Felix Salmon

Michael Lewis’s high-speed journalism

Felix Salmon
Apr 7, 2014 15:17 UTC

My full review of Flash Boys is now up at Slate. Tl;dr: he’s right for the wrong reasons. HFT is a bad thing, but not because it rips off small investors.

There’s a separate question worth asking, though: why is this book weaker than Lewis’s other books?

Partly, it’s because Lewis took a bet on the unknown. Lewis tells stories by focusing on individuals, and he clearly felt that he hit the jackpot when he found Brad Katsuyama, the founder of IEX. Katsuyama would in any case have been a compelling choice as the person through whom to explain HFT. But in this case Lewis managed to go one better: he caught Katsuyama at a very auspicious time, which meant that he could actually follow him, in person, through the launch of his new company. As a result, Lewis found himself unable to control the arc of his story: from the point of view of the narrative, Flash Boys was going to go wherever IEX went, even if IEX’s future was very unclear at deadline.

And while first-person access should in principle make the book better, because Lewis can add the kind of details which he can never find by talking to participants ex post, in practice, it often doesn’t. Rather, it means that Lewis seemingly felt compelled to, well, add the kind of details which he could never find by talking to participants ex post:

Don leaned with his back against the window, along with Ronan, Schwall, and Rob Park, while Brad stood in front of the whiteboard and took a whiteboard marker out of a bin…

Schwall looked over the desks and shouted, “Whose phone is that?”

“Sorry,” someone said, and the ringing stopped.

This isn’t novelistic color, it’s more akin to the famous drunk looking for his keys under the lamppost. When Michael Lewis knows exactly what story he wants to tell, he can talk to people and piece it together like no one else in the business. But in this case, Lewis chose the story before he knew how it was going to end, and so he ends up writing what he saw. Which is sometimes important, and sometimes isn’t. It’s a common problem when a journalist gets exclusive access to something: just because you’re the only person to witness something, doesn’t mean it’s particularly worth witnessing.

To make matters worse, Lewis felt the need to bulk up the book by dropping in, more or less verbatim, his entire Vanity Fair article on Sergei Aleynikov. That story was excellent: one of the best things that Lewis has written, which is a very high bar. But its narrative doesn’t fit with that of Brad Katsuyama; in some ways, the two are diametrically opposed. Aleynikov should probably have appeared in the book somehow, as an example of the way in which the big banks were thrown in panic by the rise of HFT. But that would have required Lewis writing the Aleynikov story all over again, a second time around — when the first time was already such a success. So he simply did a copy-and-paste job, which is not what his bigger story really required.

Overall, a lot of the weaknesses with this book are ironically the same as the weaknesses with the stock market: it’s just too fast. With a bit more time and care, Lewis could have broadened his story a bit, put Aleynikov into better context, and explained the real dangers of HFT rather than just the “you’re being ripped off” hyperbole. He could also have avoided some silly mistakes: Secaucus is west of Weekhawken, for instance, not east.

Then again, maybe Flash Boys is a sign of where book publishing is going. It will probably be read more on electronic devices than in print; it will probably be read mostly in the next few weeks, and become dated very quickly. It’s an event; it’s highly salient right now, but it doesn’t have the legs that, say, Liar’s Poker does. Books used to be objects with permanence; as they become increasingly electronic, they can start moving towards the more disposable model of, say, Vanity Fair style magazine journalism.

Which makes possible what you might call the Reputation Arbitrage. The Newsweek cover story on Satoshi Nakamoto got enormous amounts of attention just because it was a Newsweek cover story, appearing, in print, on the front page of a physical magazine with a storied brand. If exactly the same story had appeared on a lesser-known website, it would have caused much less of a fuss. Similarly, Flash Boys is getting enormous amounts of attention just because it is a book by Michael Lewis. If he had simply written the NYT Magazine story, without a book behind it, the article would still have been shared a lot, but I don’t think we would have seen the same response from, say, US law enforcement.

In the digital age, media are converging faster than people think they are. Certain formats — the magazine cover, the hardback book — retain a certain amount of vestigial reputational capital, which can cause people to write about them more than maybe they should. If only the same amount of attention had been paid to, say, Matt Taibbi’s scoop about SEC document-shredding. That is something to really get angry about.


Michael Lewis… Come on… You have been a great investigative journalist over the last 25 years… But the publication of Flash Boys is no more than a “Mad Man” advertising publication for the IEX exchange which is currently on a road show with Brad and Ronan.

