today, the complexity of the life cycle of orders is the new opacity. Indeed, a more global systemic approach based on transparency was needed to regulate the now well-integrated EU financial system.
The objective of this report is to briefly review and comment on the recent attempt by the EU to adapt to the risks of the modern algorithmic and HFT paradigm through Mifid 2 DirectiveFootnote 3 and the new Market Abuse Regime.Footnote 4 To that effect, we will briefly review the evolution of algorithmic trading to identify the major problems that are a consequence of automatisation of trading and interconnectedness of algorithms (II). In the third part, we will describe and analyse the new European Regime regarding HFT in light of three legal issues: access to markets, monitoring of algorithms, and prohibition of market manipulation (III). In the fourth and concluding part, we will offer a preliminary assessment of the European Regime (IV).
II Evolution of high frequency trading: from a niche to a global phenomenon
The miracles of HFT are far from new and were developed alongside the digitalisation of trading venues. First, Nasdaq offered electronic quotation in 1971, followed by the CAC in Paris or London Stock Exchange during the 1980s, but the orders were still sent and executed manually.
With the introduction of the SOES system, traders could also send their orders electronically while the Nasdaq dealers were manually refreshing their prices. The “SOES bandits”Footnote 5 started engaging in algorithmic price arbitrage between faster and slower human dealers. The “101 economics”Footnote 6 of fast computer traders inevitably pushed everyone else into the new, algorithmic realms of the financial speed race. By late 1990s, most public exchanges also offered automatic execution of orders, in addition to automatic quotation and order transmission, thus offering traders a wholly automated trading process.
Furthermore, Mifid 1Footnote 7 offered a special legal regime for the alternative trading venues with many pre-trade transparency waivers for “dark pools”,Footnote 8 fostering their proliferation. Alternative trading venues, copying the business model of Island and Instinet, offer automatic execution of orders inside their electronic communication networks before they are publicly displayed on the exchanges. Such “dark execution” avoids exchange fees and lowers trading costs for large block trading. Mifid 1 also denationalised European public exchanges and eliminated concentration rules.Footnote 9 The entire life cycle of orders is now fully automated and the algorithmic traders can devise numerous strategies combining light and dark execution of several thousand financial products on several dozen trading venues in practically every country of the world. To accommodate such light fast trading, new optic fiber connections were laid under the Atlantic, lines of microwave towers were built from New York to Chicago and from London to Frankfurt, and new satellites were sent into space to cut latencies with Asian markets to a few milliseconds.
Still embracing the old human-based trading paradigm, Mifid 1 did not specifically tackle HFT. While most HFT were proprietary traders, viz traders risking only their own capital, Mifid 1, Article 4(d) specifically excluded them from its scope. But by late 2000s, a series of “disruptive events” revealed the systemic importance of algorithmic and HFT and the need to regulate the automated and interconnected markets they created. Soon after the 2010 FlashCrash, investigations showed that some exchanges and information providers were secretly selling early access to market-moving data. In 2012, algorithms crashed the initial public offerings of BATS Exchange and Facebook, Knight Capital Group’s algorithm lost $440 million in 30 minutes, and Nasdaq’s servers broke down on several occasions. Algorithmic orders were described as “flickering”, “scintillating”, or “toxic liquidity” as they were appearing and disappearing like sparks on the screen, preventing humans from matching them.
At the same time, many market participants and a growing economic literature hailed the manifold benefits of HFT: bigger trading volumes signalling more liquidity, better and faster execution, lower bid-ask spreads, etc. While algorithmic risks were on the rise, outright prohibition of HFT did not seem optimal. Rather, the regulatory objective was to adapt the market structure to the new technology-related systemic risks.
III The European regulatory answer: Mifid 2 and MAR
Mifid 2 is mainly a principles-based systemic regulation, with a clear objective to instill more transparency through harmonised EU rules on the financial market structure. The new Market Abuse Regime consists of the Market Abuse Regulation and an accompanying Directive on criminal sanctions. Both follow the Lamfalussy process, a legislative process used by the EU in the domain of financial services, developed in 2001 and based on the idea that there are four levels of creation and enforcement of rules. Level 1 measures are issued by the EU regulators in the form of a regulation or a directive. Upon consulting the European Securities Markets Agency (ESMA), the EU Commission issues Level 2 measures in the form of RTS/ITS to further specify their implementation.Footnote 10 Level 3 measures, guidelines and interpretations, are adopted by ESMA. Level 4 measures can result in the Commission’s actions for non-compliance against Member States.
We classify the new provisions into three categories of legal problems. In the first part, we will review obligations regarding how traders access markets (1). In the second part, we will review requirements relating to the monitoring of the algorithmic and HFT activities on the markets (2). In the third part, we will review new rules concerning market manipulation (3).
1 Regulating access to markets
We define market access broadly to include obligations concerning authorisation by regulators, and also those concerning how trading firms interact with their trading venues, viz co-location, fee structures, minimum tick size, order to trade ratio, and market-making obligations.
