Electronification of trading

Speech by Edwin Schooling Latter, Head of Markets Policy, FCA, delivered at the FIX 2016 EMEA Trading Conference. This is the text of the speech as drafted, which may differ from the delivered version.

With thanks to the FIX trading community for their invitation to today’s event, I’d like to take the opportunity to make a few remarks on how the FCA views the increasing electronification of trading, its implications and consequences. 

Electronification has many advantages – which members of the FIX community will be very conscious of – facilitating trade capture and straight-through-processing, post-trade processes, risk management and trade reporting. At the same time in markets where electronic and automated trading are already the norm, such as cash equity and more liquid exchange-traded derivative markets, we have seen a continuing race to be the fastest, with firms competing to shave off milliseconds, then micro-seconds, and now even less from transaction speeds. The debate about the risks and benefits of such high frequency algorithmic trading continues. Fast, automated trading, including algorithmic and high-frequency trading, has also been becoming an important part of a wider range of financial markets – becoming more established in FX and, in some countries, also in fixed income markets such as sovereign bonds. This has raised various questions worth considering in greater depth. Is electronification, or the algorithmic and high-frequency trading that can follow it, or an associated change in the type and nature of firms acting as market makers a contributory factor in price volatility and some of the flash price spikes that have been observed? How could greater electronification be affecting market quality and resilience?

In markets where electronic and automated trading are already the norm, such as cash equity and more liquid exchange-traded derivative markets, we have seen a continuing race to be the fastest, with firms competing to shave off milliseconds, then micro-seconds, and now even less from transactions speeds

Price volatility can of course have many causes – changes in perceptions of economic prospects, changes in risk appetite, changes in supply of an asset made available by an issuer, or changes in demand for an asset based on borrower or investor needs – factors that will usually have no link whatsoever with whether the market microstructure is one that is voice or electronic, or does or doesn’t support high-frequency trading. And, of course, it is a desirable feature of effective markets that prices do change in response to changes in economic fundamentals, such as the likely future income from an asset, and changes in interest rate expectations. But that still leaves an important question as to whether, sometimes, the changes in market structure associated with the move to high-frequency electronic trading might themselves exacerbate price volatility. Large price spikes in extremely short timeframes – so-called ‘flash crashes’ – have been a striking phenomenon over recent years, and extended across a wide range of markets, from equities to FX, from derivatives to sovereign bonds.

The most prominent of these events – for example the May 2010 flash crash, the October 2014 'flash rally' in US sovereign bond prices, the dramatic spike just after the removal of the Swiss Franc-euro peg in January 2015, gyrations in equity prices in August last year, or the triggering of market-wide circuit breakers in Chinese equity markets in January this year, have not been UK focused. But they do present us with the question of whether similar events could happen here in the UK, and whether UK market structure, or micro-structure, is as robust as we can reasonably expect it to be.

Looking first at what we can learn from experiences in US markets, it is important to note the structural, regulatory and technological differences between UK and US markets. These may explain some of the different experiences in our respective markets over recent years. Take the equity and ETF market volatility experienced in US markets on 24 August 2015: NYSE’s Rule 48 allows market makers to suspend on volatile days their normal obligations to publish opening prices of stocks. This rule was triggered on 24 August. The reduced access to pricing information had consequences for other markets. For example, without available prices for many stocks, and the consequently reduced opportunities to manage risk and exploit arbitrage opportunities vis-à-vis positions in exchange traded funds, large discounts relative to net asset values emerged in the prices for the these ETFs. The UK does not have an equivalent to Rule 48. In the event of significant volatility pre-opening, the LSE’s opening auction period can be extended by a few minutes, but the indicative price is always disseminated and market participants know that they can expect the market to be automatically opened within less than 10 minutes of the usual time.

But there are also common features across US and UK markets, such as the prominent role of high frequency trading in our equity markets. The 2010 flash crash raised concerns about the vulnerability to price spikes of electronic trading infrastructures characterised by the prevalence of algorithmic and high frequency trading strategies. While electronic trading is, currently, much less established on UK sovereign bond markets than US equivalents, the 2014 flash rally in the price of US Treasuries raises further uncomfortable questions about whether changes in market structure are making key fixed income markets more susceptible to price spikes. Given that prices in these markets can have a major impact on the value of so many other assets, including on the valuation of collateral, and on marked-to-market balance sheet positions, the consequences of unstable prices in these markets cannot be lightly dismissed.

