Sleepless in Seattle: Is Living with Threat of Market Flash Crashes the New Normal?

The New Yorker magazine published a story this week about the inevitable destruction-by-earthquake of the Pacific Northwest. The frightening mood of the story is of course enhanced by the impossibility of predicting when this event will occur. The story reminded me of some of the debates around high-frequency trading (HFT). With modern market structure being built on this HFT fault line, commentators are warning of the increased fragility and risk of this system. However, with no way of predicting the crash, it is difficult to justify making sweeping systemic changes or even determining what these changes should be, in the face of markets with historically low spreads and deep liquidity. The result is largely business-as-usual with some work on circuit-breakers and other system fixes — the equivalent of tightening up building standards in Seattle rather than building somewhere else in the first place.

However, tremors occasionally occur — the latest of which was the US Treasury market flash crash on 15 October 2014. On that day, between 9:33 and 9:45 a.m., the 10-year yield made a 16-basis-point round-trip for no apparent reason. This is a big move for the Treasury market. Similarly to equities markets, the Treasury market has seen an increase in electronic trading (particularly for on-the-run securities — those most recently issued — since these are especially liquid).

The US Department of Treasury recently released a report into this “flash crash.” The key takeaway from the report is that no single cause (or, arguably, any cause) was found for this event. However, some points raised in the paper are interesting:

• The report finds that the share of HFT liquidity provision increased during the event window (from around 50%–65% to 75%) with bank dealers withdrawing in relative terms (from around 30%–40% to 15%–20%).
• Two distinct HFT strategies were in operation — passive HFTs provided market making activity during the event window while aggressive HFTs traded in the direction of price moves and accounted for the bulk of the observed trading imbalances on the way down and on the way up. This confirms the existence of two distinct types of HFT — “good” market-making HFTs and “bad” predatory HFTs, which I wrote about in this blog post.
• HFTs quoted tight spreads during the event window and managed risk by reducing quoted sizes. On the other hand, bank-dealers that remained in the market managed their risk by quoting higher spreads. This is an interesting finding and shows the value of having heterogeneous participants in the markets, something that will be the focus of a forthcoming CFA Institute research paper.
• The crash was not associated with any significant net change in positions by any market participant, neither was there any evidence for “fat-finger error” nor any unusual market orders.

These US Treasury market findings will be very familiar to students of the S&P “E-mini” flash crash. In fact, to a first approximation, they appear the same. I suspect looking for economic or order-flow causes for these flash crashes is a dead end.

Answers might be found in an interesting paper, previously cited in one of our blog posts, and since published in Nature, by Johnson et al. (2013). The authors look at HFT from a systems perspective rather than a financial one. Many financial markets have essentially transitioned from being based on human interactions to being based on machine interactions on machine timescales. Once humans become unable to respond to or intervene in events, the authors argue that the dynamics of the market change and instantaneous price crashes and spikes, over very short durations, happen more frequently.

The authors examined these dynamics and noted that the instantaneous reversals from the crash or spike are likely caused by automatic exchange responses or predatory algorithmic traders. Interestingly, they found that the cumulative amount of such “mini black-swan” events (those that are unexpected and extremely difficult to predict) rose exponentially leading up to the financial crisis of 2008 and has continued a linear rise since then (although at a lesser rate of increase). The authors interpret this finding as suggesting that, as in engineering systems more generally, a prevalence of micro-events in financial markets can accompany and possibly precede large changes, such as a market-wide flash crash, in the same way that tiny cracks in an aeroplane wing will lead to metal fatigue and ultimately failure.

Are market flash crashes foreshocks of “the really big one” to come?

High-frequency trading, like the Pacific Northwest, is probably here to stay. It seems to me unlikely that it will be possible (or desirable) to reconstruct the market in a fundamentally different way. More likely, markets will have to learn to live in an earthquake zone with the certainty that random crashes will happen for no predictable (even ex-post) reason, and that “the big one” may be around the corner. Like civil engineers, maybe the best we can do is ensure the systems’ foundations are designed to ameliorate the damage.

If you liked this post, consider subscribing to Market Integrity Insights.

Image Credit: istockphoto.com/SteveCollender