[OFF TOPIC - Program Trading (more arguments lending towards TA)]
December 16, 1997
Mathematical Market Model
Lets Firm Squeeze Out Subjectivity
By GREG IP
Staff Reporter of THE WALL STREET JOURNAL
RADNOR, Pa. -- If you want to win in the stock market, you can take
the usual route: Pore over companies' financial statements, study their
management, consult some leading securities analysts, then call up and
place an order with your broker, and wait.
Or you could forget about annual reports and
company management and instead get a Ph.D. in
math, design a model of the stock market, load it
into a hundred high-powered computers around
the world, and then let it tell you how to buy and
sell hundreds of millions of shares a week.
That's how Andrew Sterge does it.
His formulas have helped turn BNP/Cooper Neff
Advisors, a firm based in this town near
Philadelphia but owned in Paris, into one of the
biggest stock traders in the world. Every week, it
trades 100 million to 150 million shares on the New York Stock
Exchange, from 4% to as much as 6% of the Big Board's total volume. It
also buys and sells millions of Nasdaq shares, while sister operations trade
massive volumes in Germany, France, Italy, Spain, Australia and Japan.
Yet the firm is almost unknown outside Wall Street, and even to many
people on the Street.
Its faceless obscurity seems appropriate to an approach based on
quantification and more quantification, with an absolute minimum of human
judgment about companies. Traders at the firm don't know the
price-earnings ratios or dividend yields of stocks they buy and sell. They
don't know the companies' debt levels or book values or product pipelines
or revenue growth. In fact, they don't even know their names.
"You don't want to know them," says Mr. Sterge. "That's extraneous." At
BNP/Cooper Neff, stocks aren't companies in industry sectors. They are
"mathematical objects."
Ordinary research tools matter little. "I don't really know how to read a
balance sheet," this buyer and seller of millions of shares a day says. "I've
never looked at an annual report."
Do the Math
What he knows is mathematics. Armed with a doctorate from Cornell
University and a model he has refined for six years, Mr. Sterge and
BNP/Cooper Neff carry a standard piece of investment advice -- stick to
your discipline and don't let emotions sway you -- to its logical extreme. In
the lingo of Wall Street, they make their decisions with a "black box."
The 38-year-old Mr. Sterge, who runs BNP/Cooper Neff's proprietary
trading, thinks studying news and financial reports to determine a stock's
intrinsic value is a waste of time. "There's no god who knows the fair value
of any traded asset," he says with almost religious fervor. "The only way
you can know anything about value is how the market tries to find an
equilibrium. That's all the market is -- a big feedback mechanism trying to
find equilibrium."
And the only way to beat the market, he adds, is to trade thousands of
stocks, by the millions of shares, in search of tiny inefficiencies. The
computer won't be right all of the time, not by any means, but if it is right a
little more than half the time, those huge volumes will turn that narrow edge
into big profits.
Minimizing Risk
Though BNP/Cooper Neff has a major options-trading operation, Mr.
Sterge's trading arm, BNP/Cooper Neff Advisors, has only one client,
Banque Nationale de Paris, which bought the firm three years ago. The
bank has its own brokerage firm, BNP Securities, so commissions on the
massive trading aren't a problem. The results of this proprietary trading
needn't be made public, and aren't. Cooper Neff's only published results
are those of two new hedge funds that it has started and hopes to offer to
investors early next year.
One of them, open since May, beat the S&P 500 by 14 percentage points
in its first six months. The other, which merely attempts to beat the return
on Treasury bills, is up 21% in its first 13 months. T-bills returned about
6% in that time. In that fund, for every dollar of stock owned, a dollar's
worth is sold short, or borrowed and sold in the expectation of buying it
back later more cheaply.
Mr. Sterge concedes that those track records are too short to mean much.
But he points to what the firm considers an even more important
achievement: It gets its results with low risk. The hedge funds' volatility is a
small fraction of that of major market indexes. Volatility for a given return
in the proprietary trading, Mr. Sterge says, is a mere one-fifth that of the
S&P 500. More reward with less risk is the holy grail of investing.
So what goes into the model? Well, those many things the traders don't
know are mostly in there, though just what weight they have and how they
figure in the model Mr. Sterge isn't saying. Although he and his traders
don't bother to know the details of individual stocks, the computer model
is constantly sweeping through a variety of databases and updating such
things as earnings, book value, price momentum, volatility and liquidity.
But some things it ignores because Mr. Sterge has found them of little use,
such as "earnings surprises" -- companies beating or missing analysts'
estimates. Too many others trade on that, he says. In fact, he says,
investors often think they have found a pattern, particularly a historical one,
that they can exploit, without realizing that when everyone with a computer
can spot the pattern, the profit opportunities vanish.