The book is uninteresting, uninformative, and very biased,,, something that I thought you would never be.

IEX Group, the six month-old US stock exchange led by the heroes of Michael Lewis’s controversial new book Flash Boys, is to visit Europe for the first time in June as it seeks to build relationships with trading firms in the region.

Ronan Ryan, one of the heroes of Michael Lewis’ IEX’s chief strategy officer and head of business development, will start a 10-day European roadshow from June 18, and will visit the UK, France, Holland, Germany, Norway and Sweden.

Morgan Stanley, Goldman Sachs and JP Morgan plan to help introduce the exchange to European institutions during the visit, one person said. The banks declined to comment.

Come on Michael…. Whose side of the fence are you really on? How many shares of IEX do you own? You should be as transparent as your journalism has always tried to be.

Good luck…

Posted by competitive-mkt | Report as abusive

Don’t cry for “the little guy on Wall Street”

Felix Salmon
Sep 2, 2013 23:08 UTC

This happens every time something goes wrong on the stock market — every time there’s a flash crash, or a high-frequency trading firm blows up, or the Nasdaq is forced to go dark for three hours. A bunch of editors who don’t really know anything about HFT ask for stories about it, and they all want the same thing: a tale of how a small group of high-speed trading shops, armed with state-of-the-art computers, are using their artificial information advantage, and their lightning-fast speed, to extract enormous rents from the little guy.

The result is a spate of stories like Rob Curran’s latest piece for Fortune, which appears under the headline “Make $377,000 trading Apple in one day”. Of course, there are lots of ways to do that: one way would be to buy about 77,000 shares of Apple, for $37.7 million, and then watch them rise by 1%. But Curran reckons he’s found a better way — indeed, an easy profit which involves no risk at all. What’s more, this method is particularly evil, since apparently all of the profits that it generates are coming straight out of your pocket.

Curran’s story is based in large part on a “study” by Berkeley professor Terrence Hendershott. This study is never named, or quoted, or linked to, and I can’t find it on Hendershott’s web page, so I’m not going to blame Hendershott for any of the content of Curran’s article. Specifically, for instance, Curran’s sub-hed says that “A Berkeley professor finds out just how much a certain type of high frequency trading costs the average investor”. I suspect that Hendershott’s study actually purports to do no such thing*, and that “average investors” aren’t even mentioned in it. I say this because Hendershott is a smart guy, and I can’t believe that this kind of thing fairly summarizes any of his work:

It’s well known that some high-frequency computer geeks at firms like Getco LLC take advantage of latency, just as it’s well known that some Blackjack-playing computer geeks count cards in Las Vegas casinos. But it’s never been clear how much this type of trading costs the little guy on Wall Street.

Terrence Hendershott, a professor at the Haas business school at the University of California at Berkeley, wanted to find out. He was recently given access to high-speed trading technology by tech firm Redline Trading Solutions. His test exposes the power of latency arbitrage the way Ben Mezrich’s Bringing Down the House exposed the power of card counting.

According to his study, in one day (May 9), playing one stock (Apple), Hendershott walked away with almost $377,000 in theoretical profits by picking off quotes on various exchanges that were fractions of a second out of date. Extrapolate that number to reflect the thousands of stocks trading electronically in the U.S., and it’s clear that high-frequency traders are making billions of dollars a year on a simple quirk in the electronic stock market.

One way or another, that money is coming out of your retirement account. Think of it like the old movie The Sting. High-speed traders already know who has won the horse race when your mutual fund manager lays his bet. You’re guaranteed to come out a loser. You’re losing in small increments, but every mickle makes a muckle — especially in a tough market.

This is deeply confused. For one thing, there’s much more to HFT than simple latency arbitrage of the kind that Curran is describing here. And in any case, this kind of strategy just doesn’t work. To see why, just do the extrapolation Curran’s asking you to do. If high-frequency traders are making $377,000 per stock per day, then that would add up — multiply by 5,000 stocks, and by 250 days per year — to total profits of almost $500 billion per year, or about 3% of America’s GDP. And that doesn’t even include the extra profits made by high-frequency trading in other asset classes, like foreign stocks, or currencies, or interest rates.

Curran’s number, in other words, doesn’t pass the smell test.

Note that Hendershott’s one-day profit was “theoretical” — Curran never asks the question of whether Redline in practice makes anything like that of money, or if they don’t, why they don’t.

