The Quants Part 6

Simons got serious about making money. He started an investment firm called Monemetrics in a strip mall near the East Setauket train station. He made a call to Lenny Baum, an IDA crypta.n.a.lyst who'd done work on automated speech recognition technology. Simons thought Baum, one of the sharpest mathematicians he'd ever met, could use his quant.i.tative brilliance to make hay in the market.

Baum's chief achievement at IDA was the Baum-Welch algorithm, which he and fellow IDA mathematician Lloyd Welch designed to unearth patterns in an obscure mathematical phenomenon called a hidden Markov process. The algorithm proved to be an incredibly effective code-breaking tool, and also has interesting applications for financial markets.

A Markov process, named after Russian mathematician Andrey Markov, models a sequence of events in a system that have no direct relation to one another. Each roll of the dice in a game of Monopoly, for instance, is random, although the outcome (which square you land on) depends on where you are on the board. It is, in other words, a kind of random walk with contingent variables that change with each step along the way.

A hidden hidden Markov process models a system that depends on an underlying Markov process with unknown parameters. In other words, it can convey information about some kind of underlying, random sequence of events. For instance, imagine you are talking on the phone with a friend who is playing a game of Monopoly. He yells "Darn!" each time he lands in jail, or "Eureka!" each time his opponent lands on his Park Place property, as well as a sequence of other exclamatory giveaways. With enough data and a powerful computer, the Baum-Welch algorithm can tease out probabilities about this process-and at times even predict what will come next. Markov process models a system that depends on an underlying Markov process with unknown parameters. In other words, it can convey information about some kind of underlying, random sequence of events. For instance, imagine you are talking on the phone with a friend who is playing a game of Monopoly. He yells "Darn!" each time he lands in jail, or "Eureka!" each time his opponent lands on his Park Place property, as well as a sequence of other exclamatory giveaways. With enough data and a powerful computer, the Baum-Welch algorithm can tease out probabilities about this process-and at times even predict what will come next.

Baum was skeptical. He'd never been interested in investing. But Simons was persistent. "Why should I do this?" Baum asked during one of their many phone conversations. "Will I live longer?"

"Because you'll know you lived," Simons replied.

Baum gave in. He started commuting to Long Island from Princeton to work at Monemetrics. Both were still relative novices in the investing game, and Baum found little use for his mathematical skills in the financial realm. Instead, he proved to be a brilliant fundamental trader, wagering on the direction of currencies or commodities based on his a.n.a.lysis of the economy or twists and turns in government policies.

But Simons was stuck on the notion of creating mathematically grounded trading models. He turned to a Bronx-born math professor he'd hired while running the math department at Stony Brook, James Ax.

Ax looked at Baum's algorithms and determined that he could use them to trade all kinds of securities. In the mid-1980s, Simons and Ax spun a fund out of Monemetrics called Axcom Ltd. In 1985, Ax moved the operation to Huntington Beach, California. Axcom was to act as the trading advisor for the fund, which was nominally run as an investing firm owned by a company Simons had founded in July 1982 called Renaissance Technologies.

Soon Simons's growing crew of quants added another math wizard, Elwyn Berlekamp, a game theory expert at Berkeley. Like Ed Thorp, Berlekamp had worked with Claude Shannon and John Kelly at MIT. He'd briefly met Simons during a stint at IDA in the 1960s.

The fund put up solid returns for several years, even managing to trade through Black Monday with relatively little damage. In 1988, Ax and Simons renamed the fund Medallion in honor of a math award they'd both won. Almost as soon as they'd renamed the fund, things started going south for Medallion. In the second half of 1988, losses were piling up, and getting steeper every month. By April 1989, it had dropped nearly 30 percent. Alarmed by the s.h.i.+ft in fortunes, Simons ordered Ax to stop trading. Ax resisted, convinced he could turn things around. He hired a lawyer, threatening to sue. Simons spoke with his own lawyer.

In June, Berlekamp, who'd been gone for several months on a trip to Egypt, swung by Medallion's office. He was surprised at how the situation had deteriorated. He quickly provided a solution, offering to buy out Ax's stake, which represented two-thirds of its a.s.sets. Ax agreed. So did Simons.

With Ax gone, it was time to get to work on revamping the fund's trading system. Berlekamp moved Medallion's headquarters north to Berkeley so he could focus on overhauling the strategy without worrying about the commute. He rented the entire ninth floor of an office building on Shattuck Avenue near the university and s.h.i.+pped in the fund's computers. For several months Berlekamp and Simons sweated over how to turn around Medallion's fortunes.

A crucial change was a s.h.i.+ft to higher-frequency trading. Typically, the fund would hold on to positions for several days, even weeks. Berlekamp and Simons decided to shorten average holding periods to less than a day, or even just an hour, depending on how far a position moved. From a statistical point of view, they realized, the ability to predict what will happen tomorrow, or in the next few hours, is far better than the ability to predict what will happen a week or two down the road.

To Berlekamp, it was like betting strategies in card games such as blackjack. In blackjack, the bettor's edge is small. But that's okay, since the law of large numbers is on his side. If the bettor plays ten thousand hands a month, his chances of being down are very small (if he plays his cards right). With just one bet, a gambler has to be very sure that his advantage is quite large. That's why the goal was to make a lot of bets, as many as possible, just as long as there was also a slight statistical edge.

