Anatomy Of A Flash Crashing Black Swan

Let’s see, in 2015 there was the Swiss franc in January, German bunds in April/May, Chinese stocks over the summer, the Chinese yuan on August 11, and then global stocks less than two weeks after that.

I’m talking, of course, about black swan events.

Or perhaps I should say “grey” swan events because while rare, some of them were not entirely unexpected. For instance, anyone who was paying attention could see that China would need to devalue sooner or later in order to preserve the illusion that the export-driven economy was growing at a (near) 7% clip. Similarly, it wasn’t hard to predict that Germany might end up experiencing its own version of 2003’s Japanese VaR shock, given that ECB QE was artificially suppressing volatility and thus allowing vol sensitive investors to lever up in safe haven German debt.

Have a look at the following set of charts (which I’ve shown before) that, when taken together, certainly seem to suggest tail events are happening more often in the post-crisis world:

(Charts: Citi)

What’s causing these 6,7,8,9,…16 sigma events to occur with alarming regularity, you ask? Well, some are policy driven but we can’t forget the role that illiquid markets and batshit crazy, hair-trigger, headline-scanning algos play in creating chaos. Put simply: the market structure is now more vulnerable despite what the HFT lobby will tell you about how great they are at providing liquidity.

While black swans are never far from my mind (these anomalous creatures are indeed something of an obsession), I’m even more inclined than usual to bring them up thanks to the recent turmoil in the British pound.

Over the past 36 hours or so I’ve documented cable’s rapid descent:

Brexit angst was already running high and on Sunday we got an early look at what PM Theresa May is expected to say at a speech to be delivered tomorrow. As those who frequent these pages are no doubt aware, May’s decision to pursue a “hard and clean” split with the EU put the pound under tremendous pressure (although, as more than one analyst has noted, it’s not like this was really a surprise). For those who missed it, have a look at O/N vol:

(Bloomberg)

Here’s the latest from Bloomberg (as of ~1:15 EST on Monday):

GBP/USD now -1% at 1.2060, versus 1.1986-1.2085 range; pair gapped sharply lower with the Friday close at 1.2182

With the U.S. off on Monday, pound liquidity has been an issue after early London trades went through, in both spot and options market; especially in the latter, investors struggled to find offers within ’’reasonable’’ spread from mid-pricing, a Europe-based trader said
Overnight volatility surged above 30% handle for the first time since August ahead of PM May’s speech tomorrow even though an exact time hasn’t been announced yet and options expiring at New York cut off may not capture the event after all

As indicated above, at one point sterling dropped below 1.20 for the first time since the October flash crash an event which, getting back to the topic at hand, was yet another in our long list of black swan events. Here’s a visual reminder of what happened:

(Bloomberg)

The BIS (that bastion of integrity whose board consists of all the world’s most powerful central bankers) recently put together a study of the madness that unfolded in GBPUSD on October 7, 2016. Below, for those interested in reading the “official” explanation, is what happened. Note that this should serve as a cautionary tale going forward – especially considering the fact that given how arduous and drawn out the Brexit process is likely to be (despite May’s push for a clean, swift break), and given how on edge FX markets already are, more multi-sigma events are (paradoxically) a foregone conclusion…

Stage 1: Shortly after midnight British Summer Time (BST), equivalent to Greenwich Mean Time (GMT) plus one hour, on 7 October, trading volumes picked up sharply and sterling began to depreciate against other currencies (Graph 5). Over a period of around eight seconds (00:07:03 to 00:07:11 BST), sterling fell from 1.2600 to 1.2494 against the dollar, based on the Reuters mid-price.3 During this time, around £252 million of GBP/USD (Graph 7) and €52 million of EUR/GBP was traded on Reuters, of which the vast majority represented so-called ‘aggressive’ sales of sterling — pointing to a very significant imbalance in order flow.4 Despite the magnitude of the move and the volumes transacted, GBP/USD bid-offer spreads remained little changed until around 00:07:14 BST and measures of the price impact of transactions over this period were relatively low.

At 00:07:13 BST, the Financial Times (FT) published an article entitled “Hollande demands tough Brexit negotiations” on its website. The story was repeated on a variety of news wires. Market participants note that this release would have been interpreted as somewhat sterling-negative, but it did not represent new information. The comments from President Hollande were made at a widely attended event earlier that evening and had featured in similar form on a variety of well known news websites (including in English) prior to the publication of the FT article. Indeed, as can clearly be seen in Graph 5, the initial moves preceded its release. However, in the midst of the broader moves, it is conceivable that at the margin this may have acted to exacerbate the existing volatility.

