Slashdot Mirror
SEC Blames Computer Algorithm For 'Flash Crash'
Lucas123 writes "The US Securities and Exchange Commission and the Commodity Futures Trading Commission today issued an 87-page report (PDF) on the results of a months-long investigation into the May 6 'flash crash' that sent the Dow tumbling almost 1,000 points in a half hour. The Commissions are holding a single trading firm's automated trade execution platform responsible for the crash, saying it dumped 75,000 sell orders into the Chicago Mercantile Exchange over a period of minutes causing an already volatile market to come crashing down. The SEC has already enacted some quick rules to pause trading if a stock price should rise or fall by 10% in a five minute period, but the regulators said they expect the results of the investigation to prompt additional rules limiting the functions of automated computer trading systems."
Not quite. A stock's quote price is the last price at which a bidder's offer matched a seller's asking price. A "level 2" quote has two parts, the highest bid price that hasn't been matched up yet, and the lowest asking price that hasn't matched up yet. The true value is somewhere in between these two, but nobody knows where until somebody steps in between the high bid or low offer or somebody moves their price to get a deal. Any computer programmer should check that there's matching orders on the other side of their trade before placing a large order. A program that doesn't will execute just fine, but crash the economic system.
I just finished reading through the whole report. It's fascinating, if you're into this.
First, none of this involved a "bug" . All systems involved functioned as designed.
What's going on here is a logical consequence of the way the markets are set up. The Chicago Mercantile Exchange ("CME", the futures market, which started by trading grain) has a tradeable commodity called the "E-mini", which is a derivative security based on the S&P 500 stocks. Anyone can buy or sell contracts in E-minis, and can also buy or sell the underlying stocks. This generates a frantic amount of short-term trading from market players trying to profit from the differences between the two, which keeps the price of the E-mini close to the prices of the S&P 500 stocks.
None of this is productive activity, of course.
There's a consolidated feed from all markets that everybody gets. It has a few seconds of lag. To obtain an advantage in fast trading, some of the players buy direct exchange feeds with an average of 8ms (yes, 8 milliseconds) of lag.
What started the crash was that a fundamentals trader (one who actually pays attention to the companies involved) was selling $4 billion in stocks. Ordinarily, this isn't a big deal. They had a program throttling their rate of sale to 9% of market volume in the last minute, to avoid depressing the market. That's normal. So far, so good.
However, in response to this sale, the "high-frequency traders" started frantically trading back and forth to balance their portfolios. Their net effect didn't move prices much, but it pushed volume up. So the big seller started selling faster.
This generated enough volatility that some market players started dropping out, decreasing liquidity. That generated market imbalances which other traders started to exploit. Then, because of all this frantic trading, the consolidated market feed and the millisecond feed differed enough that some trading firms had data quality alarms and dropped out of trading. Since traders who are "market makers" are required to maintain buy and sell bids in the market, they defaulted to their default bids - buy at $0.01, sell at $100,000. Some trades actually took place at those prices. 895 shares of Apple stock were sold at $100,000. The price of Accenture fell from $30 to $0.01 in seven seconds, then recovered within the next minute.
Then "At 2:45:48, trading on E-Mini was paused for five seconds when the CME Stop Logic Functionality was triggered in order to prevent a cascade of further price declines". Yes, a 5-second automatic trading halt. That was enough to start to stabilize the E-mini contract trading on the CME. But by then, the E-mini was enough out of sync with the underlying stocks (mostly on the NYSE) that trading on the NYSE started to move stocks there to resync with the E-mini.
The NYSE still has a trading floor, which slows it down. This didn't help. But that's another story.
Nothing failed. Nobody did anything wrong. The original seller's strategy for unloading $4 billion in stock was reasonable. This is all a consequence of normal market operation. The report concludes that speeding up the consolidated market feed to get the 5-second lag (which was more than fast enough before program trading) down should be done. That's it.
Whether or not society should support an "efficient market" system to this extent is an question one is not supposed to ask.
Here are few important facts:
1. Waddell & Reed is the company whose aggressive selling triggered drop in S&P 500 futures price. The company is not HFT shop but rather long-term investment hedge fund. More here: http://www.bloomberg.com/news/2010-09-30/waddell-reed-e-mini-trades-are-said-to-have-helped-trigger-may-6-crash.html
2. According to SEC report, HFT traders played their intended role: smooth out short-term price volatility. However, due to enormous size of Waddell & Reed selloff (about $4 billion dollars in 75000 futures contracts during 20min.) they can do only that much. W&R just cut right through the order book on CME.
3. Slowing down the trading on NYSE did not help but rather hurt by locking up liquidity. Shitty NYSE Arca systems that handle ETFs overloaded and further exacerbated the problems.
4. At the end of day market returned to pre-crash levels. Long term investors were not hurt, W&R payed between 100 and 200 millions for their mistake.
5. Overall, market worked as expected.