The Joint Report on the Treasury Spike: Unanswered Questions, and You Can’t Stand in the Same River Twice
The Treasury, Fed (Board of Governors and NYFed), SEC, and CFTC released a joint report on the short-lived spike in Treasury prices on 15 October, 2014. The report does a credible job laying out what happened, based on a deep dive into the high frequency data. But it does not answer the most interesting questions.
One thing of note, which shouldn’t really need mentioning, but does, is the report’s documentation of the diversity of algorithmic/high frequency trading carried out by what the report refers to as PTFs, or proprietary trading firms. This diversity is illustrated by the fact that these firms were both the largest passive suppliers of liquidity and the largest aggressive takers of liquidity during the October “event.” Indeed, the report documents the diversity within individual PTFs: there was considerable “self-trading,” whereby a particular PTF was on both sides of a trade. Meaning presumably that these PTFs had both aggressive and passive algos working simultaneously. So talking about “HFT” as some single, homogeneous thing is radically oversimplistic and misleading.
But let’s cut to the chase: Whodunnit? The report’s answer?: It’s complicated. The report says there was no single cause (e.g., a fat finger problem or whale trader).
This should not be surprising. In emergent orders, which financial markets are, large changes can occur in response to small (and indeed, very small) shocks: these systems can go non-linear. Complex feedbacks make attribution of cause impossible. Although there is much chin-pulling (both in the report, and more generally) about the impact of technology and changes in market structure, the fundamental sources of feedback, and the types of participants in the ecosystem, are largely independent of technology.
Insofar as the events of 15 October are concerned, the report documents a substantial decline in market depth on both the futures market, and the main cash Treasury platforms (BrokerTec and eSpeed) in the hour following the release of the retail sales report. The decline in depth was due to PTFs reducing the size (but not the price) of their limit orders, and banks/dealers widening their quotes. Then, starting about 0930, there was a substantial order imbalance to the buy side on the futures: this initial order imbalance was driven primarily by banks/dealers. About 3 minutes later, aggressive PTFs kicked in on the buy side on both futures and the cash platforms. Buying pressure peaked around 0939, and then both aggressive PTFs and the banks/dealers switched to the sell side. Prices rose when aggressors bought, and fell when they sold.
None of this is particularly surprising, but the report begs the most important questions. In particular, what caused the acute decline in depth in the hour leading up to the big price movement, and what triggered the surge in buy orders?
The first conjecture that comes to mind is related to informed trading and adverse selection. For some reason, PTFs (or more accurately, their algos) in particular apparently detected an increase in the toxicity of order flow, or observed some other information that implied that adverse selection risk was increasing, and they reduced their quote sizes to reduce the risk of being picked off.
Did order flow become more toxic in the roughly hour-long period following the release of the retail number? The report does not investigate that issue, which is unfortunate. Since liquidity declines were also marked in the minutes before the Flash Crash, it is imperative to have a better understanding of what drives these declines. There are metrics of toxicity (i.e., order flow informativeness). Liquidity suppliers (including HFT) monitor it in real time. Understanding these events requires an analysis of whether variations in toxicity drive variations in liquidity, and in particular marked declines in depth.
Private information could also explain a surge in order imbalances. Those with private information would be the aggressors on the side of the net imbalance. In this case, the first indication of an imbalance is in the futures, and comes from the banks and asset managers. PTF net buying kicks in a few minutes later, suggesting they were extracting information from the banks’ and asset managers’ trading.
This raises the question: what was the private information, and what was the source of that information?
One problem with the asymmetric information story is the rapid reversal of the price movement. Informed trades have persistent effects. I’ve even seen in the data from some episodes that arguably manipulative (and hence uninformed) trades that could not be identified as such had persistent price impacts. So did new information arrive that led the buyers to start selling?
A potentially more problematic explanation of events (and I am just throwing out a hypothesis here) is that increased order flow toxicity due to informed trading eroded liquidity, and this created the conditions in which pernicious algorithms could thrive. For instance, momentum triggering (and momentum following) algorithms could have a bigger impact when the market lacks depth, as then smallish imbalances can move prices substantially, which then triggers trend following. When prices get sufficiently out of line, these algos might turn off or switch directions, or other contrarian algorithms might kick in.
These questions cannot be answered without knowing the algorithms, on both the passive and aggressive sides. What information did they have, and how did they react to it? Right now, we are just seeing their shadows. To understand the full chronology here–the decline in depth/liquidity, the surge in order imbalances from banks/dealers around 0930, the following surge in aggressive PTF buying, and the reversal in signed net order flow–it is necessary to understand in detail the entire algo ecosystem. We obviously don’t understand it, and likely never will.
Even if it was possible to go back and get a granular understanding of the algorithms and their interactions, this would be of limited utility going forward because the emergent ecosystem evolves continuously and rapidly. Indeed, no doubt the PTFs and banks carried out their own forensic analyses of the events of 15 October, and changed their algorithms accordingly. This means that even if we knew the causal connections and feedbacks that produced the abrupt movement and reversal in Treasury prices, that knowledge will not really permit anticipation of future episodes, as the event itself will have changed the system, its connections, and its feedbacks. Further, independent of the effect of 15 October, the system will have evolved in the past 9 months. Given the dependence of the behavior of such systems on their very fine details, the system will behave differently today than it did then.
In sum, the joint report provides some useful information on what happened on 15 October, 2014, but it leaves the most important questions unanswered. What’s more, the answers regarding this one event would likely be only modestly informative going forward because that very event likely caused the system to change. Pace Heraclitus, when it comes to financial markets, “You cannot step twice into the same river; for other waters are continually flowing in.”