Stage 2: At 00:07:15 BST, the CME triggered its velocity logic mechanism, which pauses trading for 10 seconds on the futures exchange, in response to the large moves in the preceding two seconds. At this point, bid-offer spreads in the spot market widened significantly. It is not possible to disentangle the impact of the trading pause on the CME, if any, on spot market liquidity from the impact of the preceding large price moves. After reaching 1.24 on Reuters (at 00:07:15 BST), GBP/USD accelerated its fall and market functioning continued to deteriorate. From this point onwards, and particularly past the 1.22 level in GBP/USD, price gapping between trades is increasingly visible (Graph 5). And by 00:07:34 BST, 19 seconds later, GBP/USD had reached 1.20 and the move had exhausted the resting sterling bids across a variety of electronic trading platforms’ order books. This unusual phenomenon occurred at several points during this period, with EUR/GBP also affected by an intermittent evaporation of offers to sell euros (and buy sterling).

While bids quickly returned to the market, overall depth in the order book remained extremely low for several minutes. Those wishing to trade could only execute in relatively small sizes and at prices at a wide spread to the implied midprice (Graph 8). For example, on Reuters prices as low as 1.1491 in GBP/USD traded at 00:07:41 BST — which represented a fall of almost 9% from the pre-event level of 1.26. Other platforms reported transactions at even lower prices. In EUR/GBP, the highest traded level on Reuters was 0.9403, at 00:09:49 BST — representing a 6.3% appreciation of the euro relative to sterling from the level around three minutes prior.

During this period, it is likely that the trading activity of individual participants could have had a significant impact on market functioning and prices traded, given the lack of depth. Indeed, UK supervisory data point to a significant increase in certain market participants’ share of trading activity as others withdrew — suggesting a role for idiosyncratic factors in driving the extreme dysfunction observed as sterling traded at levels well below 1.20 against the dollar.

It is worth noting that some of the trades executed during this period were subsequently torn up or had their prices revised, but this was often the consequence of bilateral agreements or specific contractual arrangements as there is no single methodology for determining the low in FX markets. Market contacts suggested they did not feel able to co-ordinate in determining a low point (which is relevant for certain derivatives contracts) as they feared this would breach competition and/or conduct requirements.

The lows in sterling were extremely short-lived. Between 00:08:00 and 00:09:00 BST, GBP/USD traded at levels between 1.20 and 1.22 — still a wide range. Market functioning took longer to recover. Sweep-to-fill costs (a measure of market depth) on Reuters remained volatile until nearly 00:20:00 BST, a period over which the CME triggered a number of further trading halts (Graph 9). Shortly after the first pause in trading on the CME ended, the futures price hit the daily lower limit of 122.17.6 At this point, futures trading was floored at this price for a period of two minutes. Some transactions were completed on the CME over this period, despite the spot price continuing to fall on other platforms. But when prices had not risen by the end of the two-minute window, a two-minute trading halt was imposed at 00:09:29 BST. The market reopened at 00:11:29 BST with a new lower limit, but conditions remained impaired (for example, as seen in measures of the price impact of trades). Finally, amid the continued heightened price volatility, a second velocity logic event was triggered at 00:11:57 BST, again halting trading for 10 seconds on the futures exchange.

Stage 3: Over time the market began to recover, although it is difficult to identify a clear shift to the recovery phase. By around 00:20:00 BST, prices in both the futures and spot market had settled around 2.2% lower against the dollar than their levels immediately prior to the event, although relatively high trading volumes (for the time of day) persisted for a period of three to four hours. Bid-offer spreads on Reuters remained wider than usual, but not at extreme levels, for the rest of the night.

Higher than usual GBP/USD volumes were observed on Reuters during the day on 7 October, and measures of market illiquidity were slightly elevated (see Appendix B for more details). But broader spillovers were generally limited. UK government bonds registered a relatively large move as trading opened on 7 October, but the moves were orderly. And there appeared to be little impact on risky asset prices. GBP/USD closed in London on 7 October 1.7% lower than on the previous day. In the weeks following the event, there was little observable evidence of a lasting impact on market functioning. Option-implied volatility in sterling FX pairs remained elevated for a number of days before retracing, but it is difficult to determine whether or not this reflected factors other than the flash event. Bid-offer spreads appeared to quickly recover to around normal levels, with no evidence of impaired market functioning when trading recommenced on the evening of Sunday 9 October. And measures of the price impact of trading activity showed no signs of any persistent impact on liquidity in the futures market in the weeks that followed (Graph 10).