Why fixing algos will only cause trouble

Author: | Published: 29 Aug 2012
Email a friend

Please enter a maximum of 5 recipients. Use ; to separate more than one email address.

By journalists in London, New York and Hong Kong

Investors in US equity markets have seen three high-profile systemic errors this year, two of which involved high frequency trading (HFT) firms praised for developing some of the fastest, most efficient, liquidity-enhancing algorithms. The impact of these errors was limited by regulations implemented in the wake of the May 2010 Flash Crash – when the Dow Jones plunged and recovered 1000 points over 20 minutes.

However, not surprisingly, there are now stronger calls to prevent errors in the first place. Commentators suggest that exchanges and firms have become overconfident in their systems, hastily launching new algorithms without contingency plans in an attempt to beat competitors. Regulators across the globe are scrambling for solutions – considering transaction taxes, speed restrictions and post-trade monitoring. But it seems the solution could be as simple as stronger risk management.

In March, Bats Global Markets cancelled its initial public offering (IPO) on its own HFT platform after a software glitch. Two months later, Facebook's botched launch on Nasdaq occurred.

The most recent trading glitch to catch the world's attention took place on August 1 when Knight Capital launched a new algorithm designed to lower prices for retail investors. This was in preparation for a new New York Stock Exchange (NYSE) retail liquidity plan.

Things went awry almost immediately as Knight executed wild trades for 45 minutes, triggering six of the Securities & Exchange Commission's (SEC) single stock circuit breakers. The trades resulted in a $440 million trading loss for the market-maker, affirmed after the SEC denied Knight's request to retroactively block the bad trades. HFT firms, lawyers and academics were all astonished Knight did not respond to the error sooner.

"They turned on the car, let it drive without a driver down the street and went back into the house," Haim Bodek, founder and former CEO of Trading Machines, says of Knight's new system.

Part of the problem is thought to result from an inability, or unwillingness, of exchanges and firms to retain talented programmers to oversee new systems. Regulators should bear in mind the strong human element in this year's high-profile trade errors.

During the so-called Bats Crash, the exchange was without its most senior systems developer, Paul Rose, when a bad algorithm forced the HFT platform to pull its IPO in March. Rose had left Bats to join Tradebot Systems a few months earlier.

Facebook's IPO also suffered from a depletion in human capital. IFLR has been told that a key programmer was not present for the erratic IPO.

HFT in US stocks has increased drastically over the past five years Courtesy of Nanex

Speed bumps and risk controls

Irene Aldridge, Able Alpha Trading

Shortly after Knight's glitch, the SEC began an investigation into the firm's risk management procedures. The Commission's market access rule, finalised in November 2010 and effective from last year, requires broker-dealers like Knight to maintain reasonably designed risk management controls and supervisory procedures. But it has its shortcomings.

"Unfortunately, it's all very vague [and] it's very subjective," says Irene Aldridge, managing partner of research, development and implementation of high-frequency algorithms at Able Alpha Trading. "There are no mandates to what kind of risk management a firm has to have in order to do this kind of work."

That could change. According to an August 3 SEC release, Chairman Mary Schapiro has asked the Commission's staff to hurry-up a rule proposal that would require exchanges and other market centres to have specific programmes in place to ensure the capacity and integrity of their systems. Matt Andresen, co-CEO at quantitative trading firm Headlands Technologies, says common best practices include systemic testing of orders sent to duplicative test environments, pre-trade risk checks done for all orders to be sent by a new algorithm, looping checks in which firms monitor for suspicious and unexpected systemic behaviour, market data health checks to make sure data is flowing in properly, and kill switches to stop errant trading.

"If you don't design those speed bumps and risk controls, if you don't have those operations then you will probably have an event like this at some point in time," Bodek says. "The number of times risk controls intercept these events must be enormous – people screw up all the time."

The SEC has also recently backed an initiative to introduce greater certainty into the US market by limiting the number of cancelled trades.

European fragmentation

Dr Andreas Wieland, Shearman & Sterling

How US markets were structured had a considerable impact on its vulnerability to trading glitches. The trade-through rule fostered inter-linkages between US trading venues. Consequently, whenever there was an erroneous series of trades caused by rogue algorithms it could affect the entire market. This structure is thought to have contributed to the May 2010 Flash Crash.

