So let’s make you unbearably jealous first. Even if you had never been so before. Even if you are a self-proclaimed cheerer-on-of-others’ success:
1,238 trading days with a single negative day. That is Virtu, 2009–2013.1 And at our high frequency shop, too, we had 2.5 years of straight profitable days, with a single slight negative at the very very end.
Wouldn’t you love to have come up with that strategy?? Come on now, admit it, you are strategy-jealous and it is eating away at your strategy builder ego... The green-eyed Halloween monster has been awakened, as I present to you the ultimate financial boogeyman.
Now you, too, face the visceral sensation that I think is often behind the villainization of High Frequency Trading. The reason is that envy, and the fact that a low-profile tech company that tries to stay under the radar is an easy target, especially when the explanations of a breakdown are not simple.2 Add a bestseller, and Michael Lewis on 60 Minutes in 2014: “The United States stock market, the most iconic market in global capitalism, is rigged.”3 Hm... And just like that we should all be believers.
Which brings me to the misunderstanding that I discuss in an upcoming podcast about what a high frequency trading shop really is. It is a technology company first and foremost, with trading as an output. You don’t need to take my word for it, though. Read Virtu’s IPO prospectus:
“Technology is at the core of our business.”
(Incidentally, we know about those 1,238 days because Virtu had to publish that prospectus.)
NOT a money printing business
But I cannot get to 99.9% profitable days, so what is this if not rigged, you say? The classic unfair advantage, isn’t it?
Before you Jump4 to the obvious conclusion, let’s look at the numbers a little closer.
Historical estimates put US equity HFT revenues at about $7.2 billion in 2009 and $1.3 billion in 2014.5 Four fifths disappeared, and these are revenues, not even profits. That is one hell of a whimsical printer to me.
So where does the money they “rake in” actually go? Back to Virtu, 2013, the last full year in that prospectus:
Revenue: $664.5 million.
Brokerage, exchange and clearance fees, net: $195.1 million.
Communication and data processing: another $64.7 million.
The entire payroll, for 151 people: $78.4 million.
Nearly 40 cents of every revenue dollar went to the plumbing. Brokers, exchanges, clearing houses and data vendors collectively took more than three times what the firm paid all its people.
Yes, that is about $4.4 million in revenue per employee. And even at only $519,000 per head, nobody is asking you to feel sorry for them.
So the strategy you were jealous of a few paragraphs ago? It comes with quite an infrastructure spend. In the meantime the edge on any single trade is minuscule.
I think it is an open secret I can share—the last year or so at our little HFT shop we made all the money on rebates. Only rebates! Without rebates we were losing money; consistently.
So what makes the daily numbers add up across the whole HFT operation is: turning trades over an enormous number of times, hedging instantly, and unwillingly paying for the piping. Rinse and repeat, hundreds of times before you blink.
The daunting task of trying to win
And then there is the cut-throat competition: the advantage doesn’t stay yours for long. Spread Networks built a fibre route from Chicago to New Jersey at an estimated cost of $300 million, bringing the round trip down to about 13 milliseconds. From 16! Just a little later over-the-air links did it in just over 8 millis.6 Would you be willing to pay that price for an elusive three millisecond gain, and possibly be faced with such a short shelf-life for it? How is the jealousy now?
Additionally, these companies also run on a different kind of risk, not just a humungous bill.
On 1 August 2012, Knight Capital’s systems sent more than 4 million orders into the market while trying to fill just 212 customer orders. Long story short: the eventual loss exceeded $460 million.7
Our single negative day was under 200 bucks. Knight’s was $460 million before most people had finished their coffee.
In this business the market can be less dangerous than your own error. Or your stupidity.
Where the criticism may be fair
Make no mistake, the races are real. And some of the criticism deserves more than an eye roll.
When a price moves elsewhere, a resting quote can become stale. Its owner races to cancel it; someone else races right ahead to trade against it. You can see why the person getting picked off would be pissed.
Aquilina, Budish and O’Neill measured these races on the London Stock Exchange. The average FTSE 100 stock had 537 a day, roughly one a minute. The most common winning margin was just 5 to 10 microseconds.
They estimated a “tax” of about 0.42 basis points on trading volume. Extrapolate their findings to global equity markets, and the authors put the sums at roughly $5 billion a year.8
That is a real cost to liquidity provision. And if you are spending millions to win the next few microseconds, it is fair to ask how much better a market that buys for everyone else.
A little pushback on academia
Measuring the cost of picking off a stale quote is fair and useful. However, treating that as the verdict on HFT is a rather bigger leap.
