News

That Citi flash crash revisited

On Monday morning, with volumes thinned by the UK’s bank holiday, Citigroup briefly crashed the European market. Nearly two trading days later, we’re still struggling to figure out how.

Citi has placed the blame on one unfortunate in its London office, saying a unidentified trader “made an error when inputting a transaction”. As excuses go, its sounds plausible. The London Stock Exchange’s absence meant volumes across European bourses were less than half the normal level. It wouldn’t take much to set off an algorithmic chain effect, particularly among the backwater Nordic markets.

Our colleagues Joshua Franklin and Philip Stafford report:

One trader said the basket was for trades on the Euro Stoxx 50 — an index that tracks blue-chip companies in the eurozone. The issue was further compounded by an error in a conversion to Swedish krona, which meant Stockholm-listed shares bore the brunt of the mistake.

The gossip on Monday involved a basket valued at about $350m. Immediately, that sounds wrong. Though Citi’s compliance systems have not always been faultless, any big bank program trading desk should have risk controls in place that would catch such an order placed by a lone trader. Based on the slightly more than 1bn of Stoxx 600 shares traded on Monday it’s also a bit underpowered.

A fat finger explanation needs to be weighed up against the breath and unevenness of Monday’s brief selloff. A basket of mostly Nasdaq Nordic stocks liquidated at the wrong price would show up in specific places, but the damage here looks much more general. Via Bloomberg, here are tick charts for the main European indices:

And sure, it’s ugly for the OMX 30 Stockholm (orange line) — but it’s even worse for the OMX 25 Copenhagen (purple). That’s a challenge for the “error in a conversion to Swedish krona” thing, given Copenhagen constituents are priced in Danish krone.

Limiting the view to the underperforming OMX Oslo, Copenhagen and Stockholm indices looks this way:

Which raises one possibility. Might Citi have accidentally set a whole pan-European programme trade to go through in euros, thereby trashing everything priced in a Nordic currency?

One euro’s worth about 10.41 Swedish krona, 7.44 Danish krone and 9.93 Norwegian krone at pixel. A mistake along those lines might mean wrong-currency bid prices chew quickly through order-book liquidity and result in charts that look a bit like the above. The same error could be repeated in the Swiss Market Index (magenta line) but would be much less obvious, given the Swiss franc trades near parity with the euro.

Is a currency denomination error enough to dunk the wider European market? Maybe, in thin liquidity, but only with big assumptions about the trigger-happy nature of algo trading. Nordic nations don’t have much influence — there’s just one representative in the Stoxx 50 index, Finland’s Kone, and the Stoxx 600 weightings for Sweden and Denmark are 5.5 per cent and 3.9 per cent respectively — so a lot depends on believing in the bot-driven ripple effect.

Thing is, the breakdown of OMX 30’s slump reveals a synchronised but dissimilar performance among its member stocks. That’s unexpected.

In the first instance, basket sells hit whatever’s in the basket. Correlation chasers and stop-loss liquidations then cause everything else to fall more-or-less in lockstep. Neither of these things happened here. OMX members AstraZeneca (teal line) and ABB (red) are barely affected throughout the tumble while Assa Abloy, Hexagon and Swedish Match all sink by double figures.

Might this be another symptom of a currency mistake? Like a majority of OMX stocks Assa Abloy, Hexagon and Swedish Match trade almost exclusively in krona. The Stockholm quote is a reference price. But for AstraZeneca and ABB, Stockholm is irrelevant: their krona-denominated daily trading only accounts for approximately 10 per cent of the total. ABB’s dominant quote is in Swiss francs and for AstraZeneca, even when London’s on holiday, the Cboe sterling quotation provides the reference.

The minutes going into the OMX 25 Copenhagen’s selloff show a similar pattern among its dom and non-dom stocks. Novo Nordisk, which has a reasonably active non-kroner quote, outperforms. Pandora, which doesn’t, doesn’t.

Nasdaq and Euronext have said they are standing by the trades, so Citi needs to wear any losses from its mistake. What that means in practical terms is not known. It might involve buying back a single basket sold on the dip, or it could require repairing the damage from a misaligned program that was unilaterally hoiking out everything Nordic down to approximately a tenth of its actual value.

We don’t know which explanation (if either) is the right one, because we’re not experts in this stuff and Citi’s not telling us. Theories welcome below.