One argument is that in more traditional market making arrangements, with humans making judgments based on news, and with time to think, rather than computers interacting on the basis of pre-programmed instructions in response to other transactions more quickly than humans can think, we are less likely to see price swings that manifestly cannot be justified by any change in fundamental factors. It is also argued that HFT liquidity providers don’t have the deep balance sheets of more traditional market makers and are consequently perhaps less able and less willing to take positions that can help keep prices stable during temporary imbalances of buy and sell orders. But we should probably be wary of an illusion that, in the past, more traditional market makers were always willing to offer stable and continuous pricing. One possibility is that electronic markets simply give increased visibility to price volatility. In an old-style voice market, would the spike in prices have been manifest instead as the dealer not picking up the phone, or declining to offer a price but asking to call back in a few minutes? As CFTC Chairman Timothy Massad saliently pointed out in a recent speech: 'Contrary to what some have suggested, I suspect it was difficult for market makers in the pre-electronic era to routinely maintain tight and deep spreads during volatile conditions. They likely took long coffee breaks.'

Electronification and volatility – is there a link?

But before we get our own coffee break, I’d like to explore a bit further the possible links between faster trading and volatility.

This is an area where detailed analysis is difficult and time consuming.

Looking at London markets where high frequency trading is most established, it is not easy to identify with confidence a clear trend in intraday price volatility. For example, looking at LSE order book data, intraday volatility measured across 1-minute time intervals has largely remained stable over the past decade, with the exceptions of the 2008 and sovereign debt crises (calculating the intraday volatility as the standard deviation of 1-minute midpoint returns), the drivers of which were clearly not market micro-structure.

The most recent studies we have seen suggest that, in 'normal' market conditions, the presence of HFT players can be beneficial for market liquidity and reducing price volatility.

But that does not necessarily mean that the impact of high-frequency algorithmic trading on liquidity and volatility is equally benign during episodes of stress. There is the hypothesis that HFT firms, without the advantage of deep balance sheets and perhaps without clear market making obligations, are more likely and more able to reduce their provision of liquidity at the very point when it is especially important - in stressed markets. From being a substantial provider of trading volumes in ‘normal times’, the argument runs that these firms might temporarily exit financial markets during extreme episodes of volatility – reducing liquidity and increasing the likelihood of wide price moves.

And alongside the concern that price changes caused by external factors might sometimes be exacerbated by such features of an HFT market micro-structure, there is also the fear that 'rogue' algorithms themselves can trigger price volatility.

FCA market monitoring work… focusing on occasions where single security price spikes have been caused by the impact of unintended orders, indicates that these have usually been due to human error, rather than algorithms behaving in unexpected ways

Drawing reliable conclusions from analysis of the impact of HFT in times of market stress is perhaps even harder than assessing their role in normal times, in part because such episodes are infrequent. But the FCA continues to explore through the data available to us how market participants using HFT strategies traders behave during price-spike episodes.

Turning first to the question of whether it is algorithms that trigger the price dislocations, we have observed, through the confidential order book data available to us, examples of algorithms not functioning as intended. But FCA market monitoring work analysing experience over the past year and a half, for example, and focusing on occasions where single security price spikes have been caused by the impact of unintended orders, indicates that these have usually been due to human error, rather than algorithms behaving in unexpected ways. That’s consistent with findings from our supervisory work on pre-trade controls on automated trading, which indicate that the controls employed by proprietary trading firms using HFT are generally more comprehensive and sophisticated than those of other market participants.

The harder question is the extent to which the action and interaction of HFT algorithms has exacerbated or mitigated price volatility triggered by other events.

The Joint Report by the SEC and CFTC on the market events of 6 May 2010 noted that while selling pressure in the E-Mini S&P500 futures contacts was initially partly absorbed by high-frequency trading firms and other intermediaries, as prices continued to decline these firms then aggressively sold contracts to reduce their long positions and pulled back from providing limit order bids in any significant depth. Resting limit orders to buy during the depths of the flash crash summed to less than 1% of the depth of market bids observed at the beginning of the day. But as those numbers demonstrate, HFT firms clearly were not the only type of market participant temporarily to step back from trading or making markets, and arguably this could be considered rational behaviour given the circumstances.

A more optimistic theory is that HFT firms are actually less likely to withdraw from markets during volatile periods than their ‘slower’, more traditional counterparts. Volatility represents an opportunity to profit for nimble trading firms, and liquidity providers who can confidently and adequately manage the risk.

Recent academic research, notably studies using Nasdaq data, indicate that HFTs provide liquidity during extreme price movements and play an important role in absorbing imbalances created by non-high frequency traders.[1][2]

Indeed, preliminary findings published by US authorities on the ‘flash rally’ in US Treasuries seem to provide mixed evidence on the role HFT firms played in the event. HFT trading volumes were very high on the day but HFT firms continued to make markets throughout the period of heightened volatility – albeit in smaller size than normal – and this helped contribute to continuous trading (ie no ‘gaps’ in pricing).

This is an area where further careful study is warranted, and, notwithstanding that the biggest price-spike headlines have, to date at least, been generated in other jurisdictions, the FCA will continue to monitor carefully the markets where we have regulatory responsibility. We have observed what you might call 'mini-flash events' in individual stocks, or sometimes multiple securities, across a variety of UK markets in recent years. The fact that none has been as severe or headline-grabbing as the other examples I’ve mentioned does not mean we can afford to be complacent.

