Episode 16. Markets at the speed of light

How to Build a Stock Exchange

This episode explores the technological transformations that have led to markets at the speed of light: algorithmic traders and flash crashes. Yet for all the images of terrifying AI we discover that stock markets in the cloud are more rooted in material than ever before, pushing against the laws of physics in the pursuit of speed and profit. We see a culture war between hoodie and suit, techie and yuppie, but find – no surprise here – that whatever the uniform, the elites win out in the end.

Transcription

The Frankenstein story – the monster that bursts out of the laboratory and pursues its creator – is firmly embedded in our collective imagination. The novelist Robert Harris gives it a spin in the Fear Index, published in 2011. But the monster is not a thing of flesh and blood. It is an artificially intelligent trading algorithm launched by a Geneva-based hedge fund. It is fantastically, malevolently intelligent: able to penetrate secret files and to discover the worst imaginings of its creator, to conduct a reign of terror through purchase orders and sub-contracts. As its creator attempts to burn down the servers that house it, the algorithm uploads itself into the digital netherworld where it roams free, doing as its code instructs: feeding off fear for financial profit.

Harris has a keen ear for details in the news, and the financial cataclysm sparked off by this machine actually took place, just over ten years ago, in the afternoon of 6 May 2010. A wobble in the US markets, and then a spectacular collapse: the Dow Jones losing 998.5 points in 36 minutes, a trillion dollars of capital evaporating in five. Circuit-breakers – automatic cut outs designed to stop the market self-destructing – halted trading. When the market opened again, prices climbed quickly back to the morning’s levels. Although individual traders may have made or lost fortunes (we don’t know – and Harris deftly weaves fiction into the gap) very few ripples spread into the economy as a whole. This was the ‘Flash Crash’.

There may have been fear but there was no panic, no shrieking or shouting. The whole affair was conducted algorithmically, as high-speed trading machines did the electronic equivalent of yelling ‘sell, sell’, unloading stock to each other at ever-falling prices, and creating a self-fulfilling cyber-crash. Algorithms don’t panic, but they do form expectations, and they do so in thousandths of a second.

An initial investigation found that a large sell order had triggered the flash. There was a veiled reference to a problem with the timing of data feeds, a technical, structural problem. If you follow the news in the UK, though, you might have heard of the Hound of Hounslow, Navinder Singh Sarao, a solitary London trader with unusual personality traits who built an engine to ‘spoof’ the Chicago algorithms and made millions trading from his bedroom. American regulators became convinced that his activities had sparked off the crash, though this seems a lot less plausible than the fiction of malevolent artificial intelligence. Sarao may have made $70 million but most of his money seems to have ended up in the hands of fraudsters and questionable entrepreneurs. The only thing he purchased was a second-hand VW which he was too nervous to drive. He was extradited to the United States to face justice. The judge, expecting a criminal mastermind, saw instead a 41-year old man with autism who still lived with his parents and laid down a lenient sentence of a year of house arrest, even if Sarao had threatened to cut off the thumbs of a market administrator.

Hounslow, for those who don’t know London, is an unremarkable borough to the west of the city: suburbs, offices, few tourist attractions. Though the pun on Wolf of Wall Street may have been too tempting to avoid, it tells us something. In the place of the champagne and cocaine fuelled highlife of Jordan Belfort, we have a super-trader in an upstairs bedroom clad in hoodie and jeans, the global uniform of the techie. The Hound is just one manifestation of the culture war that has shaped financial markets over the last two decades: hoodie and baseball cap versus shirt and tie, techno wizard against Princeton-educated Master of the universe. That he was extradited to America and tried for market malfeasance shows, however, that market and state still walk hand in hand, whatever uniform the managers are wearing. That the only person of colour in this whole narrative so far is stood in a court of law says something else about financial markets, something that needs to be dealt with in a later episode.

