Share |

Keeping an eye on superfast transactions on Wall Street

Using a supercomputer to catch a supercomputer
Image of Time Series Variation of Missing Trades graph.

Graph of time series variation of missing trades from research paper: What’s Not There: The Odd-Lot Bias in TAQ Data. This figure illustrates total level of missing trades and volume from 2008 to 2009. Panel A demonstrates the volume not reported to TAQ data as a percentage of total volume. Panel B demonstrates the trades not reported to TAQ data as a percentage of total trades. Click on image to enlarge. Image courtesy Mao Ye.

The superfast-paced financial trading world of Wall Street has been put under scrutiny with the help of supercomputers. The researchers found a general lack of transparency in the marketplace, as well as evidence of high-frequency trading algorithms — capable of carrying out transactions in just billionths of a second — being used to congest the market by placing orders only to cancel them within seconds. These findings are helping regulators make decisions to rebalance the trading arms race.

First, researchers from the University of Illinois at Urbana-Champaign and Cornell University, US, looked at the systematic bias of trading decisions. Specifically, they examined how a lack of transparency in trades of fewer than 100 shares skew perceptions of the market in their paper called What’s Not There: The Odd-Lot Bias in TAQ Data. The researchers looked at high-frequency trades, which now take up 73% of all trades in equity markets according to the paper.

Today, human traders have to be both savvy traders and computer experts, but they’re mostly removed from the equation. Supercomputers and algorithms have largely replaced human-based trading in the financial markets, with the raw data generated easily reaching ten gigabytes per day.

With this torrent of data, only sophisticated algorithms and supercomputers can hope to identify errors in high-frequency trading. The researchers used the shared-memory systems’ of Blacklight, at the Pittsburgh Supercomputer Center to perform an analysis of excluded trades for less than 100 shares of stock, also known as odd-lot trades, and Blacklight and Gordon, at the San Diego Supercomputer Center to look at order cancellations.

Because of this research the Financial Industry Regulatory Authority may revise their odd-lot trades policy and include them in consolidated tape, which holds official records of latest price and trade volumes. This November, the US regulators will vote on a proposal to eliminate odd-lot trades altogether. "For the 120 stocks in our sample, about 20 percent of trades are missing, Eliminating odd-lot trades would result in much greater disclosure,” says Mao Ye, principal investigator of this study.

The chief economist of the NASDAQ, Frank Hatheway, says Mao Ye’s research paper has received interest from top financial economists and regulators alike. 

“I was privileged to be in the audience when NYU Professor Rob Engle, Nobel Laureate 2003, cited Mao’s paper as the only existing research which had been able to identify and explore this unseen aspect of the markets on both a theoretical and empirical basis,” says Hatheway. “The paper has also triggered a discussion about whether the US equity markets should change their handling of odd-lots in order to provide more information and transparency to the investing public.  A decision on that point is expected in the next few months.” Those who look at impact of recent regulatory and technology changes will also benefit according to Hatheway.

Image of Histogram of Quote Life for Orders with a Life Less than One Second graph.

Histogram of quote life for orders with a life less than one second from research paper: The Externalities of High-Frequency Trading. This graph shows the histogram for all the orders with a lifespan of less than one second. Each bin represents a 5-millisecond interval. The sample includes 118 stocks from 19 March 2010 to 7 June 2010. Click on image to enlarge. Image courtesy Mao Ye.

In a second, more recent study, entitled The Externalities of High-Frequency Trading, Ye and his students, Jiading Gai and Chen Yao from Illinois University, analyzed 55 days of NASDAQ trading from 2010. In total, this worked out to roughly 15 terabytes of raw data, which is about the size of the entire US Library of Congress.

"If we had tried to use anything other than supercomputers, it would have taken years just to process the raw data," says Ye. 

While Blacklight’s memory capacity enabled the researchers to handle this huge amount of data, Gordon enabled the researchers to run instructions in parallel on multiple processors to understand why trades are canceled so quickly. XSEDE (Extreme Science and Engineering Discovery Environment) staff helped optimize the effectiveness of the C++ code run on these two systems. 

The code was used to analyze the ratio of orders cancelled to orders executed, and evidence was found of a manipulation called ‘quote stuffing’. This is when a high-frequency trader places an order and then cancels it within 0.001 seconds. As traders invest aggressively at speed, an abnormal amount of these orders, followed by cancellation congests the market, which can give the trader a few microseconds or milliseconds advantage. Today, over 95 percent of daily orders are cancelled almost immediately after they’re placed.

The researchers conclude that the Securities and Exchange Commission should change their policies and rules in regards to these types of trades. Measures could include limiting the speed of orders or charging a fee for order cancellation. Ye says, "A small fee is necessary, but only at a certain threshold to allow genuine cancellations to occur without penalty."

Now, a separate study is looking at the issue of hidden orders or hidden liquidity, by which a trader places an order that bets on where a stock price will go without exposing their position, therefore not impacting the market price. Another study is looking at options trading, which requires even more data than the odd-lot analysis.

Your rating: None Average: 4.6 (14 votes)

Comments

Post new comment

By submitting this form, you accept the Mollom privacy policy.