In the real world, it should probably go without saying, hundreds of billions of dollars in annual risk-free profits aren’t just sitting on trees, waiting to be plucked. The idea behind latency arbitrage is simple: you’re essentially trying to buy or sell at yesterday’s prices, in the knowledge of where the price is today. (Except, we’re talking about a time lag measured in milliseconds, rather than days.) If you were to actually enter the market with a simple latency-arbitrage algorithm like this one, however, you would almost certainly lose your shirt in no time: a thousand other algobots would immediately recognize your pattern, and pick you off systematically.

But Curran seems to be convinced that Hendershott’s theoretical profits correlate to actual profits in reality: latency arbitrage alone, he says, is worth billions of dollars to high-frequency traders. What’s more, he says, those billions of dollars are “coming out of your retirement account”.

I have to say I’m weirdly impressed by Curran’s sophisticated argument for why these theoretical profits must be costing small investors billions of dollars a year: “every mickle”, we’re told, “makes a muckle”. This argument has the advantage of being unfalsifiable — but, sadly, it’s also complete nonsense. (I especially love the idea that mickles are more likely to become muckles “in a tough market”, whatever that’s supposed to mean.)

The fact is that “the little guy” has never had better execution than he has right now. To oversimplify wildly, let’s divide Wall Street into two groups: the sell side, the price-makers who provide liquidity, and the buy side, the price-takers, who simply decide whether to accept the market’s offer or not. If you’re looking at the current bid-offer spread on a stock (also known as NBBO, for national best bid/offer), then the bid is the best current price at which a sell-side firm will buy the stock from you, while the offer is the price you’ll have to pay to buy it. The difference between the two prices, these days, is lower than it has ever been, and small investors can normally buy or sell as much of any given stock as they like, right at NBBO, with execution in a fraction of a second. That wasn’t the case ten years ago.

Looked at through Curran’s eyes, the “little guy” is always a price taker. He doesn’t go out there into the market posting offers and waiting to see whether anybody will hit them; he just looks to see what offers there are, and if he likes the price being offered, he takes it. That kind of investor — and there are a lot of them out there — has never had it so good, precisely because there are so many HFT shops these days, competing to provide liquidity to the buy side and to receive the small sums of money that exchanges pay to the price-makers rather than the price-takers.

High frequency trading, along with its close relative decimalization, has been fantastic for price takers. They get better prices, they get them faster than ever, and the transaction costs associated with a “round trip” — buying a position and then selling it again — have never been lower. There’s some debate about whether it’s easier or harder than it used to be to trade in size; the jury’s still out on that one, but technology like dark pools has helped there, too. And if you’re big, then there are no shortage of VWAP algorithms and the like which you can use to try to beat the HFT bots at their own game.

Curran disagrees, and cites another paper — this one by Michael Wellman and Elaine Wah University of Michigan. (He didn’t link to this one, either, but a bit of googling found it here.) “Like others before them,” writes Curran, “Wellman and Wah’s study found latency arbitrage was eating investor profits.”

In fact, the Wellman-Wah paper finds no such thing: it’s not an empirical paper at all, and makes no attempt whatsoever to quantify investor profits, be they real or foregone. Instead, it’s an entirely theoretical thought experiment, where an “infinitely fast arbitrageur profits from market fragmentation” at the expense of “zero-intelligence trading agents”.

It’s easy to agree with Wellman and Wah that if there were a lot of risk-free latency arbitrage going on, then the victims would be “zero-intelligence trading agents”, or, as Larry Summers likes to call them, noise traders. But there’s a lot more to HFT than “rent space in a co-located server rack, find risk-free latency arbitrage opportunities, profit!” And while there are, still, idiots (look around), there are fewer of them than there used to be during the go-go day-trading days of the late 1990s. They learned their lesson during the dot-com bust, and with the rise of HFT there are very few small investors left who really believe they’re competing on a level playing field.

Note that it’s traders who lose money when HFT bots make money; if you’re a buy-and-hold investor, you really don’t care what’s going on behind the scenes at all. You just want the best execution for your orders — and right now, in general, execution for small investors is excellent. If you’re day-trading leveraged ETFs, on the other hand, then you’re basically just gambling: intraday moves are essentially random. Those people, over time, will end up losing money to the high-frequency traders.

