Supercomputers could generate warnings
for stock crashes
POSTED:
04/22/2013 12:01:00 AM MDT
UPDATED: 04/22/2013 02:01:11 AM MDT
UPDATED: 04/22/2013 02:01:11 AM MDT
By Lisa M.
Krieger
San Jose Mercury News
San Jose Mercury News
Powerful computers can wreak havoc on
U.S. stock markets, creating hair-raising volatility and eroding investor
confidence in the lightning-fast search for profit.
But far more powerful computers could
help save it.
High-speed trading now dominates U.S.
stock markets, buying and selling in a fraction of the time that it takes to
blink. Computers detecting rapid swings in prices and instantly reacting builds
volatility and more trades, generating a sea of chaotic data and a vicious
feedback loop that can lead to nightmares far worse than May 2010's infamous
"flash crash."
Faster still is "Edison," a
supercomputer tended by Lawrence Berkeley Laboratory scientists in a former
Wells Fargo Bank building in downtown Oakland, Calif. Edison-like computers
could track ultrafast trading across the nation's many markets, detecting precursors
to a crash — and sounding early warnings for regulators seeking to avert a
gruesome economic wreck.
It would be like a NASCAR race yellow
flag warning drivers to slow down, scientists say.
"If improved monitoring and
regulation can build some greater trust in the market, everyone benefits,"
said David H. Bailey, director of the lab's new Center for Innovative Financial
Technology, which is building a bridge that links computational science and
financial market communities.
Edison loves big data. Its idea of an
average day is simulating a supernova explosion, measuring the expanding
universe's rate of acceleration. Or modeling 150 years of Earth's future
climate change — in three dimensions.
When fully deployed later this year,
Edison will perform as many as 2 quadrillion operations a second. How big is
that number? Two quadrillion cups of water would fill Lake Tahoe twice.
Tracking every trade, in real time,
on every U.S. stock exchange? No big deal.
"That data size — we routinely
do 10 times that much. Easily. It's a trivial matter," said Bailey, a
leading figure in both high-performance scientific computing and computational
mathematics.
Not so long ago, life was simpler.
U.S. stock markets moved at a human pace, simply matching buyers with sellers.
But now many exchanges take place. And more than half of all trading is done by
high-speed computer "traders" that live their electronic lives in
server parks.
Most agree that computer trading is good
for the average investor because it's inexpensive. But it also triggers
unpredictably large price swings — causing widespread Maalox moments. It's
breeding distrust in the market.
The "flash crash" of 2010
was triggered by a single firm using algorithms to rapidly sell 75,000 futures
contracts.
In moments, the Dow Jones industrial
average fell more than 700 points, or almost 10 percent, and quickly recovered.
Since then, numerous mini-flash
crashes and other anomalies have slapped around stock values and investor
confidence.
"Electronic markets ... seem
capable of impressively flaky behavior," said Lawrence Berkeley Lab's
David Leinweber, a computer scientist, former investment manager and
algorithmic trading specialist.
"We are lost in the jungle when
it comes to our ability to understand ever-faster markets well enough to keep
them safe, stable and secure," he said.
Regulators with weak and incompatible
computer systems have set safeguards. One uses shutdown switches —
"circuit breakers" — to halt all trading. A second, called
"limit up, limit down," cancels trades outside a normal price range.
But these, asserted Leinweber, are
like applying the rules of the road to aircraft. Slowing, rather than suddenly
halting, markets is less traumatic.
In early 2011, using the lab's Cray
XE6 "Hopper" supercomputer, Leinweber's team found that a
supercomputer could use a recently identified measure to warn of a looming
flash crash.
Called Volume-synchronized
Probability of INformed trading, or VPIN, it detects an imbalance between buy
and sell orders, and growing volatility, about 45 minutes before a crash. It
reveals "flow toxicity," that is, when high-tech traders' computers
generate so many buy or sell orders that it becomes all but impossible to match
orders. Amid this volatility, one side will stop trading to stem its losses,
causing a sudden drop of prices that triggers an avalanche of similar
withdrawals called a "flash crash."
A second measure of market
instability, the Herfindahl-Hirschman Index, or HHI, also rose sharply for some
stocks, although not for others.
These two signals of instability —
and that uneasy feeling that someone else has more information than you — do
not give a clear direction of specific trades, said the lab's John Wu. So
they're not likely to be exploited by someone hoping to jump in and profit.
But they might be useful "for
regulators to impose some rules that might slow down the market so we don't get
into a sort of feeding frenzy," he wrote in an e-mail.
Comments
Post a Comment