Even the distinction between liquidity-providing HFT and the “sniper kind” needs more care.9 The same shop can make a market here and pick off someone else’s stale quote there. The good guys and the bad guys very often share a payroll. I made a related point in Not All Informed Trading Is Research.
If the question is how to improve today’s market, then comparing it with a better version of itself makes sense. But if the conclusion is that HFT made markets worse, I would want the market that preceded it measured, too.
The cost of a particular activity and the overall effect of a technological shift are different questions. Finding the former does not settle the latter.
I have already argued with some of the proposed cures in the batch-auction piece. Let’s leave those there for now, and look in the rear view mirror for a snap second.
The pre-HFT world, in numbers and headlines
The fairest question isn’t “is HFT perfect?” It’s “compared to what?”
Nasdaq, early 1990s. Many of the largest, most active stocks were quoted almost exclusively in even eighths, keeping their spreads at a quarter of a dollar or more. Christie and Schultz exposed the pattern in 1994. The Justice Department charged 24 major Nasdaq firms, and the separate class action settled for $1.027 billion, the largest antitrust class-action recovery up to that point.10
NYSE, 2004. Five specialist firms paid $241.8 million in penalties and disgorgement for trading ahead of their own customers’ orders and inserting themselves between buyers and sellers.11
That is the market electronic trading inherited, and competed away. Regulation, decimalization and competition all helped the process. When the NYSE automated quote dissemination in 2003, Hendershott, Jones and Menkveld found that the resulting increase in algorithmic trading narrowed spreads and reduced adverse selection, particularly in large stocks.12
By 2024, the SEC was adopting rules for half-penny quote increments in qualifying stocks.13 We have come from defending quarter-dollar spreads to debating half-penny increments. An efficiency shift, if there ever was one.
And the people changed, too.
The people I met from that older world of soft dollars and the like all seemed, well, rather dodgy. They would want to meet at the office, and then automatically offer to take me to a strip joint afterwards — obviously paid by the company (one could argue: by the spread). As if it were the most natural thing in the world. Simply the next item on the agenda.
The new generation of CTOs and CEOs were very different. Business and technology. What you knew, what you could build. None of the “restaurant” trips. Technical prowess and knowledge first, resulting gains next.
As big a shift as the spread reduction, if not bigger, if you ask me.
Are the boogeymen really the HFTs? Or rather those old-timers with their shifty eyes, luring you to the dark side?
You may want the odd Gordon Gekko14 for diversity and fun, I get it, but I can assure you: you wouldn’t like to trade in a market where all your counterparties and CEOs are Gekkos.
Virtu Financial, Form S-1, 2014. The daily record measures Adjusted Net Trading Income, before various corporate expenses. Financial figures are for 2013; 151 is year-end headcount.
The SEC/CFTC Flash Crash report describes a large automated sell programme followed by HFT activity that amplified the disruption. The explanation is rather less convenient than a single villain.
Jump Trading, a major quantitative trading firm.
Congressional Research Service, 2014, citing TABB Group. Historical industry estimates; the 2014 figure was an estimate published during that year.
Budish, Cramton and Shim (2015) report the estimated $300 million construction cost and a reduction from 16 to 13 milliseconds. McKay Brothers (2013) reported microwave round trips of 8.25–8.33 milliseconds.
SEC’s Knight Capital enforcement announcement. The positions accumulated in 45 minutes; the eventual loss followed their unwinding.
Aquilina, Budish and O’Neill, Quantifying the High-Frequency Trading “Arms Race”. The global figure is an extrapolation from UK data. The 5–10-microsecond figure is the most common winning margin, not the average. Races are less frequent in FTSE 250 stocks, not absent; the global estimate assumes the UK results generalize.
Ibikunle, Moews, Muravyev and Rzayev, Data-Driven Measures of High-Frequency Trading, associate liquidity-providing HFT with greater information acquisition before announcements, and liquidity-demanding HFT with less. Their categories describe activities, not mutually exclusive firms; liquidity demand is also broader than sniping. But information arriving early does not tell you how it was acquired. Honest research? An illegal tip? The metric does not distinguish. If the advantage came from inside information, eroding it is hardly an attack on honest research.
Hendershott, Jones and Menkveld, Does Algorithmic Trading Improve Liquidity?, Journal of Finance, 2011.
The half-penny framework was adopted in 2024. Implementation has since been deferred until the first business day of November 2027.
Probably the most emblematic wall street character in Oliver Stone’s Wall Street (1987) played by Michael Douglas.