Shearman & Sterling's Dr Andreas Wieland says Europe's more fragmented trading venues make it much less vulnerable to the trading errors that spell disaster in US' highly interconnected markets."

But Europe is also introducing measures to curb HFT risks. The Markets in Financial Instruments Directive (Mifid II) will introduce a licensing requirement for HFT firms and a specific regulatory framework for algorithmic trading activities.

Germany, however, has pre-empted the proposed EU-wide reform with its own legislation. On July 30, the German Ministry of Finance published a draft Act for the Prevention of Risks and the Abuse of High Frequency Trading.

The new rules could slow down HFTs on the country's regulated markets and multilateral trading facilities. But lawyers in Frankfurt warn that closer HFT supervision may not prevent another Knight Capital-like debacle.

The Act addresses many of the regulatory methods envisaged under Mifid II and those addressed in Iosco's July 2011 report on the impact of technological changes on market integrity and efficiency. The reforms aim to better establish systemic stability in a sector that has to-date operated with little supervision.

Wieland says the suggested changes could have enormous consequences for the German market as some HFT firms struggle to comply. But he stresses that market vulnerability to algorithmic errors, such as that experienced in the May 2010 Flash Crash, and more recently by Knight Capital, was exacerbated by its specific structure and not just how HFTs were regulated.

The reforms will require HFTs to apply for licensing as a financial services institution with Germany's financial regulator BaFin (Bundesanstalt für Finanzdienstleistungsaufsicht). Proposed amendments to the German Banking Act and the German Securities Trading Act will also enhance the powers of regulatory authorities, in particular requiring HFTs to submit their algorithmic trades and strategies to the regulator.

They avoid the more controversial elements discussed in connection with Mifid II, such as the requirement for a formal market-making obligation and a minimum lifetime for orders. Both would have the potential to destroy the business models of many HFT firms and would likely have severe consequences on liquidity and market efficiency at German trading venues, if implemented.

"The reforms force firms engaged in algorithmic trading to implement adequate and sophisticated risk management strategies and to ensure the trading system has sufficient capacity and effective safeguards to prevent erroneous orders," Wieland says.

"The newly-introduced maximum order-to-trade ratios, could have the effect that certain trading strategies cannot be deployed with the same leverage or volume as currently used, thereby curtailing the activities of HFT firms on German venues," he adds.

Wieland said enhanced focus on HFT firms' risk management process was undoubtedly a key tool to prevent a repeat of events like Knight Capital's trading loss. But he questioned the efficacy of Germany's approach.

"It is as much surprising as questionable that Germany has decided not to wait for a common European solution given that the Mifid II proposals are already at a fairly advanced stage," he says. "Instead, Germany appears to opt – not for the first time – for a national solution that precedes and pre-empts legislation at an EU level."

The compliance race

According to Wieland, the move had the potential to irritate the market.

The August 17 deadline for comments on the draft legislation left little room for the industry to influence matters further. "This will create substantial challenges for trading venues, investment firms and their trading and compliance departments," says Wieland.

Interested parties will need to swiftly analyse the impact of the draft legislation on their business models and practices and decide whether and how these can be brought into compliance with the new legislation.

Training of personnel, surveillance and strengthening a culture of compliance will also be crucial to avoid trading practices that may subsequently be perceived as market manipulation, and consequential investigations, fines and penalties.

What's more, if the new legislation is adopted as it stands, HFT firms that wish to continue trading in the country will need to quickly prepare for a licence in Germany or may have to cease trading at German trading venues. "It may be worth examining alternative solutions such as whether a Mifid licence could be obtained in another EU jurisdiction with a passport into Germany," he says.

"Some HFT firms could avoid the need for getting licensed in Germany and retreat from the German market," he says. "There is a risk that this could result in less liquidity on German trading platforms."

Institutions engaged in algorithmic trading should closely monitor the development and impact of the proposed legislation. "It may well be that the proposed legislation will be subject to changes and clarifications during the parliamentary procedure," he adds.

"Whatever the outcome, these suggested changes make clear that the time for HFT firms, be they US or European, to operate below the radar of regulators is over," he says.

Nonetheless, people should be aware that even with the best regulation and risk management systems in place, mistakes do happen. "The German approach may help regulators better supervise HFT firms, but that does not entirely rule out singular events like Knight Capital from happening again," he says.

Who is liable?