Hello, and welcome to How to Build a Stock Exchange. My name is Philip Roscoe and I am a sociologist interested in the world of finance. I teach and research at the University of St Andrews in Scotland, and I want to build a stock exchange. Why? Because, when it comes to finance, what we have just isn’t good enough. It’s been a while since the last episode, my apologies, but there is some stuff going on. If you’ve been following this podcast, however, you’ll know that I’ve been talking about how financial markets really work, and how they became so important. I’ve been deconstructing markets: the wires, and screens, the buildings, the politics, the relationships, the historical entanglements that make them go, all in the hope of helping you understand how and why finance works as it does. As well as these, I’ve been looking at the stories we tell about the stock market. You might be surprised how much power stories have had on the shape and influence of financial markets, from Daniel Defoe to Ayn Rand. I’m trying to grasp the almost post-modern nature of finance, post-modern long before the term was invented, the fact that finance is, most of all, a story. Start-ups are stories, narratives of future possibility; shares and bonds are promises based on narratives of stability and growth. Even money is a story, circulating relations of trust written into banknotes, credit cards and accounts. Stories set the tone, make the rules, determine what counts and what does not. A good stock market needs a good story, so if we’re serious about rebuilding financial institutions then we need to take control of those stories.

Markets populated by algorithms scarcely understood by their creators raise all kinds of new and pressing problems. Fictional physicists living in Geneva and leveraging their experience of quantum mechanics into monstrous artificial intelligence; autistic coders living with their parents; transaction speeds that push up to the possibility of natural laws; there’s something different in contemporary finance…

This episode is all about the technological projects that transformed financial markets beyond recognition. I need to offer a caveat here. Ethnographies of high finance are not at all my domain, and I’ll be relying more than usual on the work of colleagues: Donald MacKenzie, Daniel Beunza, Juan-Pablo Pardo-Guerra, Christian Borch, Anna-Christina Lange, Marc Lenglet and others. You’ll find full references to my sources in the transcript on the podcast website.

We can think of changes that have swept through financial markets in two ways. First of all, they are a technological project driven by the endeavours of engineers. The result has been a wholesale transformation in the materiality of markets. To step into a trading pit now, and we can think of pit only in the most metaphorical sense, is to step into a warehouse of humming and chattering servers. Like the traders of old, they jostle for space around a central exchange, but space measured out in fibre-optic cable and milliseconds. We can also, though, think of these transformations in terms of a wholesale change in our understanding of how exchanges should work, as the metaphor that underpins them changes from one of the market as a fundamentally social entity to market as a computational device where efficiency becomes of paramount importance. The market ceases to be a concrete thing in a specific place and becomes a distributed network located nowhere, and everywhere: Wall Street, Chicago and Houndslow.[1] This change in our understanding of what the market actually is, what it is all about, reflects longer term moves in our understanding of the economy under neoliberalism. From von Mises and Hayek onwards we have grown accustomed to thinking of the economy – the market (in scare quotes) – as a vast dis-embedded computational device as opposed to a specific set of social and material situations.

Let’s start with a story of technological progress. We may recall from episode eight how automation had long been a dream of economists and policymakers who fastened on the possibilities for efficiency and surveillance that a mechanised market might offer: moving trades from the inaudible whispers of brokers to the easily supervised daylight of a centralised system. Pardo-Guerra’s study of the automation of the London stock exchange shows how the process began with the automation of tedious routine work of settlement and clearing, work previously conducted after hours in the rooms beneath the Exchange’s trading floor. Allowing the technologists in, even here, cracked open the closed world of the LSE. Treated at first like second-class citizens, the engineers built a series of systems that incrementally advanced the automation of trading until, on the day of Big Bang, 6 October 1986, the LSE opened in a fully electronic form. We saw in episode eight how this change took many by surprise, not least the LSE’s own management which had expected to operate a hybrid face-to-face and electronic trading system. But within days the trading floor was dead, and within months it had been closed. The engineers had built their own networks of power within the organisation and suddenly they were running the show. We saw how TOPIC, the LSE’s dealing screen, augmented by the FTSE 100 trigger page, created a completely new space for the market: a series of digital representations of trade accessible anywhere. It was still, however, a hybrid solution with dealers advertising prices that would be transacted by phone or voice, ‘folding’ existing practices into a new technological arrangement; the engineers’ institutional advancement did not really upset the money-making hierarchies of the LSE.