What role does market microstructure have in reducing price volatility?

That takes us to the important question of the role the design of market microstructure, and in particular the controls put in place by trading venues, can play in limiting vulnerability to price spikes.

One dimension of that discussion is when it is appropriate to have circuit breakers in place, and how these can best operate. The 2010 flash crash prompted US markets regulators to review whether better-calibrated circuit breakers could have helped mitigate that crash. On the other hand, their operation in Chinese markets earlier this year stimulated wide-ranging debate about their cons as well as their pros, including their impact on market confidence, and whether they can in fact accelerate price moves, for example as market participants seek to sell in advance of the circuit breakers being triggered.

In our view, well-designed circuit breakers do have an important role to play in supporting market resilience. The use of trading halt mechanisms to control price volatility is well established in the UK, both in equity and exchange-traded derivatives markets. You may be surprised to learn that on the LSE, for example, circuit breakers were triggered some 9,700 times in 2014. The majority of these cases can be easily linked to news that affected views on price, though there are also examples where moves related to orders inputted incorrectly or carelessly. One example is the price spike in HSBC plc’s share price which triggered LSE’s circuit breaker on 30 Jan 2014 – the details of which can be found in Market Watch 46. Nearly three-quarters of the halts related to more illiquid and volatile securities, such as small-cap shares on AIM or exchange-traded products, where a short trading halt may be considered unlikely to hit the headlines. But they were also triggered on 188 occasions in relation to FTSE 250 stocks, and 14 times in relation to FTSE 100.

Well-designed circuit breakers do have an important role to play in supporting market resilience

MiFID II is going to make it a requirement for all trading venues to be able to temporarily halt or constrain trading in times of extreme volatility, for example through a circuit breaker, as one part of a wider set of other controls to limit the risks from automated trading. MIFID II does not impose a single design for these circuit breakers, but ESMA will be working in the year ahead on guidelines for their calibration.

One interesting question is whether there is a case for more harmonisation of circuit breakers where an asset is traded on multiple venues, or even where related assets (for example the derivative and the underlying) are traded on different markets. We’ve seen from US experiences how quickly volatility in one asset class can spill over into other asset classes because of arbitrage, most notably between futures markets and markets for the underlying securities, but also how lack of available prices in one market can undermine price formation in others. It is questionable whether closing markets any more than is absolutely necessary is compatible with maximising their resilience, or compatible with maintaining confidence in their effectiveness. If market participants consider prices are temporarily dislocated from fundamentals in a way that is disadvantageous to them, for example in a flash price spike scenario, they will usually be able to hold back the trade until later, perhaps just a few minutes later. It may be better to preserve their ability to choose whether to trade, and to try and maintain opportunities to trade out of an unhedged risk position for those who want or need to, rather than to make trading impossible.

MiFID II will also bring a number of other market strengthening reforms designed to limit the risks that may accompany some of the benefits of electronification of markets, and greater use of algorithms.

For example:

  • investment firms will be required to test all trading algorithms prior to their deployment, and trading venues will in turn be required to confirm their members have done so
  • trading venues will be required to have systems that reject orders that exceed pre-determined volume and price thresholds or are clearly erroneous
  • to provide confidence that price dislocation will not be caused by trading venues’ systems being overwhelmed by computer-generated messaging traffic, trading venues will be required to ensure their systems have sufficient capacity to deal with peak order and message volumes, and will be required to have throttle limits (eg limits per member on the number of order sent per second)

Put together, the controls required by MiFID II seek firstly to reduce the likelihood of price dislocation being caused by weaknesses in market microstructure (algorithm testing, capacity controls, throttles), secondly to reduce its impact should it occur (circuit breakers), and thirdly to encourage liquidity back into the market (through requiring market maker agreements).

Conclusion

Given that the increasing electronification of markets will probably gradually continue, and is likely, at least in more liquid markets, to be followed by increasing algorithmic, and high-frequency trading, we need to continue to develop our understanding of how this affects market quality.

In some areas we have already decided on a regulatory response – and in the EU this is embodied in particular within the tighter rules that will come into force via MiFID II.

In other areas, continued analysis and further discussion is necessary on how best to design and calibrate in our market infrastructure the safeguards that protect against price volatility that does not relate to economic fundamentals. Industry involvement in that discussion is essential, and the FCA looks forward to dialogue with market participants as we begin to develop the ESMA guidelines over coming months. Thank you.

[1] High Frequency Trading and Extreme Price Movements (Jonathan Brogaard, Allen Carrion, Thibaut Moyaert, Ryan Riordan, Andriy Shkilko, Konstantin Sokolov, December 2015)
[2] Limit Order Placement by High-Frequency Traders (Subrahmanyam, Zheng, 2016)