Based on all its data about the market's internal workings and stocks'
historic and theoretical relationships, the model rates thousands of stocks
against one another as overvalued or undervalued.
John Culbertson, an executive vice president at the firm, hopes eventually
to take human beings out of the hedge-fund management altogether,
except as monitors. The model's consistency can't be confirmed until the
last "fuzzy slice of subjectivity" is removed, he says. If "a manager is going
through a divorce, for example, you want to make sure that's not seen in
the results of the fund."
For now, there are people. But headquarters here in this Pennsylvania
town are nothing like the frenetic trading desks of some big Wall Street
firms. There are no traders screaming into phones, waving research
reports or scanning the newswires for market-moving items.
Easy Does It
Instead, half a dozen recent college graduates sit quietly staring into
computer screens. Periodically, a window pops up on the screen. A red
one offers a basket of stocks to sell, a green one a basket to buy. The
computer spits out the expected return over a typical 10-day holding
period and the risk, in dollars, of the basket. The trader clicks a mouse to
accept or reject the trade.
Taking a break from the seemingly leisurely pace, head trader Robert
Cavallaro says, "The misconception of the trading environment is it's
something terrible that burns you out and kills you." With crewcut hair,
jeans, a T-shirt and workboot-clad feet up on the desk, he looks more
like a punk rocker than a stock trader. "The way we run trading is fresh
and energizing, not taxing," he says.
His is by no means the only firm that uses a quantitative approach. In the
first place, everybody is quantitative to some degree, even such a
legendary stock picker as former Vanguard Windsor Fund manager John
Neff (no relation). "I'm not against the quantitative approach because
mine's quantitative," he says, noting that a computer can help with one of
his goals: keeping emotion out of investing.
But even the more highly quantitative trading firms have long had plenty of
competition -- and it keeps getting stiffer. As vast computer power has
come within the reach of more people, the most profitable inefficiencies in
the market have been all but arbitraged away. For example, the
simultaneous purchase of stock-index futures and sale of stocks, or vice
versa -- the type of program trading blamed in part for the 1987 crash --
is much less profitable than 10 years ago. To make money, "quants" need
ever-increasing volumes and ever-more-sophisticated models.
In the same circle as BNP/Cooper Neff is a cluster of other low-profile
but highly active firms: D.E. Shaw Investments L.P. in New York,
Susquehanna Investment Group in Bala Cynwyd, Pa., and Hull Trading
Co. in Chicago. Among big Wall Street firms, there is the
proprietary-trading desk at Morgan Stanley, Dean Witter, Discover &
Co. Many investment funds also pick stocks according to quantitative
models.
But nobody else, it appears, does so in as much volume as BNP/Cooper
Neff. To other traders, the firm and what it does are one big mystery.
"We've seen the volumes these people put up in the program-trading
statistics every week, and for the life of us we can't figure out what kind of
strategy they're doing," says Eric Noll, an executive at Susquehanna.
Part of it is the firm's location. Mr. Sterge and his staff are strangers on
Wall Street and like it that way. It keeps them from falling into boring
strategies like "the cash-index arbitrage thing," Mr. Sterge says, with a
trace of derision.
Quantifying Everything
Besides, hanging out with Wall Street types won't improve your math,
which has always been the way Mr. Sterge looks at the world. His
doctoral thesis built a mathematical model of how Congress operates.
During several years as a pro tennis player, he found himself bedeviled by
statistical frustration: "I beat some highly ranked people and lost to people
who weren't highly ranked," he says.
He was trading interest-rate options for CoreStates Financial Corp. in
Philadelphia when he met Roy Neff in 1989. The two had been riding the
same commuter train. Mr. Neff, a founding partner of Cooper Neff, which
was then primarily an options firm, had noticed Mr. Sterge's reading
matter. One day, seeing him buried in a book about "stochastic
processes," Mr. Neff introduced himself.
Mr. Sterge joined Cooper Neff shortly afterward and two years later
began experimenting with program trading. He began with a conviction
that a stock's value could be seen only in how it trades with thousands of
others. He bet that by designing a model that tried to find variances
between where stocks traded and where they theoretically should trade,
he could make money.
He was wrong, at first. He lost money for months. But not faith. He kept
rewriting the model and reloading it into the firm's computers, until it began
to work.
What Multiple?
A simplified example suggests what the model looks for. Gold stocks
historically have traded at much higher prices relative to their earnings than
the average stock, even though their earnings grow much more slowly.
That doesn't seem to make sense. But gold stocks have the unusual
property of usually rising when the rest of the market is falling, and vice
versa. Thus they reduce the volatility of a portfolio. Investors are willing to
pay for that.