A different kind of challenge came from outside the LSE. By the mid-1990s, as Pardo Guerra shows, an industry had sprung up in the provision of computerised infrastructures which could be bought almost off-the-shelf by anyone with the desire to set up a new exchange. “Within this sprawling ecology,” he writes, “there was increasing recognition of the dominant design… electronic order books that allowed for the direct interaction of instructions from investors without the intervention of humans to coordinate transactions.”[2] Three engineers, named Peter Bennett, Michael Waller-Bridge and Stephen Wilson, had spent years at the LSE trying to set up a pan-European order book system. Blocked in this endeavour they set out on their own. They called their start-up system Tradepoint, and parked it symbolically out of the City, in the architect Lord (then Richard) Rogers’ building in Thames Wharf, also home to the renowned River Café, the first of London’s great stripped-down continental-fare gastro hubs. All of this was a performance, even if the restaurant did help bring visitors to the office and allow them to make their case on home territory. What was it a performance of? Of difference, of outside status, of the power of technology to break up cliques and upset apple carts. Another performance took the form of a ‘computer room’, an ordinary room equipped with a huge ventilation duct and mains cable, out of bounds apart from the sign on the door, that helped to convince visitors that the market was backed by sufficiently weighty technology. In reality, the computer system was quite moderate, enhanced by the programming skills of a colleague Ian McLelland, who customised a software package bought off the shelf from the Vancouver Stock Exchange. As for the upset apple carts, that was a performance as well: Tradepoint brought to bear an impressively deep network of social relationships with existing players, including making an agreement with the London clearing house and inviting its boss, Sir Michael Jenkins, onto the Tradepoint board.

There was, as Pardo Guerra points out, a moral imperative to the Tradepoint offering: ‘by allowing competition beyond the control of the LSE’s market-makers, their electronic order book would narrow spreads, driving down costs for end investors’. The order books, and the practices that came to be associated with them, notably anonymity, were attractive to overseas investors, derivatives trades and hedge funds. It was a venue for early robot traders, market participants ‘represented by installed boxes literally sporting flashing lights’. Tradepoint never amassed the volume of orders necessary to be a commercial success, but it did, in Pardo-Guerra’s words, change ‘the language of what was possible and permissible’[3]. Although an attempt in 1995 to forcibly introduce an order-driven system led to a members’ rebellion and the sacking of chief executive Michael Lawrence, order-driven trading was now inevitable and in October 1997 the LSE introduced its new system, SETS. Order books began to diffuse through the institution from the most senior markets downwards.[4]

Pardo-Guerra’s observation that Tradepoint changed the language of the possible is crucial here. It takes us back to our second causal factor, the evolving understanding of the purpose of an exchange. Moving away from thinking of a stock exchange as an institution rooted in geographic and social place to a distributed network of information processing shifts what we value. Speed, efficiency and structural elegance are the things that matter. This is the engineer’s aesthetic rather than the financier’s and it flows from a wellspring of technological expertise. But you will remember also our account of markets as comprising organisational fields, a social theory that sheds light on the evolution of institutions as high status actors seeking to consolidate their advantages at the expense of the less powerful. As the Tradepoint episode shows, these new technologies and conceptions of market organisation become the next battleground in struggles for institutional dominance. You might recall how, when the LSE designed its junior market AIM, a group of influential market-makers managed to hold off electronic order books and preserve their profitable positions. But order books remained a contentious issue and by the early noughties, with AIM internationalised and home to stocks larger than British SMEs, the LSE began to employ them in its junior market. What could the market makers do? External competition seemed to be the only way for the market-makers to resist the power of the LSE but there was no competitor ready to hand. Or was there?

————–

We left OFEX in dire straits, with a failed fundraising, and the Jenkins family evicted from the firm. Into this void of leadership stepped Simon Brickles, the barrister who had been instrumental in setting up the constitution of AIM and had later become head of the market. He had left the LSE in 2003, frustrated by an increasing emphasis on order books and its move away from his vision of a market with light-touch regulation, a high temple of capitalism. Brickles sensed that the way out of OFEX’s problem was a headlong charge – not away from the LSE but towards it.[5]

His shareholders agreed. The market-makers who had supported the rescue fundraising to become major shareholders in OFEX were chafing at the high fees imposed by the London Stock Exchange – now a demutualised and revenue-focused global corporation – for settlement and transaction. The European MiFID regulations, expected in 2007, sought to open up competition between markets, but there was no possibility of competition unless a vehicle to challenge the LSE could be found. Brickles therefore began to expand the market’s offering. The company announced a £2.5 million fundraising to pay for an expansion in the number of securities traded, stating ‘the company intends to markedly broaden its existing trading services to encompass an extended range of securities. The enlarged trading service will allow brokers and investors flexibility in selecting their execution venue’. In other words, the junior market was to be positioned as a direct competitor to LSE’s smaller company markets and AIM. On November 10, 2005, the Times reported a private meeting at the offices of mid-tier broker Charles Stanley: ‘Present at the meeting were representatives from Stanley and dealers such as Seymour Pierce, Peel Hunt and Winterflood Securities, which has led the opposition to the LSE. Some brokers are upset at the extension of the LSE’s SETS part-electronic trading platform to various small-cap and AIM stocks, for which they claim it is unsuitable.’ And there you have it, an outbreak of strife over the rights and privileges to make money in the markets.