Before accessing markets, all HFT firms must be authorised by the regulators according to Mifid 2, Article 2. A precise definition of HFT was therefore necessary. As per Mifid 2, Article 4(14) high frequency trading:
− uses special infrastructure to minimise latencies;
− automatically handles orders without human intervention;
− exhibits high intraday rates of messages, viz orders and cancellations.
In comparison to other forms of algorithmic trading, HFT automates all stages of order generation, from initiation to execution. While most agree that HFT is “[by] any measure […] a dominant component of the current market structure”Footnote 11 its precise scope can vary significantly, depending on the quantitative definition of the “high intraday rates of messages”. In ESMA’s 2014 study the lowest estimates of HFT presence in the EU were set at 24% and the highest at 76%.Footnote 12 In fine, the EU Commission defined HFT as trading that exhibits message rates of more than 2 per second per financial instrument or 4 messages per all instruments on a given trading venue.
Authorisation requires a detailed description of how algorithms work and who writes, operates, and controls them. The algorithmic code will not have to be deposited but its activity will be followed through a unique Legal Entity Identifier (“LEI”). LEI is a world-wide initiative based on the ISO 17442 standard that will allow tracking traders all around the globe.Footnote 13
Mifid 2 also regulates co-location, ie the way traders physically connect to trading venues. HFT uses co-location to get early access to market data, so other market participants claimed the practice gives HFT traders an illegitimate advantage. Article 48(8) allows co-location agreements for different fees, but they must be transparent, fair, and non-discriminatory. Trading venues will have to publish co-location offers on their websites and will also be responsible in case of outsourcing.
Many smaller traders do not pay for co-location but connect through a Direct Electronic Access (DEA) provider. DEA providers will have to put in place appropriate controls for algorithms and will be responsible for their clients’ activities.
Mifid 2 also regulates fee structures of trading venues, two of which are closely associated with HFT. Article 27(2) prohibits payments for order flow when they infringe on the conflict of interest requirements. The orders HFT collect around the markets might be directed to venues offering the largest rebates to HFT rather than to those offering best execution to the end clients. The second fee structure associated with HFT are maker-taker fees, whereby the passive limit orders that wait on the order book and make the market for others are paid a fee, while the aggressive orders that are almost immediately executed and take the market are charged a fee. While this attracts market-making orders from HFT, it can obscure the calculation of trading costs. Article 48(9) does not prohibit any other fee structure inasmuch as they are based on “non-discriminatory, measurable and objective parameters” and published on venues’ websites.
The speed of trading also depends on the minimum tick size, which is the smallest price increment allowed for a financial product. Article 48(6) endeavours to harmonise tick sizes, because if too small, they might encourage unnecessary volatility by HFT. Trading venues must enforce minimum tick size regimes provided by ESMA.
Trading venues will also have to impose maximum order to trade ratios, viz the maximum ratio of all sent orders relative to the actually executed trades.
Deficient market-making exacerbated the 2010 FlashCrash and many HFT firms are de facto market-makers but were never obliged to register as such. Mifid 2 defines market-making as posting “firm, simultaneous two-way quotes of comparable size and at competitive prices relating to one or more financial instruments on a single trading venue or across different trading venues, with the result of providing liquidity on a regular and frequent basis to the overall market for at least 50% of the daily trading hours”. Those who exhibit such trading will have to sign a special agreement with their trading venues to continuously provide liquidity even in stressed market conditions.
2 Regulating monitoring of algorithms
Algorithms will be closely monitored and back-checked on several levels, before deployment and in real-time.
Internally, Article 17 (investment firms) and Article 48 (trading venues) impose a compliance function. Responsibility for disruptive events can now be attributed to all relevant persons at all hierarchical levels, from the developer to the trader, the trading supervisor, as well as compliance and the firm’s management. Trading venues will have to provide testing platforms with historical data to ensure resistance of trading and order matching algorithms under stressed market conditions.
Control will also be imposed on several levels of the life cycle of orders, from the investment firm, to the direct electronic market access provider (DEA), and the trading venue. Authorised market participants can be held responsible for their clients’ activity. Algorithmic orders will be flagged and “unusual orders” must be blocked before being sent to the venue by pre-trade controls with predetermined thresholds in terms of prices and volumes. When trading activity seems unusual, the firm and the venue should be able to either stop an algorithm with a kill function or stop overall trading in a given instrument by a circuit breaker.
To allow ex post reconstruction of trading activity by regulators, firms and venues throughout Europe must synchronise their clocks, time-stamp trading activity with a 100-microsecond accuracy, and store data for five years. In case of cyber-attack or physical breach of their systems, trading venues must ensure continuity of their business arrangements and maintain data losses close to zero.