While regulators search for ways to prevent system errors, the more immediate question for the market is one of liability. Knight shareholders are expected to file class actions on the grounds the company was negligent in responding to the trading error and did not include important information on the functionality of its trading systems in disclosure statements. Claims against broker-dealers are typically heard on an individual basis before Financial Industry Regulatory Authority (Finra) arbitration panels – a process sometimes criticised for a lack of transparency and fairness.

"While there are many problems with arbitration, it has been getting somewhat better the last number of years," according to Jacob Zamansky, founding partner at Zamansky & Associates.

A Finra proposal to allow fully public arbitration panels, rather than a panel including one financial industry professional, was approved by the SEC last year – a move expected to result in fairer decisions by the three-person panels. This may be a moot point, though.

The biggest losers in Knight's debacle, aside from the company, were its shareholders. This is because trades in the six other stocks most affected were declared erroneous and undone following the trigger of single-stock circuit breakers. Broker-dealer shareholders are not prohibited from filing class-action law suits like broker-dealer customers are by their contracts.

"I think that is going to go to court," Zamansky says. "I don't think Knight Capital will be able to avoid a class action in a federal court as opposed to arbitration."

Ironically, Knight is pursuing remittance of some of its own damages alleged to have resulted from negligence at the hands of Nasdaq during the launch of Facebook's IPO in May. The firm reported losses between $30 and $35 million and has been evaluating all remedies available, according to a May 23 filing with the SEC.

Other losers in Facebook's IPO have already filed their claims against Nasdaq. Most notable of those is a collection of retail investors who have claimed losses resulting from a 30-minute delay in trading along with subsequent delays in order confirmations that caused investors to unknowingly submit repeat buy or sell orders. The class action, lead by Philip Goldberg, alleges negligence and was filed on June 12 in the US District Court for the Southern District of New York.

Suits against Nasdaq have brought to the forefront an immunity exchanges are afforded as self-regulatory organisations. Immunity will depend on whether Nasdaq's system errors arose as part of a government function.

The US Court of Appeals for the eleventh circuit has provided some guidance on where liability begins and ends for exchanges. In the 2006 case Weissman v National Association of Securities Dealers, the court affirmed that an exchange can be held liable when performing non-governmental functions, but recognised earlier case law regarding immunity in the performance of those functions. That case involved misleading Nasdaq advertisements.

Zamansky is one of the lawyers acting for the class of investors against Nasdaq. He says immunity does not apply in this case because Nasdaq was not acting as a regulator when its systems failed.

"They have an obligation to provide a system that works properly, and if they are on notice of a problem they have to act timely," Zamansky says. "When you are talking about markets that execute in a millisecond, two and half hours of a glitch is an eternity."

While it is unclear how the court will interpret Nasdaq's liability, Bodek thinks exchanges and HFT need to be more exposed to liability over the market-wide impact of bad algorithms.

Asia: ahead of the game?

As regulators in the US and Europe attempt to catch up with increasingly-sophisticated HFT and algorithmic trading strategies, Asian regulators are one step ahead. The trading strategy has received widespread media coverage across the continent, despite affecting Asian markets less than it does elsewhere.

Sources tell IFLR that Asian regulators are implementing frameworks for electronic trading before it becomes an issue. But counsel fear that increasingly prescriptive regulations could halt the natural evolution of market technology and create a substantial compliance burden.

The Securities and Exchanges Board of India (Sebi) released norms for electronic trading in March, while Hong Kong's Securities & Futures Commission (SFC) and Australia's Securities & Investments Commission (Asic) have recently released consultation papers detailing new regulatory frameworks for HFT and algorithmic trading.

Though global standards for HFT and algorithmic trading regulation are advised by organisations such as Iosco, each regulator has taken a different strategy, noting that their respective jurisdictions are in varying stages of these strategies' prevalence.

Sebi's trendsetting

Of these jurisdictions, India's exchanges are the smallest and least sophisticated. However, they are governed by the strictest rules. Though India's regulators have been criticised for the country's uncertain regulatory environment, counsel have quickly adapted to Sebi's norms on algorithmic trading and HFT since they were announced in March.

Because of the market's size, an errant algorithm on an Indian exchange is unlikely to destabilise markets globally. Instead, issues arise from the uneven environment created by HFT and algorithmic trading. Manisha Kumar, partner at J Sagar Associates, says small players face unfair competition from those able to afford the technology for these strategies.