On 30 November 2005, after a period of intensive work, the PLUS service (as it was now called) was launched. It enabled brokers to trade any stock on the Official List, ‘everything from Vodafone, down to the smallest FTSE All-Share.’ But it was not yet a fully-fledged stock exchange and another funding followed, pegged to the ambition of achieving a licence as a Recognized Investment Exchange. According to the offer document, the firm, currently focused ‘on providing cost-effective quote and trading services dovetailed to the needs of small and mid-cap companies… is seeking to expand into offering services to meet the quotation and trading needs of larger companies and the UK institutional community.’ In February 2007 the offer, heavily oversubscribed, valued the company at £43m.

Central to the whole endeavour was PLUS’ trading system. It had to be fast. Tradelect, the LSE’s new £40m system, went live on 18 June 2007, cutting order processing time to 10 milliseconds and greatly reducing trading costs. PLUS’ efforts show that the process of setting up a new stock exchange had evolved from a primarily social to a material and technological project. It ordered a platform from the Scandinavian firm OMX, but that was just the start: it needed to connect to market-makers, brokers, data vendors and the internal surveillance system. It had to be robust. It was, as Brickles said, ‘a huge spider’s web, and if any one of those bits of the spider’s web doesn’t connect you cannot launch the market.’ July 2007 saw the granting of the RIE license, and the OMX X-Stream platform launched in November, just as MiFID came into force. Both took up quantities of management time and were finished in time for the November deadline: ‘No mean feat. We were running pretty hard’, said one of the executives.

But, as Tradepoint’s founders had clearly understood, starting a market isn’t just a technological project. PLUS’ concentration on the material infrastructure perhaps overwhelmed the social and discursive labour involved in setting up a new exchange. Despite a shared management expertise, PLUS failed to engage in the processes that had made the AIM launch a success: prolonged, interactive consultation with the investee community. Indeed, many in the smaller company community felt that PLUS was no longer seriously committed to its original constituency. They levelled the same critique that PLUS had been making against the LSE: a steady drift upstream towards bigger companies and more lucrative business. John French, the businessman who chaired the advisory panel, described the task of maintaining a focused market for smaller company shares as being like ‘pushing water uphill’ in the face of scant interest from institutional investors and the market’s own management.

Any doubts over the market’s direction of travel – from smaller company nursery to discount trading and trade reporting venue – would have been settled by the Turquoise affair, a significant and ‘traumatic’ distraction for management in the autumn of 2007. Turquoise was a dark pool, a lightly regulated trading venue, that would offer anonymity and low fees. Like PLUS’ move to compete with the LSE’s small-cap markets, Turquoise sprung from the fact that in the mid-2000s ‘people hated the LSE,’ then run by Clara Furse, ‘it was…vicious.’ It had formidable backers, a number of senior executives of global investment banks , ‘big swinging dicks,’ according to one interviewee, ‘…big players, nothing to do with small company investing but big players…[who] got it into their heads, probably rightly, that the LSE was taking too much of the pot in trading terms…’ Although it had first been mentioned in the press in April 2007 it had not made much progress, earning itself the sobriquet ‘Project Tortoise’. These executives needed infrastructure and expertise in market operation, and on 6 October 2007 the Daily Telegraph ‘revealed’ that PLUS was negotiating the terms of a ‘takeover’ with Turquoise, while the Independent announced a ‘merger’. PLUS shares were suspended at 28p following the announcement of a ‘non-binding heads-of-terms agreement with a third party’. But nothing happened. By 19 October talks were over, and Turquoise was reported as looking for a deal with Cinnober, a Swedish technology firm. Still no progress was made and eventually the whole thing was quietly absorbed by the London Stock Exchange, now run by the shrewd and politically aware Xavier Rolet.