3 Redefining market manipulation provisions in light of HFT
Mifid 2 does not itself tackle market manipulation but refers to the separate EU Market Abuse Regime.Footnote 14 The definitions of market abuse took the form of a regulation, having direct effect on the EU Member States because differing interpretations of the previous 2003 Market Abuse Directive provided for regulatory loopholes. Broadly speaking, manipulation consists of artificially influencing the prices with the objective to secure a profit. Attempting to manipulate is also prohibited. Article 12(1)(a) and (b) are the general anti-manipulation provisions that distinguish between: (a) the open market manipulation; and (b) the use of fictitious devices or contrivances. Open market manipulation consists of prima facie legitimate orders or transactions but that are only intended to influence the price. Fictitious devices mostly involve out-of-market actions to influence a price, such as disseminating false information on the internet. These general provisions are illustrated by Article 12(2) through a list of manipulative behaviours. Annex I gives a further list of indicators that in themselves do not constitute manipulation. The list is supplemented by the Commission Delegated Regulation 2016/522, which provides a longer, albeit non-exhaustive list of indicators.
MAR targets HFT and algorithmic trading in Article 12(2)(c) and prohibits sending orders that would disrupt or delay the functioning of trading venues’ servers, make it difficult to identify genuine orders, and create or likely create misleading signals about the supply and demand of a financial product. The indicators include the extent to which trader’s orders represent a significant proportion of trading activity, the extent to which orders are concentrated in a short time-span, position reversals, the extent to which orders change the order book, and whether orders are sent at a specific time in the trading session. Some provisions specifically mention spoofing, layering, and quote stuffing, which are trading patterns that have been closely associated with HFT.
When investment firms or trading venues detect suspicious trading activity they must immediately send a Suspicious Order and Transaction Report to their regulator. Some such trading might still be exempted from sanctions if it figures in the list of accepted market practices or if the trader proves the orders were part of a legitimate activity.
IV Early assessment of the new European legal framework for algorithmic and high frequency trading
The new European legal framework for algorithmic and HFT is a welcome initiative as the world’s first and most comprehensive set of rules to tackle the risks of the modern trading paradigm. With a systemic and principles-based approach, it brings all HFT and algorithmic traders back on the regulatory radar and offers regulators a wide arsenal of prohibitions.
Systemic approach is necessary because of the global and interconnected nature of algorithmic trading. Regulators will be able to track trading strategies and their effects in the entire EU financial system. It is less clear why Mifid 2 created a special regime for HFT. The evolution of trading shows that a large majority of traders now de facto use HFT-like strategies. In a matter of years, possibly all traders will fall under this regime.Footnote 15
Because every order has at least an indirect effect on the whole system that can hardly be defined ex ante, the principles-based approachFootnote 16 is also very welcome. The principles, unlike rules, prohibit undesired results and effects of algorithmic activity rather than specific ways to arrive there, like the obligation to prevent “disruptive trading” or the obligation to “maintain orderly markets”. They are particularly useful with regard to artificial intelligence algorithms that find different ways to arrive at the same result.
However, while algorithms operate with precise quantitative data, the prohibited effects are expressed in broad qualitative terms. Examples include “extreme volatility”, “large/small orders”, “significant effects”, etc. Such regulatory practice allows for substantially differing interpretations and has already revealed some drawbacks, and has already spurred some resistance. For example, Virtu claimed against AMF in France that the anti-manipulation provisions similar to MAR were unconstitutionally vague.Footnote 17
We therefore expect an ensuing cooperation between the regulators and all market participants, in particular their compliance departments. Market “gatekeepers” like trading venues, DEA providers, or firms selling algorithms to detect manipulation, will be heavily involved in the determination of rules. Level 3 Lamfalussy measures and constant communication by and with regulators will be very welcome in order to avoid legal uncertainty and over-enforcement. Total transparency fostered by the EU still requires drawing the fine line of legality, and trust in regulators not to misuse the data collected.
Furthermore, cooperation with regulators is important because Mifid 2 offers a wide(r) arsenal of prohibitions. HFT-related strategies are particularly complex and sometimes it is not clear which prohibition they fall under, or whether they break several of them at the same time. For example, AMF sanctioned the Euronext Paris exchange for secretly offering, as part of a test of a new business model, an unlimited order to trade ratio free of charge to an HFT market-maker, Virtu. This allowed Virtu to send a large amount of orders that misrepresented the state of the order book on five different trading venues in Europe, and to cancel them up to five milliseconds before others could possibly match them. AMF sanctioned Euronext for not keeping “orderly markets” by violating the requirement of impartiality in offering a commercial advantage to Virtu. Within Mifid 2, the strategy might also violate requirements regarding co-location, fee structure, order to trade ratio, market-making, monitoring of algorithms, and detection of market manipulation at the same time.
Surely, the compliance costs will significantly increase, which might favour larger firms with more resources. An interesting proposition to frame the cooperation, reduce compliance costs, and increase trust is the idea of regulatory sandbox, already implemented in the UK.Footnote 18 With the accord and strict supervision by the regulators, firms can test in the real markets their business models that “do not easily fit into the existing regulatory framework”. Regulators can offer restricted authorisation, individual guidance, waivers and no enforcement action letters but can also prohibit the activity at any later time.
Though paved with good intentions, the road to harmonised rules in Europe is not finished. The technical standards and interpretations that will emerge from private and regulatory practice will materially fill the under-defined provisions of the principles-based approach. While Mifid 2 and Market Abuse Regime are a great leap forward, we must be careful in our next few steps toward concrete implementation. Some already sense Mifid 3 around the corner.
