The USDJPY flash crash

I’ve ranted previously about people calling things “flash crashes” that don’t deserve the name. Fat fingers; gunning stops; these things aren’t a patch on the original flash crash. But what happened in USDJPY two days ago, in the witching hour between the Wednesday session in New York and the Thursday session in Singapore – that was worthy of being called a flash crash.

The least liquid time of day in the FX markets is after 5pm in New York: the NYC traders are in the pub; the Sydney traders are still waiting for their first cup of coffee to kick in; and the Tokyo and Singapore traders (except for a few dedicated souls) aren’t at their desks yet. And on Wednesday afternoon, around that time of day, someone started selling USDJPY. And AUDJPY. And EURJPY.

And then more people started selling. Stop-loss sell orders started firing. Exotic options barriers – set at levels below the all-time USDJPY low – started firing, and those tend to have stoploss orders attached to them as well. But while all these forced sellers were lining up to hit bids, all the buyers went on strike… and this was the result:

USDJPY tick chartThis chart (with timestamps in Singapore time) only shows ninety seconds of trading – and in that ninety seconds, USDJPY dumped 160 points (or a bit over 2%). Normally, USDJPY doesn’t move two percent in a week.

Here’s the tick-by-tick data from the most violent part of the move – on the chart, this corresponds to the cliff-dive at 5:19:50 (the tick data’s timestamped UTC, yes it’s confusing):

image002.png

Each of these lines is one USDJPY trade on Icap’s EBS system – the biggest and most liquid interbank trading platform for JPY (and EUR as well; for historical raisins that are beyond the scope of this post, Reuters’ D3000 platform is the biggest venue for GBP and the commonwealth currencies. Any other platform that says they’re bigger than D2 or EBS is, I suspect, fibbing).

Look at what happens at 9:19:47. A bid gets hit at 77.42, exposing the 77.40 bid behind. The 40 bid gets hit, exposing a 77.25 bid behind. If you hit that bid at 77.25, you’re crossing a 15-tick spread – when the normal spread in interbank USDJPY is one or two ticks. Normally you wouldn’t do it – not least because hitting that bid is going to cost you $1500 USD compared to the last trade level. Normally you’d be insane to sell it. But someone had to sell it and had no other choice. At 9:19:50, someone hits the 77.25 bid… and finds open air behind it, just like Wile E. Coyote when gravity kicks in.

At this point, nobody in their right mind would show a bid. No human trader is going to post a bid in these markets, because they’re going to get given and it’s going to keep on going lower. No machine is going to post a bid, because their humans will have pulled the plug on them in a market like this. So in the space of three hits – 77.349 given, 77.25 given, 77.00 given – USDJPY’s tanked half a percent.

And that creates more problems. Exotic option barriers tend to lie at round numbers – 77.50 and 77.00, for example – and when those levels get hit, people generally need to sell into a falling market (or buy in a spiking market). And anyone with a stoploss order between 77.50 and 77.00 will be selling into a falling market as well. So the decline feeds on itself: you see more traders hitting bids, nobody paying offers, and that scares even more people off.

78 seconds later, the market hits its ding-dong low at 76.25:

image001.png

And the situation is even more miserable. The average spread has blown out to 25 or 30 ticks – again, the normal spread in USDJPY is a tick, two on a bad day. If you cross that spread to hit a bid, it’s $2500 or $3000 per million USD notional – and by now, if you’re a big bank, you might have 20, 30 or 50 million to sell. Do the math on that and tell me it doesn’t make your stomach drop.

At one point (9:20:58 to 9:21:01), people are hitting 76.30 bids and paying 76.50 offers basically at the same time.

But why did it all go so horribly wrong?

Consensus seems to be that it was mostly Mrs. Watanabe’s fault. When Tokyo Forex (one of the larger retail FX brokerages in Japan) released its positioning data that day, it turned out that Japanese retail traders had got themselves long a record amount of USDJPY ($3 billion or more) just the previous afternoon, with billions of USD of stops lined up below the level where the BoJ was thought to be intervening.

If you want to point the finger at a proximate cause, EU energy commisioner Gunther Oettinger’s ill-considered statement (“the situation in Japan is out of control!”) is as good as any. That startled the market and pushed it toward the level where Mrs. Watanabe’s stops were lurking.

But when those stops started firing in the least-liquid time of the day, it started off a selling cascade. In the original flash crash, Waddell and Reed’s massive SPX sell order triggered the selling; this time around, it wasn’t so much a single player hitting the bid as a mass of forced sellers all rushing for the exit at once. In that sense, Thursday’s USDJPY collapse was more like a re-run of Black Monday and its portfolio-insurance-driven stoploss selling.

Either way, though, the effect was the same. By the time Singapore walked in and sat down and choked on its coffee, USDJPY had dumped four big figures, rallied two big figures, and USDJPY implied vols (which reflect past volatility as much as they reflect expected future volatility) had – for the shortest-dated options – more than doubled.

And the news media had yet another bad-Japan-news story on their hands. “What does it all mean? Is it the dreaded repatriation? Is it a nuclear meltdown we didn’t hear about? What made this happen?” Sometimes, though… sometimes a flash crash really is just a flash crash.

This entry was posted in Money. Bookmark the permalink.

4 Responses to The USDJPY flash crash

  1. yen trader says:

    you’re such a wannabee trader…

  2. Alex Sebastian says:

    I noticed you used the standard Waddell & Reed story for the Flash Crash. I would be interested to know whether you have read NANEX’s reports refuting this story and what you make of them?

    http://www.nanex.net/FlashCrashFinal/FlashCrashAnalysis_W&R.html

    • Josh says:

      I’m not really sure Nanex’s reports refute anything. They’re interesting, sure, but they show the data and then go immediately on to say “the W&R volume could not possibly have caused the crash” – which sounds to me like argument by assertion.

      The SEC report seems pretty thorough to me.

Leave a Reply

Your email address will not be published. Required fields are marked *