By the early noughties, then, we have reached a situation where the scuffles between markets – battles for domination and profit among rival market participants – are played out through technological systems. The ‘market in markets’ longed for by regulators materializes quite literally in the wires of market systems and the code that flows through them. US markets followed the same trajectory. Throughout the 1970s and 80s ongoing institutional bricolage had led to the electronic NASDAQ system, where brokers displayed prices and dealt with each other by phone. Although the network spanned America, it encoded existing patterns of dominance and buttressed the power of the New York Stock Exchange and NASDAQ’s broker-dealers. These latter colluded, at least by habit and practice, to offer prices in even-eights only, keeping the commission to a quarter of a dollar per trade.[6] At around the same time, the New York Stock Exchange was mired in its own scandals, including the payment of $139m to CEO Richard Grasso as a ‘compensation package’ – the number so big it certainly warrants a euphemism.

The scene is right for a coup, or at the very least a culture war. Just as Tradepoint had set itself up as a self-consciously outside challenger to the LSE, all River Café and Thames Wharf, so in the US a new generation of code-writing techno-libertarians started to play in the markets. Their innovations cracked open the long established monopolies of NASDAQ and the New York Stock Exchange.

A subset of the NASDAQ automated system was the small order execution system, or SOES. After the crash of 1987 when market-makers just stopped processing orders, regulators made it compulsory for brokers to publish quotations and honour them. The unexpected consequence of such a move was that it provided a facility for outsiders to day trade smaller sums in the NASDAQ markets: youngsters in T-shirts and jeans and baseball caps staring at screens and hoping to catch out the brokers with a speedy click here or there. Traders congregated in the offices of firms like Datek with its headquarters in Broad Street, just round the corner from Wall Street. These youngsters became known as SOES bandits, revelling in their outsider status as they needled the established players in ways that transgressed the established etiquette of trading. Tensions often flared. MacKenzie and Pardo-Guerra quote an episode where a member of staff of a NASDAQ broker-dealer located at 43 Broad St, infuriated at being ‘SOES-ed’ by Datek’s traders, crossed to no 50, and barged into Datek’s trading room, screaming ‘You did it again, I’ll fucking kill you!’ He leapt at one of the Datek traders, so a more senior trader picked up a letter opener and stabbed him forcibly, fortunately only in the shoulder. This trading was edgy, all-in, as another colourful detail shows – ‘No one blinked when a chalk-faced guy doubled over a garbage pail and puked violently, never leaving his seat and trading right through the puke’.

Josh Levine was an engineer who tumbled into this world. At Datek, Levine began to build hard and software hacks that avoided more longhand operations, for example allowing quick keystrokes, or hijacking the printer feed from a NASDAQ terminal into a computer system. These eventually became a slick trading system in their own right, helping to crack open the closed shop of the NASDAQ broker dealers: faster, sharper, leaner than anything NASDAQ could provide. A crucial step forward came when Levine realised that he could cut out the expensive NASDAQ dealers altogether by allowing Datek traders to exchange stock between themselves. This required a matching engine, and Levine built one. He called it Island. It had low fees and even offered rebates to those posting sell orders. Levine built systems that worked elegantly from an engineering viewpoint, completely rethinking the organisation of algorithm and exchange, a programmers aesthetic that valued speed and efficiency above all else. Trade time dropped from two seconds to two milliseconds; Island’s engine was so quick that users realised the distance between their own server and the central machine mattered, and the practice of ‘co-locating’ servers in the exchange building – for a fee – appeared.

The offices in Broad Street, Manhattan, maintained the flavour of the dot-com start-up: T-shirts, hoodies, junk food and eccentricities, but soon enough, as MacKenzie and Pardo-Guerra put it, Island became a continent. By 2005, through a series of acquisitions, it had become part of NASDAQ and transformed the giant exchange from the inside out, rebuilding NASDAQ’s technological infrastructure along the Island model. Other programmers moved from Island to exchanges elsewhere and spread the technology as they went. Traces of Levine’s code still flow in the veins of NASDAQ, and his vision of how the engine of the market might work has been enacted worldwide. Order books as we know them today began life on a screen surrounded by junk food wrappers, in the office of a day-trading outfit in a Manhattan backstreet. The hackers won, intellectually at least.

Technological upheaval transformed not just the exchanges, but also their customers. It wasn’t long before the robots arrived, the real world equivalents of Harris’ fearsome algorithm. Program trading, where algorithms made suggestions to brokers, had been around since the mid-eighties. Indeed they had taken some of the blame for Black Monday in 1987, but they still depended on humans to get the orders transacted. Levine’s Island was perfectly suited for entirely automated trading, even down to the hacker-libertarian politics. In another study MacKenzie tells the story of one such firm, based in Charleston, Carolina, set up by academic statisticians who had previously built a model to predict the outcomes of horse races and figured the methodology would transfer to the stock market. In good times the firm came to be one of the leading tech firms in the county, though these good times came and went.