Still, Wellman and Wah — and Curran — are concerned enough about the plight of the zero-intelligence trading agents that they propose a solution to this problem:

The authors suggest that the perpetual motion tape be replaced by a stop-motion tape. Instead of a continuous, free-for-all market, the session would take the form of a series of lightning-fast-auctions at intervals of a few milliseconds. This would give exchanges a reasonable amount of time to disseminate information (most only take a few thousandths of a second to catch up on the “direct access” feeds). It would also give traders a reasonable amount of time to place bids and offers on a given stock. The average investor would not see the difference because prices on active stocks would still be changing many times per second.

I’ve proposed something similar myself — last year, I said that a stock market where there was a mini-auction for every stock once per second would cause no measurable harm to investors, and would make the stock market as a whole less brittle. I’m no fan of HFT, and a discontinuous market would indeed put a stop to most of its excesses.

But what kind of a world do we live in when someone like Curran can claim with a straight face that “a few milliseconds” is “a reasonable amount of time to place bids and offers on a given stock”? Or where being able to see a stock price “changing many times per second” is considered an important feature of any stock market? The answer actually tells you a lot about the real financial victims of HFT. If you’re just a bystander, looking at stock prices changing many times per second, then you are not losing money to the algobots; in fact, you probably benefit from them, when you make a trade. If, on the other hand, you are not a high-frequency trader but you are the kind of person who thinks that “a few milliseconds” is “a reasonable amount of time to place bids and offers on a given stock”, well, then in that case you might indeed be a victim of the HFT crew: you’re trying to compete with them, and you’re probably losing.

The point here is that making improbable claims about the costs of HFT to small investors is not going to get you very far. The real costs of HFT are found in fat tails and systemic risks and the problems that are endemic to ultra-complex systems. It would certainly be rhetorically very neat and easy if we could plausibly declare that small investors are being hurt by high-frequency traders. But the truth is that they’re not: they’re actually being helped by HFT. It’s the market as a whole which is being put at risk by these algorithms, not the “little guy”. And while I’d welcome a move to a discontinuous market, I don’t for a minute think that such a move would save small investors any money at all, let alone billions of dollars.

*Update: Eric Hunsader has found a cached version of the paper, and — as I suspected — it doesn’t say anything like what Curran said it says. Hendershott did not apply any kind of trading strategy to Apple’s price history on May 9, and did not come up with $377,000 in theoretical profits by doing so. Here’s the relevant bit of the paper, which tries to recreate a “synthetic” NBBO and then compare it to the official SIP NBBO:

For 3.51 milliseconds of each second the SIP NBBO and synthetic NBBO differ. This could result in a buy or sell market order going to the wrong market roughly half that often: 0.175% of the time. Figure 5 shows that the average price dislocation is $0.034. Simply multiplying this times the percentage of the time a dislocation occurs yields an expected price dislocation of $0.006 per 100 shares for a market order entered randomly throughout the day. Multiplying this dollar amount by Apple’s May 9 trading volume of 17,167,989 shares yields $942, representing 0.001 of a basis point of dollar volume traded. This suggests that investors randomly routing market orders are unlikely to face meaningful costs due to data latency.

Yep, the “cost to the little guy” is not $377,000 per day; in fact, according to the paper, it’s just $0.006 per 100 shares, or one thousandth of a basis point. Which adds up to a whopping $942 per day. None of which can be captured by latency arbitrage. Hendershott’s conclusion is not that the little guy is losing out on billions: it’s that “investors randomly routing market orders are unlikely to face meaningful costs due to data latency”.

So where does the $377,000 figure come from? It comes from hypothetical latency traders trying to pick off other (equally hypothetical) active traders who do things like place orders in dark pools at the NBBO midpoint:

Assume BATS updates AAPL bid price from $530 to $531, and the ask price remains at $532. This changes the mid-price from $531 to $531.5. In the first 1.5 milliseconds, slower traders are not aware of the price change. If some such regular traders have placed an order to trade at mid-price in a dark pool, then a faster trader can buy the stock at $531 in dark pool when the synthetic NBBO gets updated. After 1.5 milliseconds, the trader can sell it for $531.5 in the dark pool. In this case the trade gains 50% of the price dislocation. Dark pools represent roughly 11% of trading volume, corresponding to 1,888,478 share of AAPL on May 9. If half of the average dislocation of 0.034 cents is captured on this volume then the fast trader would make a profit of $376,900 in a single stock on a single day. While Apple is one of the highest-volume stocks and this almost certainly represents an upper bound on the profits of strategies based on latency, the dollar figure illustrates the possible magnitude of profits and costs stemming from latency for traders continuously in the market.

