Small-scale stock trades surging
Day after day in early 2011, a supercomputer in Oakland slogged through tons of stock exchange data looking for trading patterns.
Blacklight, one of the mega-memory supercomputers at the Pittsburgh Supercomputing Center, analyzed about 7.5 terabytes — a terabyte contains 1 trillion bytes — of trading data on 7,000 stocks listed on the Nasdaq exchange.
Mao Ye, the University of Illinois researcher who was borrowing Blacklight, made a startling discovery: An increasing amount of “odd-lot” trades in stocks — 100 or fewer shares — were not being included in exchanges' official tallies of daily trading volumes.
“By executing a whole bunch of small trades, you can effectively do a large trade and escape notice,” said Ralph Roskies, scientific director at the supercomputing center.
When Ye presented his findings to Nasdaq, exchange officials agreed there was a greater need for transparency. So now, Nasdaq as well as the New York Stock Exchange will include odd-lot trading volume data in their daily volumes starting in October.
“That study influenced the decision,” said Frank Hatheway, Nasdaq senior vice president and chief economist.
“The conclusion of the study, that there's substantial trading in odd lots, is something we accepted and discussed,” said Hatheway. “The study made those conversations easier.”
The data was analyzed by the supercomputing center's Blacklight and the San Diego Supercomputer Center's Gordon. The Pittsburgh center is a joint effort of Carnegie Mellon University, the University of Pittsburgh and Westinghouse Electric Co.
Blacklight saved Ye “lots of time” because he didn't have to do a lot of computer programming, said Roskies.
Traditionally, odd-lot trading data has not been included in the stock exchanges' daily trading volumes reported to regulators. The thinking was that odd-lot trades came from small investors whose actions were unlikely to affect the market much.
But with the increase of high-frequency trading in odd lots at light speed, small trades add up. For instance, instead of trading 10,000 shares in one order, a program can make 200 trades of 50 shares, which potentially can manipulate a stock's price.
The study found that odd-lot trading that accounted for 2.25 percent of Nasdaq trading volume in January 2009 had grown to 4 percent by the end of that year because of high-frequency trading.
“Part of the concern is whether these automated trades sometimes spiral out of control and cause problems with the market. We've had crashes because programs run amok,” said Roskies. He cited the “flash crash” on May 6, 2010, when the Dow Jones industrial average lost about 1,000 points in minutes, then made up nearly all of that by day's end.
The reporting change also will provide more trading data about high-cost stocks, such as Google. With its shares selling for more than $900 each, investors tend to buy and sell the stock in quantities of fewer than 100 shares. As a result, more than half of trades in Google go unreported, said the study.
New York Stock Exchange spokesmen declined comment on the supercomputer-aided study but acknowledged odd-lot trading would be included in daily volumes starting in October.
Thomas Olson is a Trib Total Media staff writer. He can be reached at 412-320-7854 or firstname.lastname@example.org.