What MacKenzie shows, however, is that for all the barefoot, T-shirt, take on the world hacker aesthetic, the firm only really flourished when it discovered pockets of systematic advantage that were already being exploited by human actors. So, for example, the programmers learnt about the SOES bandits and built an algorithm that mimicked what these humans were doing, looking out for tell-tale signs of big movements in the markets. Then it was a question of machine competing against human, a simple race where the ones outcompeted weren’t the incumbent NASDAQ brokers but the human bandits in Broad Street. Trading at that speed needed a matching engine capable of managing the order flow and the algorithm plugged straight into Island’s, sometimes breaching the order limit of a million trades per day. It was trading figures like these that forced NASDAQ to buy Island, inviting the algorithms into the mainstream. And, of course, once trading becomes a race then only speed matters and everyone has to run; some 90% of global stock trade is now conducted algorithmically.[7]

One of the ironies of high-speed trading is that, just as the market has slipped into the cloud, so designers have had to pay attention to the place where trading actually happens. HFT has foregrounded the brute material from which markets are made, and this material is political. Automated markets are housed in heavily guarded warehouses outside major cities, New Jersey in the US or Slough the United Kingdom. As the market is literally and actually made in these places, the speed with which prices travel back to the trading algorithms is crucial. Co-location has become a sine qua non of high-frequency trading, with firms paying to locate their boxes as close to the exchange’s engine as possible. Links between exchanges come to matter. Michael Lewis’s book Flash boys is held together by the story of an extraordinary construction project, the building in secret of a fibre-optic link between New York and Chicago, drilling through the Appalachian mountains. Fibre-optic cables had already been laid along the railway track but that bends and twists through the mountains. The few milliseconds that could be saved by travelling in a straight line made the difference between being able to make a profit trading in the markets and never being able to do so. The investors who funded the line could hold traders to ransom. But the speed of light through glass is only two thirds of the speed of light through the air, so rivals have installed chains of microwave dishes between the cities, and finally a major project has built a line as close to the geodesic as possible. It’s faster on a clear day, but slower in the rain, and at certain phases of the moon the line is blocked by the tidal pull on Lake Michigan. We are literally at the limits of physics and yet, as MacKenzie points out, this is an economic arms race of the classic kind: enormously wasteful with huge rents being paid just so players can stay in the game. Even the players can see this: in the middle of describing how engineers have worked day and night to shave five to 10 nanoseconds from the processing time of specialised chips one of MacKenzie’s interviewees pauses to reflect that all that training, all that expertise could have done something else… something different.[8]

Though we might like to think of algorithmic trading as possessing the diabolic intelligence conjured up by Harris, it is much more a case of early bird catching the worm, where early bird is measured in power consumption, heat dissipation, and metres of fibre-optic cable. This in turn has thrown up serious questions about the fairness of high-frequency trading. Michael Lewis argued that we – pension holding, long-term investing citizens – are being scalped by these traders. Part of the difficulty is that algorithms are programmed to spot predictable trades and large buy and sell orders are by their nature predictable, despite the best efforts of brokers to hide them through their own high-speed slicing and dicing. Meanwhile machine learning and huge datasets have started to undo the formal anonymity of electronic exchanges as the most predatory algorithms learn to recognise and outmanoeuvre their more docile cousins.

Even if we do accept the necessity of high-frequency trading there are questions about how much the interaction order that we take for granted in everyday life – queueing, or telling the truth, for example – should transfer into the world of algorithms. In a recent blog, the sociologist Christian Borch has argued that culture is needed to prevent further flash crashes – there have already been several more. He writes about a group of firms working to introduce a better moral culture in algorithms; ‘they strive to eliminate any negative effects their algorithms may have on markets, and they have developed an ethos built on ensuring market integrity in every respect… these firms expend massive, ongoing efforts to comprehend how and why their algorithms behave the way they do, alone and together with other algorithms.’ Makers of algorithms must expend massive efforts to understand how they behave precisely because learning algorithms have a degree of autonomy. Indeed, writes the sociologist Kristian Bondo Hansen, algorithms have a tendency to over learn, making causal associations where there are plainly none and have to be taught to be good scientists, employing Occam’s razor and the principle of parsimonious explanation. AI turns out not to be so I after all. Hansen prefers to explain machine learning algorithms as a means of making sense of the swathes of noisy data that make up contemporary markets, distributed cognitive systems organised and curated by their programmers. But this is a circular defence; as so much of global equities trade is algorithmic, those same algorithms must be the source of that noise and HFT looks like the solution to a problem that it has itself created.[9]

