High frequency trading is a wonderful subject for study because nobody agrees on what would make it Good or Bad. So academics and practitioners can write papers about how it is Good, or how it is Bad, and they don't particularly contradict each other because they measure different things and the actual thing that one wants to measure is hard to nail down and would probably be hard to measure even if you knew what it was.
So here we are with an interesting European Central Bank working paper from Jonathan Brogaard, Terrence Hendershott and Ryan Riordan about high frequency trading.* They're mostly for it, which has naturally gotten them some attention. They think that the things you should measure are along the lines of "does high frequency trading improve market price discovery?" and "does it provide liquidity?" and I guess if they thought it was Bad, they'd be asking different questions. But they answer yes to their questions: They find that high frequency trading improves price discovery, and that it does not cause instability by withdrawing liquidity during volatile periods.
You can quibble with these points if that's what you like to do with academic papers, and I don't know what else one would do with them, really. The "improves price discovery" thing comes with the important caveat, "for about three or four seconds"; after that the information is incorporated into the market. So HFT makes markets three seconds more efficient. Is that good? That question plummets quickly into metaphysics and, thus, into this footnote.**
You can quibble with the instability point too, but not up here.***
But here is an oddity. The authors look at what happens when negative macroeconomic news is announced, and then draw this chart:
So: Starting about one second before the bad news is announced, HFT firms are actively selling (that is, demanding liquidity to sell shares -- red line). Starting at around the same time, stock prices are going down (gold line, right axis). On the other hand, starting about two seconds before the bad news is announced, HFT firms are passively buying. That is, they're supplying liquidity: They've posted bids and offers, and their bids are getting hit. That's the gray-green-blue-whatever-ish dotted line. The blue dashed line is the net HFT activity. In aggregate, HFTs buy on negative macro news, and would seem to lose money doing so.
The opposite happens on positive news: HFTs end up selling into positive news and losing money. As with negative news, the price changes and HFT liquidity demand occur about a second after the HFT liquidity supply:
This is broadly consistent with the authors' view of HFT as a useful provider of liquidity -- basically, HFT firms are acting like market-makers here, taking the opposite side of trades that people want to make on news -- but the timing is quite weird. The apparent interpretation is that HFT firms are being beaten to the punch on economic news, by about one second, and losing money because of it. They leave their bids and offers up going into the news, and someone gets the news just before they do,**** and that someone trades against them and makes money off of them.
An important function of high frequency traders, apparently, is to get taken advantage of by people who are just a bit faster than them.
So that's odd, no?
But it seems to be basically true. The authors' sample of "HFT" is a Nasdaq data set of 26 large independent high frequency trading firms, which excludes big broker-dealers like Goldman who have some HFT strategies, as well as proprietary algorithmic trading firms that trade quickly but don't make a business out of trading constantly. Those firms might be more likely to react to fundamental news than the big pure-HFT firms, who trade based mainly on statistical-arbitrage-type market data (prices and order books) than on actual news releases.***** So the speedy algo traders profit off the ever-so-slightly-less-speedy-in-this-particular-case HFT traders.
What can you conclude from that? Well, for one thing, if your view of "high frequency trading" embraces "anyone who trades real fast with a computer," then you may not find this paper's positive conclusions about HFT entirely soothing. They're really about only one category of high frequency traders -- and the other category seems to be trading against them. If this paper's fast computer traders are Good, then it stands to reason that the fast computer traders on the other side might be Bad. The net effect remains murky.
Also, though, we have talked before about enforcement efforts to crack down on the sorts of fast computer traders who get economic data milliseconds before everyone else. At the time I found that crackdown puzzling, since it seems to protect not individual investors but other, slightly slower fast computer traders. If information comes out at 2 p.m. and you[r computer] gets it at 1:59:59.999 and you try to buy with your information advantage, the only person who's selling to you at 2:00:00.006 is another computer. The little guy or Fidelity portfolio manager or whoever is actually reading the news with human eyeballs is whole seconds behind and thus totally safe.
And here you go. The people getting picked off by "high frequency traders," in the loose sense, when news is announced, are "high frequency traders," in the strict sense. Algorithmic market-makers get picked off by algorithmic speculators. A whole financial-markets drama occurs in the blink of an eye, and all you have to do to avoid it is blink.
* Like a lot of things in financial academia, this has been floating around online for a while in various forms, but it came out today under the ECB's quasi-imprimatur, which is a valuable quasi-imprimatur, so now we are talking about it.
** From the paper:
The fact that HFTs predict price movements for mere seconds does not demonstrate that the information would inevitably become public. It could be the case that HFTs compete with each other to get information not obviously public into prices. If HFTs were absent, it is unclear how such information would get into prices unless some other market participant played a similar role. This is a general issue in how to define what information is public and how it gets into prices, e.g., the incentives to invest in information acquisition in Grossman and Stiglitz (1980).
You might wonder what fundamental research HFT firms are doing to ferret out new information and make it public.
A thought experiment might be, if you delayed HFT access to economic events by three seconds, would they still have three-second advantage over non-HFT traders? What if you delayed them by five minutes?
*** The authors look at HFT behavior in the top 10 percent most volatile days, compare it to the bottom 90 percent most volatile days, and find that HFTs provide similar amounts of liquidity. A plausible model of HFT might be "volatility is profitable but blind panic is bad," meaning that you'd supply more liquidity in the 90-to-99.5 tranche of days but much, much less in the top 0.5 percent. That model would lead to yearly, not biweekly, flash crashes, which is kind of what happens, but the paper looks at the more modest top-10-percent-of-volatility-days thing.
**** And before it comes out? The authors use Bloomberg time-stamps to measure second 0 (the time the news comes out), so arguably second -1 or whatever is the actual time it comes out and it takes some time to get up on Bloomberg, but the trading 2 seconds before the measured announcement time does seem weird.
***** I owe this point to a conversation with Terrence Hendershott, one of the paper's authors. Incidentally, if you accept that HFT firms react only to market prices and orders, and not to fundamental news, then that Grossman-Stiglitz point in footnote ** gets especially metaphysical.
This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.
To contact the author on this story:
Matthew S Levine at firstname.lastname@example.org