Lambert: Technology doesn’t crash markets. People who own technology crash markets. But as Mehrling points out, we’re not asking anything about those people. That seems odd. Or…. Wait a minute. Maybe it doesn’t!
Yves: The open secret that Merling may assume that readers know is that HFT is junk liquidity, as in it adds liquidity when markets don’t need it and sucks it away when they do. So the idea that HFT would amplify or even cause a crash should not come as a surprise.
By Perry Mehrling, a professor of economics at Barnard College. Originally published at his website
The official report on events of October 15, 2014, is now public and it makes fascinating reading. Most news accounts of the report have taken its bland no-smoking-gun conclusion at face value, but if you actually read the report a rather clear picture emerges, along with some rather obvious unanswered questions. Bloomberg has the best account, but there is more to be said.
The crash itself involved a very large spike in Treasury prices that was more or less completely reversed in the time interval 9:33-9:45 am on the fateful day, but the event itself is not the important thing. The report is really an investigation of the effect of changing market structure in the Treasury bond market, where high-frequency electronic trading now accounts for a majority of trading, and the vast majority of market depth. Some of this high frequency trading is being done by traditional bank dealers, but most of it is the province of a new breed that the report calls the Principal Trading Firm. The central question is how the introduction of this new animal is affecting the larger trading ecosystem.
The first step toward crash was clearly the sharp reduction in market depth in the hour after release of a negative retail sales number at 8:30am. “The drop in available depth that followed the data release at 8:30 ET was due primarily to a large reduction by PTFs of the limit orders they left standing in the cash and futures markets (Figures 3.15 and 3.16)” (p. 27). Importantly, the PTFs operate largely at the “top of the book”, which is to say near the most recent trade price, so this reduction of limit orders directly implied a larger price move for a given order size.
On the heels of this fragility, the trigger for the crash was a very large volume of messages, mostly trade cancellations, that swamped the matching engine and so increased trade latency (a measure of time delay). Some of these cancellations were in the visible trade book, but the majority were far from the trade price and so invisible to market participants. They couldn’t see the cause but they could see the effect, increased latency, and they could and did feed that effect into the algorithms driving their own trading strategies.
Subsequently, during the “event window”, PTFs initially drove prices up with aggressive buying, and then back down again with aggressive selling. Who was on the other side of these trades? Mostly other PTFs, apparently, who were adopting more passive market-making algorithmic strategies. Indeed bank dealers widened their bid-ask spreads, and there were moments in the upswing when dealer offers disappeared–there was no one willing to absorb buying pressure, so price continued to rise. In essence, the spike was mostly about algorithms trading with algorithms, indeed apparently sometimes algorithms at the very same firm since the percentage of “self-cancelling” trades spiked within the event window!
What caused the reversal? Around 9:39 price got high enough to reveal a large volume of futures sell orders that had previously been invisible to the market, and price went into reverse without ever reaching the level to trigger the sell orders. And so the whole thing went into reverse, PTFs trading with PTFs mostly on the way down, more or less back to where price was before the spike began.
This account begs some questions which unfortunately the report does not answer.
First, what caused the initial reduction in market depth? Two large trades are singled out. “For example, near the beginning of the event window, two buy market orders were executed in the 10-year futures market—one for 3,000 contracts at 9:33:45 and one for 2,100 contracts at 9:34:07—both of which coincided with reductions in market depth” (p. 26). The reader wants to know more, specifically what kind of market animal initiated these and other orders, and why they absorbed so much market depth.
Second, what caused the flurry of cancellations? The report makes clear that these were PTF messages, but the reader wants to know more. Were these cancellations some kind of response to prior reduction in market depth, and how exactly do PTF algorithms respond to observed latency? Not to put too fine a point on it, but is it possible to infer market manipulation?
Third, what about those unexecuted futures sell orders that reversed the tide? Were they initiated by PTFs, or bank dealers, or who?
It is clear that the data exist to answer all three of these questions, and clear also that the report authors had access to that data. So the real question is why are the obvious questions not being asked? Given the multiple authorship, one imagines a process of drafting and redrafting in which any strong conclusion gets edited out by one party or another. So maybe the question is which branch of government is trying to make sure the obvious questions are not being asked?