All of which goes to remind us, once again, that stock exchanges have histories and organisational path dependencies that do much to shape their present form. We see in the development of cyber markets the outcome of a series of struggles between established players and new ones. Techno-libertarianism turns out to be just another elite discourse, just as gendered and riddled with privilege as the stock market monopolies it set out to crack open. Suggesting that culture can somehow be imposed upon high-frequency trading from the outside ignores the fact that it is there already: the engineer’s aesthetic, the junk food wrappers and Star Trek posters. And sometimes the establishment wins anyway. The story of PLUS tails off in 2009, with a pyrrhic victory on the courtroom steps after the LSE blocked PLUS from trading AIM stocks; the legal action had exhausted the smaller firm and when the LSE cut its fees its customers drifted back once more. The credit crisis did the rest.

Crisis seems an appropriate place to finish. For all the talk of culture and supervision and care for creation of algorithmic systems, contemporary cyber markets are fragile things. They can move so quickly as to out run even the exchange’s failsafe mechanisms. Hostile trading conditions created by predatory algorithms make it increasingly likely that institutional investors – the eventual users of equity markets – will attempt to trade over-the-counter in a situation that ironically parallels the organisation of AIM. Cyber markets are crisis markets, the material enactment of a narrative the market as a dis-embedded information processor, free from space and time. You can trace this story downwards, from the big ideas of liberal, then neoliberal, economists to the regulation and organization of markets. Or the other way, from the bottom up, through the technological projects of engineers and the mundane wires and circuits of finance through to a conception of markets as giant computers. We should allow both. No idea was born outside of the material world, just as every engineer who thinks markets might be better built has recourse to some imaginings of how things should be organized. Even if they are just ‘one day all of this will be mine.’

I’m Philip Roscoe, and you’ve been listening to How to Build a Stock Exchange. If you’ve enjoyed this episode, please share it. If you’d like to get in touch and join the conversation, you can find me on Twitter @philip_roscoe. Thank you for listening. Join me next time – for the penultimate episode, when we’ll be talking about crisis and exploitation.

—

Sounds under creative commons license from freesound.org

Server farm: cinemafia https://freesound.org/people/cinemafia/sounds/24080/

Computer chatter and war machine: ProjectsU012; https://freesound.org/people/ProjectsU012/sounds/361018; https://freesound.org/people/ProjectsU012/sounds/337249/

—

[1] This observation is drawn from Daniel Beunza et al., “Impersonal Efficiency and the Dangers of a Fully Automated Securities Exchange,” in Foresight Driver Review, DR11 (London: Foresight, 2012).

[2] Juan Pablo Pardo-Guerra, Automating Finance: Infrastructures, Engineers, and the Making of Electronic Markets (Oxfoird: Oxford University Press, 2019), 189.

[3] Ibid., 201.

[4] Michie, The London Stock Exchange: A History, 616.

[5] This next section is taken from Philip Roscoe, The Rise and Fall of the Penny-Share Offer: A Historical Sociology of London’s Smaller Company Markets (University of St Andrews, 2017), Other report.

[6] The SEC eventually launched a huge antitrust action against the broker dealers, with damages reported to be $910m in total. see Donald MacKenzie and Juan Pablo Pardo-Guerra, “Insurgent Capitalism: Island, Bricolage and the Re-Making of Finance,” Economy and Society 43, no. 2 (2014).

[7] Adam Hayes, “The Active Construction of Passive Investors: Roboadvisors and Algorithmic ‘Low-Finance’,” Socio-Economic Review (2019).

[8] Donald MacKenzie, “‘Making’, ‘Taking’ and the Material Political Economy of Algorithmic Trading,” Economy and Society 47, no. 4 (2018): 518.

[9] Kristian Bondo Hansen, “The Virtue of Simplicity: On Machine Learning Models in Algorithmic Trading,” Big Data & Society 7, no. 1 (2020).