How big is high-frequency trading?
I have a bit more clarity on the $20 billion figure for total profits from high-frequency trading: it comes from the TABB Group. In a recent publication, TABB’s Robert Iati writes:
TABB Group estimates that annual aggregate profits of low latency arbitrage strategies exceed $21 billion, spread out among the few hundred firms that deploy them. While we know all the large investment banks such as Goldman Sachs are committed to prop trading profitability, the hundreds of smaller, private high frequency prop shops extend much greater influence in the marketplace by providing liquidity that keeps activity flowing.
The Bloomberg article, meanwhile, explains the figure thusly:
The firms compete for a slice of $21.8 billion in annual profits from equities and derivatives market making and arbitrage, according to Tabb. Among the largest are hedge funds Citadel Investment Group LLC, D.E. Shaw & Co. and Renaissance Technologies Corp., as well as the automated brokerages Getco LLC, Hudson River Trading LLC and Wolverine Trading LLC.
When John Hempton, then, says that “quantifications of this as a $20 billion issue are insane”, I think there are two questions: firstly, what is “this”, and secondly, how profitable is it, in aggregate.
It would be most convenient if the HFT algorithms were split nicely into a “trading” bucket and a “quant arbitrage” bucket, so that Hempton could complain mildly about the “trading” algos while saying at the same time that they’re not all that big of a problem, while ignoring the stat-arb shops and other high-frequency, low-latency traders. But in reality there’s very little difference: the traders all have strategies, and the stat-arb strategies are all implemented so as to maximize trading profits.
To put it another way, I don’t think people are making billions of dollars in profit just by being fast. But there are definitely people making billions of dollars in profits through strategies for which being fast is a necessary precondition.
Which leads us to the second question: if you tot up all the profits from high-frequency, low-latency traders, including big shops like Citadel, Renaissance, and Goldman, can you get to $20 billion? My gut feeling is that you can, and that the TABB estimate is not obviously unreasonable.
I also got a note from Jon Stokes yesterday which is worth disseminating more widely:
It’s quite remarkable to me that many of the econ and finance folks who insist that “HFT is the same thing we always did, just way faster” don’t seem to realize that frequency and amplitude matter a whole lot, and that for any given phenomenon when you suddenly increase those two factors by an order of magnitude you typically end up with something very different than what you started with. This is true for isolated phenomena, and it’s doubly true for complex systems, where you have to deal with systemic effects like feedback loops and synchronization/resonance.
What I’ve noticed anecdotally is that engineers and IT pros are more concerned about HFT than people who just handle money for a living. These guys have a keen sense for just how fragile and unpredictable these systems-of-systems are even under the best of conditions, and how when things go wrong they do so spectacularly and at very inconvenient moments (they get paid a lot of money to rush into the office to put out fires at 4am).
There’s an analogy here with e-voting, which I did quite a bit of work on. In the e-voting fiasco, you had people who were specialists in elections but who had little IT experience greenlighting what they thought was an elections systems rollout, but in actuality they had signed on for a large IT deployment and they had no idea what they were getting into. To them, it was just voting, but with computers, y’know? They found out the hard way that networked computer systems are a force multiplier not just for human capabilities, but for human limitations, as well.
This is why I’m sympathetic to Paul Wilmott’s view of all this: there’s simply too much complexity here for comfort, and too many things which can go wrong. When the stat-arb shops imploded in the summer of 2007, the systemic consequences were mild-to-nonexistent, and that does provide a certain amount of reassurance. But we can’t be sure that if and when such a thing happens again, the consequences won’t be much worse.