Flash Crash: Fat Fingers or Faulty Markets?
Turns out we're still not sure.
BY Scot Blythe | November 30, 2010
The May 6 flash crash has excited controversy and examination – but as yet, little in the way of definitive explanation. For one thing, mini-flash crashes – at least in particular stocks — continue to happen.
That’s something that Baruch College finance professor Bob Schwartz doesn’t find at all surprising. He doesn’t dismiss the possibility of a future flash crash. His reasoning: markets have become increasingly fragile. One piece of evidence is how much more bids and offers shoot up – and down — at the opening of the trading day.
Some may think that “fat fingers” are at work – a trader mistypes a decimal point or adds an extra zero to an order. But, brokerages are supposed to have controls in place to prevent “erroneous” or “broken” orders. So too with marketplaces – of which there are now many, instead of the old NYSE-NASDAQ oligopoly, or the TSX monopoly in Canada.
Both the U.S. Securities Exchange Commission, together with the Commodity Futures Trading Commission, and the Investment Industry Regulatory Organization of Canada have conducted inquiries into the May 6 flash crash.
Most recently, the SEC has banned “stub quotes.” Essentially, stub quotes are stink bids, offering to buy a share for a penny or sell it for thousands of dollars, allowing market makers nominally to fulfill their obligation to maintain orderly price discovery by putting up two-sided quotes: a bid and an offer. The SEC has now mandated that market maker prices be within 8% of the national best bid offer for stocks subject to circuit breakers, and within 30% for stocks that are not.
The SEC introduced single stock circuit breakers as a pilot program in June, and IIROC is looking to follow suit.
Still, the flash crash has laid bare important changes in market structure since the early 1990s – changes that affect both price discovery and liquidity – and therein may lie some of the answers to the crash.
“In comes the high frequency traders. You can always blame them if you want,” Schwartz told a Toronto audience recently. “But I think it’s just too simple to say it was them. We have yet to find any significant results that they added to volatility.”
In this, he echoes the comments in a controversial paper by Ewing Marion Kauffman Foundation researchers Harold Bradley and Robert E. Litan. They note, “In the good old days of the 1960s and 1970s, it could take a minute or more for the typical retail trade to be completed. By comparison, the [SEC] Concept Release notes that, between 2005 and 2009, average execution time for small orders of NYSE-listed stocks fell by more than a factor of 10, from 10.1 seconds to just 0.7 seconds.”
Is that a problem? It’s hard to tell, Bradley and Litan argue, since in the SEC paper “nowhere is high frequency trading defined, although we know that the practice covers hundreds of institutional and broker dealer trading approaches. In our view, trading algorithms and high frequency trading serve a tempting, convenient, and easy target for politicians and market participants who have failed to adapt to the modern technology age.”
The situation is a bit different in Canada. For 90% of Canadian stocks, trades occur less frequently than on the NYSE – about once every five minutes, says James Twiss, IIROC’s vice-president of market regulation. And, while high frequency traders make up 70% of the trading in U.S. stocks, they affect just 15% to 20% of the volume in Canada. Still, high-frequency trading can have an impact on some Canadian stocks, since 70% of the volume is in issues that are interlisted in Canada and the U.S.
That may be one reason that the flash crash was more attenuated in Canada. The index fell only half as much – 3% — and lagged the U.S. events by two minutes – a “sympathetic response” Twiss calls it. But there are other differences in Canadian market structure that are worth paying attention to.
According to Schwartz, only 40% of liquidity is visibly priced. Other orders are placed in “dark pools” that cater to institutional investors moving large blocks of stock, or, on the retail side, are internalized by broker-dealers who match orders from inventory, “free-riding” some would say, on the basis of public quotes, while capturing the spread between the bid and the ask.
But sometimes that free-riding comes at too high a cost. Say Bradley and Litan, “many internalizers of retail order flow stopped executing as principal for their customers that afternoon, and instead sent orders to the exchanges, putting further pressure on the liquidity that remained on those venues.”
Better than the famous Nasdaq hang-up of 1987, when market makers didn’t bother to answer the phones, letting the New York Stock Exchange deal with the carnage. But, Bradley and Litan suggest, the Nasdaq market maker model was never all that stable to begin with.
“During normal market conditions, internalizers can afford to pay for market orders only because they are allowed to trade these orders at the same effective price as orders transparently priced on competing exchanges. This cannibalization of market orders relies wholly on displayed limit orders that constitute the National Best Bid and Offer (NBBO) and creates yet another reason for institutions to seek solace in dark pools.”
The Canadian market structure is different. Internalization, in the Canadian version, is essentially trading in dark pools. For retail investors, there is an order exposure rule. “What the order exposure rule was designed to do was to ensure that there was enough liquidity in the books,” says Twiss. With large orders (50,000 shares), “there’s an obligation that that order be immediately ordered on the marketplace or executed at a better price than is displayed on any marketplace. In U.S., there is no order exposure rule. There can be full internalization and many of you have read the concept paper from the SEC that was released earlier, they lamented the number of firms for which the entire retail order flow never sees a transparent marketplace in the U.S.”
Canada’s toolkit is a bit better when it comes to buffering volatility, Twiss argues, but it’s far from perfect.