Indeed, such a view prompted the Delhi High Court to issue Sebi, the Reserve Bank of India, and Indian stock exchanges with notices in August, calling for responses to a plea seeking to abolish algorithmic trading at stock exchanges as the facility was not available to small investors. The petition, by retail investor group Intermediaries and Investor Welfare Association, alleges algorithmic trading puts small investors at a disadvantage in comparison to big brokers and foreign institutional investors. The matter is fixed for hearing on January 29 2013.

Kumar believes that India needs to encourage retail investors. But she warns that this would be challenging if muscled institutional investors dominate.

An in-house counsel at an international bank agrees. He notes that Indian firms would be disadvantaged because international funds primarily deploy these strategies. "In a market like India, there is a tendency to say that foreign firms have usurped the market via technology," he says, adding that Sebi would not want to encounter that level of criticism.

The most contentious aspect of Sebi's norms is its individual algorithm review requirement. Sources discourage individual algorithm review because its high cost is disproportionate to its potential effects on the market. The counsel commented that the individual review especially is an irritant when upgrading technology because the algorithm requires adjustments and therefore Sebi approval.

But he also highlights that Sebi is partially liable for approved algorithms because of the review: "We don't have protection in any other market, in which responsibility for algorithms is on the participant. In India, algorithms go through exchange approval, so someone else has to look at it. In the worst-case scenario, we may not be held entirely responsible."

He predicts that regulators elsewhere would require individual algorithm review, which has already been proposed in Germany. As for the next iteration of Sebi's algorithmic trading norms,  Kumar said that they may direct stress testing of exchange and broker systems and encourage retail participation, which would go a long way in increasing investor confidence.

Hong Kong queries liability

In late July, the SFC released its first consultation paper on the regulatory framework for electronic trading on exchange-based platforms. This was eagerly anticipated, as rules for electronic trading in Hong Kong have not changed since 1999. Though the Hong Kong Exchange is the third largest in Asia, HFT and algorithmic trading strategies comprise a negligible portion of its orders because of a long-standing stamp duty on each trade.

Overall, the SFC's proposals are less stringent than Sebi's, and do not require individual algorithm review. However, while Indian counsel are satisfied with Sebi's risk allocation, Hong Kong counsel are concerned with who is ultimately liable for malfunctioning algorithms.

Sources disagree with the consultation paper's requirement that providers of automated systems be held ultimately responsible for orders' compliance to the regulatory requirements. A director of the Hong Kong Securities Association warns that this would especially affect smaller brokers, as many who provide electronic trading services rely on external system vendors.

Aside from relying on compliance confirmation from the system vendor, small brokers may lack the technical expertise to determine whether the electronic trading systems meet the requirement proposals in the consultation, and that compliance would be costly for those without in-house IT experts. A partner at an international law firm agrees, saying that he expects a significant increase in management time dedicated to governance and compliance.

Moreover, the partner notes that providers could also face concerns regarding direct market access (DMA) services. Under the proposal, providers are responsible for pre-trade risk controls as well as post-trade monitoring. He adds that while providers of electronic trading systems should have safeguards in place, they cannot control whether third parties choose to utilise their products in an abusive or manipulative manner.

Counsel note that Hong Kong faces increasing competition from other less restrictive Asian markets for liquidity. One partner says: "It would be unfortunate to put in place unnecessarily onerous compliance requirements around HFT relative to other forms of trading, so that institutions move this technology to other Asian markets, which could mean less liquidity in Hong Kong."

Asic's search for balance

An alternative to the Hong Kong Exchange is the Australian Stock Exchange (ASX), in which HFT and algorithmic trading play a prominent role. In August 2012, ASX CEO Elmer Funke Kupper estimated that HFT account for 15 to 25% of trades in the Australian market.

Like Hong Kong's SFC, Asic is looking to revise its market integrity rules and in conjunction its guidance on automated trading. However, it is important to note that Asic has completed several iterations of consultation papers on the subject, while the SFC has not yet received market input on its first consultation paper.

Asic released a consultation paper on August 13 that elaborates on several measures similar to those detailed in the SFC's consultation paper. These include an annual review of systems and having direct and immediate control of all trading messages, including pre-trade controls. Though the rules seem similar to those proposed in Hong Kong, Asic has had more time to fine-tune its proposals.