Or, in English: if there are people trading continuously in the market who don’t have low-latency feeds, and those people are using the NBBO to determine their trading strategy, then those people can get picked off by hypothetical HFT bots. But clearly, those people are not “the little guy on Wall Street”, and no one in reality is making anything like $377,000 a day from HFT.

In fact, even the tiny $942 figure doesn’t represent HFT profits, it just represents potential losses for small investors. Curran has completely misrepresented Hendershott’s paper. His entire story, including the headline, is basically just false.

Update 2: The paper itself is still hosted on Henderson’s website here, he just doesn’t link to it from his list of publications.


Thanks for a very informative article. I was under the impression that the main problem with HFT was that it had wiped out traditional market-makers, so no one is on the hook to provide meaningful quotes during times of market disruption. So all liquidity can suddenly dry up and cause a “flash crash”.

I guess that idea is inherent in the following quote?
“The real costs of HFT are found in fat tails and systemic risks and the problems that are endemic to ultra-complex systems.”

I’m trying to get my head around this topic. Could one say that this is a failure of regulators, who allow HFTs to make profits as market makers, but don’t force them to take on the associated obligations?

Posted by ADZimm | Report as abusive

The problem with high frequency trading

Felix Salmon
Oct 6, 2012 16:23 UTC

Last night, on BBC Radio 3, I was featured reading an essay about high frequency trading. I hope it’s fun to listen to, but if you want to read it, here you go.

One of the many consequences of global warming is that it’s now, for the first time, possible to drill under the sea bed of the Arctic ocean. The oil companies are all there, of course, running geological tests and bickering with each other about the potential environmental consequences of an oil spill. But they’re not the only people drilling. Because there’s something even more valuable than oil just waiting to be found under the Arctic.

What is worth so much money that three different consortiums would spend billions of pounds to retrofit icebreakers and send them into some of the coldest and most dangerous waters in the world? The answer, of course, is information.

A couple of days ago, I called a friend in Tokyo, and we had a lovely chat. If he puts something up on Twitter, I can see it immediately. And on the web there are thousands of webcams showing me what’s going on in Japan this very second. It doesn’t look like there’s any great information bottleneck there: anything important which happens in Japan can be, and is, transmitted to the rest of the world in a fraction of a second.

But if you’re a City trader, a fraction of a second is a veritable eternity. Let’s say you want to know the price of a stock on the Tokyo Stock exchange, or the exact number of yen being traded for one dollar. Just like the light from the sun is eight minutes old by the time it reaches us, all that financial information is about 188 milliseconds old by the time it reaches London. That’s zero point one eight eight seconds. And it takes that much time because it has to travel on fiber-optic cables which take a long and circuitous route: they either have to cross the Atlantic, and then the US, and then the Pacific, or else they have to go across Europe, through the Middle East, across the Indian Ocean, and then up through the South China Sea between China and the Philippines.

But! If you can lay an undersea cable across the Arctic, you can save yourself about 5,000 miles, not to mention the risk of routing your information past a lot of political flash points. And when you’re sitting in your office in London and you get that dollar/yen exchange rate from Tokyo, it’s fresh from the oven, comparatively speaking: only 0.168 seconds old. If everybody else is using the old cables and you’re using the new ones, then you have somewhere between 20 milliseconds and 60 milliseconds when you know something they don’t.

Those are periods of time so short that humans can barely notice them. This essay, for instance, is about 900,000 milliseconds long, and it takes me hundreds milliseconds just to say the word “cable”. Which is a word with more than one meaning. To you, it probably means some kind of wire. But to City traders, it means 1.6254, or something very close to that number. Because in the City, “cable” means the pound/dollar exchange rate. And it’s named that after a transatlantic cable which was used to telegraph the exchange-rate information from London to New York as far back as 1858.

So what we’re talking about here is nothing new, in terms of kind. Nathan Rothschild built a significant chunk of his fortune by using a system of couriers who told him the result of the Battle of Waterloo a full day before anybody else in London knew it. And my own employer, the Reuters news agency, was founded on sending financial information between Brussels and Aachen using carrier pigeons.