By contrast, Merrill Lynch & Co., despite tremendous earnings growth,
trades at a lower P/E ratio than the average stock. This is apparently
because it adds volatility to a portfolio; when the market goes up, Merrill
usually goes up even more, and vice versa. The valuations of gold stocks
and Merrill Lynch make sense only as part of a larger universe.
Mr. Sterge says any reasonably smart analyst can identify some
undervalued and overvalued stocks. But it's another matter to bet on such
knowledge with as little risk as possible. For example, why not just buy
the most undervalued ones and sell -- or sell short -- the most overvalued?
The problem is that a highly overvalued stock might also be extremely
volatile, so you could watch your returns swing wildly from profit to loss
before you finished.
Instead, Cooper Neff goes beyond deciding which stocks are over- or
undervalued and tries to make bets between stocks with similar
mathematical characteristics, such as volatility and liquidity. It largely
ignores such things as their industry and market capitalization.
Ignoring News
What if terrible news breaks one day on a stock the model has bought?
Too bad. Cooper Neff owned Philip Morris Cos. the day in April 1993
that it plunged 23% on news of a big cut in cigarette prices. "We got
smoked, but ended up having a really good money-making day because
we just traded," Mr. Sterge says. "We ignore those things. If we happen to
be long a stock that gets toasted, we don't cut our losses. We let the
model do the work. If it wants to get out, we get out. If it wants to stay in,
we stay in. I can't outguess the damn model.
"Humans create the model, but I let the model make the bets," Mr. Sterge
says.
He does use his judgment, of course, to try to make the model better. He
reads academic journals and scholarly working papers for investment
ideas. He often spends Fridays at his house on North Carolina's Outer
Banks scrawling equations in the back of his day planner. The final result is
a closely guarded secret. Mr. Sterge and his research director are the only
two people in the company who can actually get at the model, which is
encrypted in each trader's computer. The traders must sign confidentiality
agreements even though few would understand the model if they saw it.
And it isn't just the model. The system also depends on massive amounts
of computing power and extremely low trading expenses and cost of
capital. Without those, even an investor as math-savvy as Mr. Sterge
couldn't replicate the results.
New recruits don't have to know math when they join BNP/Cooper Neff,
but they'd better be ready to learn. Mr. Sterge gives them lessons in linear
algebra, along with a reading list that includes Amir D. Aczel's "Fermat's
Last Theorem," about the quest to solve a 360-year-old math problem.
Back from a plane trip recently, Mr. Sterge, gazing at a picture of the
Rockies in his office, suddenly challenged his staff to calculate how far the
horizon was from 40 miles up. Then he solved the problem for them: 593
miles.
Cooper Neff's program trading went global after the firm was acquired at
the end of 1994 by Banque Nationale de Paris. The bank's BNP
Arbitrage unit gave Cooper Neff money to manage -- as much as $11
billion -- and asked it to reproduce its system in BNP offices around the
world.
The French Way
At first, cultures clashed. Setting up at the bank's Paris office, Mr. Sterge
and his fellow Americans would start work at 7 a.m. and see the French
traders roll in at 9:55, just before the market opened. The French knocked
off for lengthy lunches while the Americans, loath to stop trading, ordered
in, just as in Radnor.
Asked about this, Yann Gerardin, managing director of BNP Arbitrage,
shrugs. "Have you been in Radnor?" he asks, conjuring up images of office
parks and lawns and not much else. "Then you understand why you don't
have lunch there. Have you been in Paris? Then you understand why we
do have lunch." Nowadays, Paris traders are eating at their desks more
often.
Language, however, has posed few problems. Not many of the Americans
spoke good French. But "I can make more money with their math than
with their French," Mr. Gerardin says.
The firm now trades about 90 million shares a week in Italy, Mr. Sterge
says, along with 30 million a week in Australia and Japan and five million a
week in France, Germany and Spain. This comes to about 5% of the
national markets except for Japan, where it is less. Canada and Britain are
next up. Although each market is traded separately, Mr. Sterge doesn't
see them that way; he sees the world as 7,000 stocks to trade against one
another in one gigantic hedge fund.
In the U.S., some rivals have questioned whether the firm can do such big
volume and still adhere to the New York Stock Exchange's "collars."
These require that when the Dow Jones Industrial Average rises or falls 50
points, program traders shifting between stocks and stock-index futures or
options may only buy stocks in falling markets or sell stocks in rising
markets. One rival trading executive says firms have raised this issue with
the Big Board, but it is satisfied BNP/Cooper Neff is playing by the rules.
The exchange says it doesn't comment on members' trading.
The firm's trading may grow further if liquidity in the market, Mr. Sterge's
math and business opportunities permit. But "obviously there is a limit,"
Mr. Gerardin says. "We're not going to trade against ourselves." |