Damian Jeffree, director of policy at the Australian Financial Markets Association (Afma) said Asic was originally considering a quite prescriptive approach to HFT through the market integrity rules, but that would not have been the right approach. The aim is to have proper processes in place to ensure that systems are well-designed and well-implemented, he says.

Moreover, he cautions against targeting particular investment classes in respect to their access to and competence with technology. "Restricting technological innovation in the market carries its own risks," he says. "We expect Asic to be active in regulating HFT but this does not mean that undue constraints need to be placed on how business is done."

But the ASX may change the environment for HFT and algorithmic trading on its own. Kupper, its CEO, has stated that ASX has made submissions that suggest how to better protect Australia's markets, including charging HFT transaction fees similar to those in Hong Kong.

China’s new arrival

Algorithmic and HFT are increasingly popular strategies in China, but more prevalent in the commodities and futures markets. Local brokers are, however, becoming more interested in using HFT strategies on equity exchanges. Citic Securities, China's largest broker, purchased algorithm trading technology from Nasdaq-listed Progress Software in May, while broker Guangfa Securities agreed to use a platform to develop algorithms by Steambase, a US company.

While the Shanghai Stock Exchange says that it will implement new rules to manage HFT, including placing trading limits on firms with abnormal order volumes or frequent order cancellations, it has not yet specified a timeline.

Sources tell IFLR, however, that regulators are gathering information about HFT. In China less than 2.5% of trades are executed via automated trading systems. In New York, that figure is more than 70%. Kingsley Ong, partner at Eversheds Hong Kong, says that there is very limited systematic risk arising from automated trading in China for Chinese regulators to be concerned at this stage.

Ong expects the market for electronic trading to develop, though. "Development may be cautious and slow in China, but this does not digress from the growing trend and demand for embarrassing automated trading. As the market becomes more mature, the trading volumes will likely accelerate," Ong says. He predicts that brokers will develop more creative algorithms and innovative ideas for trading in China.

To date, the difficulty with HFT of equities has been that foreign investors likely to use automated systems can only trade A-shares on the equities exchanges via the Qualified Foreign Institutional Investor programme. Also, investors must use exchange traded fund conversions to get around the T+1 restriction in the A-share market. That restriction, however, does not apply to the futures and bond markets.

The Zhengzhou Commodity Exchange, Dalian Commodity Exchange and Shanghai Futures Exchange are among the most popular platforms for automated trading purveyors, and have recently taken measures to regulate HFT.

In November 2010, the Shanghai Futures Index implemented guidelines on monitoring abnormal transactions – a move presumed to have been sparked by irregularities caused by electronic trading. Its provisions require exchange to monitor trades and safeguard the exchange from errant trades, but state that members should stop their clients' abnormal transactions. A 2008 regulation had codified regulatory review of trading software. There are also exchange rules that limit the strategies utilised by algorithmic trading, such as allowing only 500 order cancellations per contract per day.

The rules have worked

Securities regulators should be wary of basing HFT reforms on this year's US trading glitches. Trading losses resulting from the Bats Crash, Nasdaq's untimely launch and subsequent delay of Facebook shares, and the Knight Capital debacle were almost exclusive to those companies and their shareholders. If interpreted in a sense of market-wide impact, the SEC's post-Flash Crash regulations did their job.

Single-stock circuit breakers stopped trading in the stocks most affected in each instance. This has the downside of making some trades erroneous and partly hurts market certainty, because traders might not know if their trades will be busted. But the alternative would have meant actual wild price swings.

"The Knight situation is in some ways a validation of market mechanics," Andresen says. "A private company made a series of mistakes which led them to do unprofitable trades in the market. Clients were not impacted, the exchanges were not impacted and the market data continued to flow smoothly."

The question that naturally follows revolves around an interpretation of best practices. High frequency traders tell IFLR they simply would have done things differently.

Regulators should not bow to public pressure, and not forget the enormous benefits of HFT. "It adds significant liquidity to the market," Wieland says

"A regulatory balance needs to be struck which stabilises the market by introducing effective risk management procedures, and emphasises the need for proper business continuity plans and algorithmic trading stress testing without killing the industry," he says.

"Regulators worldwide must work to establish a regulatory framework that doesn't put HFT firms out of business," Wieland adds. "That would diminish liquidity on the market and ultimately make markets operate less efficiently."