What’s new is that billions of pounds can be made by having access to information not a day in advance, or an hour, or even a second, but even just a millisecond or two. Stock exchanges aren’t physical places where human beings bargain with each other any more: they’re racks of computers in places like Mahwah, New Jersey, where the cables are carefully measured to be exactly the same length so that no one has an infinitesimal advantage thanks to the amount of time it takes information to travel an extra few millimeters down a wire.

Obviously, only computer algorithms can make money from an information advantage which is measured in milliseconds. It’s computers which are making the decisions to buy and sell: if they had to wait for a human to sign off on those things, they’d never make any money at all. That’s a little bit scary, and not only because of the classic science-fiction stories where computers become so sophisticated that they gain consciousness and start waging battles against the humans who built them.

The more obvious problem with exchanges run by computers is that computers don’t have any common sense. We saw this on the 6th of May, 2010 — the day of the so-called “flash crash”, when in a matter of a couple of minutes the US stock market plunged hundreds of points for no particular reason, and some stocks traded at a price of just one cent. It was sheer luck that the crash happened just before 3pm, rather than just before 4pm, and that as a result there was time for the market to recover before the closing bell. If there hadn’t been, then Asian markets would have sold off as well, and then European markets, and hundreds of billions of pounds of value would have been destroyed, just because of a trading glitch which started on something called the e-mini contract in Chicago.

Most of the trading on US stock exchanges is done by something called algobots, these days. These are algorithms: they’re computers which are programmed to put in orders, take out orders, trade in big size, trade in small size – all according to very sophisticated rules, called algorithms. And one of the ironies about the flash crash is that it was actually caused in large part by algobots not trading. The US has over a dozen different stock exchanges, places where stocks are bought and sold. Most of us have only ever heard of the listing exchanges, the New York Stock Exchange and the Nasdaq. But there are many more you probably haven’t heard of, with names like Arca and BATS, as well as sinister-sounding things called Dark Pools. What happened in the flash crash is that when the trading got completely crazy, the algobots just switched themselves off. This was something they weren’t used to, they didn’t know how to react, and so they just went away. And there was suddenly no liquidity in the market. No one was offering to trade. And with no one offering to trade, the prices just plunged, all the way down to one cent. Because there were no bids in the market any more.

The algobots can be very useful, on a day-to-day basis. If a normal person like me buys a few shares in some company or other, that trade doesn’t even happen on any stock exchange at all. It just happens directly with a broker, an algobot, who’s happy to take the other side of my trade because small individual investors like me are normally pretty stupid, and tend to buy high and sell low.

In any case, if any given stock exchange is an incredibly complicated thing, the fragmentation of the stock exchanges has created a much more complex system yet. Most big banks and stockbrokers — and the algobots they control — have access to all of the different exchanges, and they trade wherever they think they can get the best prices. Since the best prices tend to be found wherever the most traders are trading, you end up with something a bit like six-year-olds playing football: everybody’s running towards the ball at the same time. And the result is these huge waves of activity, where traders move en masse, from one stock exchange to the next, in very unpredictable ways. If you layer that unpredictability on top of the complexity inherent in any system of multiple stock exchanges, you end up with something which will almost certainly break in a pretty catastrophic manner at some point. We don’t know how, and we don’t know when, but there’s an ironclad rule of any system: the more complex it is, the less predictable it is, and the more likely it is to fail catastrophically in some unforeseeable manner.

If Twitter fails, that’s fine. A bunch of people get annoyed, and then they want to express how cross they are on Twitter, and then they remember that they can’t, and that makes them even more annoyed. But little actual harm is done. If the stock market fails, on the other hand, or the bond market, or the foreign-exchange market, or the oil market, that’s really, really bad news. Billions or even possibly trillions of pounds could evaporate.

And that’s the biggest reason why it’s time to start cracking down on high-frequency trading. Virtually every major financial center in the world is trying to work out what to do and how to do it: these decisions aren’t easy, partly because any crackdown on the algobots is likely to have its own significant up-front costs.

After all, high-frequency trading has been genuinely wonderful for small investors like you and me. We might not be particularly clever, but when we put in an order to buy this or sell that, the order gets filled immediately. We pay almost nothing in trading costs — just a few pounds, normally. And we get the very best price in the market: something called NBBO, for “national best bid/offer”. If you look at all the prices being quoted on all of the stock exchanges in the country, we get the lowest price of all if we’re buying, and the highest price of all if we’re selling.

That wasn’t true ten years ago. During the dot-com boom, especially, small investors generally had no idea how much they were going to end up paying for a stock they wanted to buy, and all too often their trades could take minutes or even hours to get filled. Today, all individual investors get filled in a fraction of a second: we’ve never had it so good. So if anybody tells you that high-frequency trading is bad for the little guy, and that it means there isn’t a level playing field any more, they don’t know what they’re talking about. Yes, high-frequency traders do make money from small investors, but they do so honestly, just by assuming that whatever those small investors do, the opposite thing is likely to make money. As a result, there’s always someone willing to take the opposite side of the trade whenever you want to buy or sell a stock.

This is a real improvement, which means that the rise of high-frequency trading had genuinely beneficial effects between, say, 2002 and 2007. In those years, the computers helped markets to become ever more efficient and liquid — and they were just in time, too. When the financial crisis came along in 2008, bond markets seized up, but the world’s stock markets actually came through with flying colors. They did what markets are supposed to do: they went down when people were selling, and they kept on falling until they were so cheap that people started buying again. If you wanted to sell, you could always sell, and if you wanted to buy, you could always buy. We take these things for granted, but creating a system which stays that liquid, all the way through such a big crisis, is a real achievement, and the algobots deserve a lot of credit there. After all, imagine what would have happened if you had to phone up your broker at Lehman Brothers in order to sell your shares.

But if you look at what’s happened over the past five years, since 2007, the benefits of high-frequency trading have pretty much plateaued. And the downsides are becoming more and more obvious. There was the flash crash, of course, and then there was the implosion of Knight Capital, one of the biggest and most respected high-frequency trading shops, which released a faulty algorithm one morning and was almost bankrupt an hour later, after losing somewhere in the region of $10 million per minute. If that could happen to Knight, it could happen to anybody. Then there was the botched flotation of one of the stock exchanges, BATS. Once again, its algorithms turned out to be not up to the task. And this was in an expected, rather than an unexpected, situation.

There are more subtle signs, too, which are if anything even more worrying. For instance, look at stock-market volume — the amount of money which changes hands every day. That’s going nowhere: if anything, it’s going down, even as high-frequency traders get bigger and bigger. That says two things.

The first is that real-money investors, the people who the market needs the most, are being scared away by the algobots, because even if the bots are good for the little guy, they’re really bad for big, institutional investors. For big investors, the stock market is more of a rigged game now than it has been in a long time – and they’re taking their ball and they’re going home.

The second reason that volumes are dropping is that the algobots are getting so sophisticated at sparring with each other that they’re not even trading with each other any more. They’re called high-frequency traders, but maybe that’s a misnomer: a better name might be high-frequency spambots. Because what they’re doing, most of the time, is putting buy or sell orders out there on the stock market, only to take those orders back a fraction of a second later, and replace them with new ones. The result is millions of orders, but almost no trades.

I’ll give you one example from a stock with the ticker symbol EFZ. It doesn’t matter what that ticker represents: the computers certainly don’t care. On September 11, between 6:51 and 7:08 in the morning, the US stock markets saw more than 280,000 quotes to trade EFZ. And how many times did it actually trade? Zero.

I can even demonstrate what that kind of thing sounds like. If you map all those offers to buy or sell onto a piano, and play them back, you get something which sounds like this. Remember, each note is an offer to buy or to sell; there’s no actual trading going on.


There’s no value being created here. If the economics of high-frequency trading means that fiber-optic cables get laid under the Arctic ocean, that’s good for everybody. But if it all just devolves into meaningless noise, then something has gone very, very wrong. Especially since the more noise and complexity you have, the bigger the danger that everything could just implode for some unforeseeable reason. Any one of these notes has the potential to be the butterfly wing-flap which results in global disaster. If they’re not doing anybody any good, it makes sense that regulators should crack down on them.



Well done, Felix. I thoroughly enjoyed it.

The EFZ example can be explained as algo-bot warming up with test orders, this is quite common in pre-trading hours, especially when a new algo is just deployed into production. “Flex the muscle” and get ready to go.

It’s common practice for brokerage houses to submit test orders in the morning to make sure all systems are ready. Of course, they don’t usually submit that many orders.

Many bots also need calibration in a live setting with many “test” orders to adjust latencies, optimize trading paths, etc. first before market opens. So…
is that wrong?

Posted by AllenZ | Report as abusive