By November 1989, Medallion was up and running again. And it was an immediate success. In 1990, it gained 55 percent after fees. The team at Medallion kept tweaking the models, and the performance kept improving. Simons kept bringing on board math whizzes, including Henry Laufer, another Stony Brook don, to work for Renaissance. Laufer had earned a degree in physics from Princeton in 1965 and published a book on black holes in 1971 called Normal Two-Dimensional Singularities Normal Two-Dimensional Singularities. He was an advisor to Renaissance's commodity traders in the 1980s and joined the firm full-time in January 1991.

Simons closed the fund to new investors in 1993 with $280 million in a.s.sets. He didn't think the models could handle much more cash. In 1994, returns. .h.i.t an eye-popping 71 percent. The great run by Medallion was on. Month after month, quarter after quarter, year after year the money kept rolling in. The fund's success became so reliable that its researchers and traders (all sporting Ph.D.'s) forgot what it was like to lose. When Medallion posted a rare 0.5 percent loss in a single quarter of 1999, at least one employee actually wept.

Meanwhile, Simons had tapped into Morgan Stanley's stat arb machine created in the 1980s by purchasing Kepler Financial Management, the fund set up by Robert Frey after he'd left Nunzio Tartaglia's APT group. The fund had a rough start, but it eventually started hitting on all cylinders. In 1997, it was absorbed into the Medallion mother s.h.i.+p and called the Factor Nova Funds, adding stat arb firepower to an already state-of-the-art investment machine. It was the first step in making Medallion a genuine multistrategy fund.

By then, Berlekamp was gone. He'd left Renaissance at the end of 1990 to pursue academic interests at Berkeley, where he went on to crack game theory puzzlers such as mathematical chess. But the Medallion legend continued to grow. To be sure, the fund has had a few hiccups over the years. In March 2000, when the dot-com bubble began to implode, reversing trends in technology stocks that had been in place for several years, Medallion lost $250 million in three days, nearly wiping out its year-to-date profit. But the fund quickly bounced back and put up another year of stellar returns.

Every trader on Wall Street who has heard of the fund's mind-bending performance has openly marveled: how do they do it?

Simons has let few clues drop over the years. He once remarked that the fund sifts through data for identifiable patterns in prices. "Patterns of price movements are not random," he said, a shot across the bow of the efficient-market random walkers such as Eugene Fama. "However, they're close enough to random so that getting some excess, some edge out of it, is not easy and not so obvious, thank G.o.d."

After chuckling at this cryptic statement, Simons added: "G.o.d probably doesn't care."

One day in 2003, Paul Samuelson came to speak at Renaissance's headquarters in East Setauket. The MIT economist and n.o.bel laureate had long held that it was impossible to beat the market. He qualified that statement by saying that if anyone could do it, they would hide away and not tell anyone about their secret.

"Well, it looks like I've found you," Samuelson said to the laughter of the wealthy quants of East Setauket.

How does Renaissance detect nonrandom price movements? It's almost the same as asking whether Renaissance knows the Truth. Renaissance detect nonrandom price movements? It's almost the same as asking whether Renaissance knows the Truth.

The fact is, no one outside the offices of Renaissance Technologies knows the answer to how it detects nonrandom price movements. Few people who have joined Renaissance have ever left. Those who have aren't talking.

There are a few clues, however. One is the large number of cryptographers who helped to create Medallion: Ax, Berlekamp, and of course Simons himself. Cryptographers are trained to detect hidden messages in seemingly random strings of codes. Renaissance has applied that skill to enormous strings of market numbers, such as tick-by-tick data in oil prices, while looking at other relations.h.i.+ps the data have with a.s.sets such as the dollar or gold.

Another clue can be found in the company's decision in the early 1990s to hire several individuals with expertise in the obscure, decidedly nonWall Street field of speech recognition.

In November 1993, Renaissance hired Peter Brown and Robert Mercer, founders of a speech recognition group at IBM's Thomas J. Watson Research Center in Yorktown Heights, New York, in the hills of Westchester County. Brown came to be known as a freakishly hard worker at the fund, often spending the night at Renaissance's East Setauket headquarters on a Murphy bed with a whiteboard tacked to the bottom of it. Worried about his health, he became an avid squash player because he deduced that it was the most efficient method of exercising. Often seen in the fund's office in rumpled clothes, a stack of pens stuffed in his pockets, Brown had the ability to tackle the toughest mathematical conundrums as well as wire up the most advanced computers.

Mercer, meanwhile, was simply known as the "big gun" at Renaissance. When a th.o.r.n.y problem cropped up that required focused attention, the firm would "just aim Bob at it and fire," said a former employee.

Over the following years, Renaissance hired a slew of people from IBM's voice recognition group, including Lalit Bahl and the brothers Vincent and Stephen Della Pietra. Internet searches on any of these names will spit out a series of academic papers written in the early to mid-1990s. Then the trail goes cold.

At first blush, speech recognition and investing would appear to have little in common. But beneath the surface, there are striking connections. Computer models designed to map human speech depend on historical data that mimic acoustic signals. To operate most efficiently, speech recognition programs monitor the signals and, based on probability functions, try to guess what sound is coming next. The programs constantly make such guesses to keep up with the speaker.

Financial models are also made up of data strings. By glomming complex speech recognition models onto financial data, say a series of soybean prices, Renaissance can discern a range of probabilities for the future directions of prices. If the odds become favorable ... if you have an edge ...

It's obviously not so simple-if it were, every speech recognition expert in the world would be running a hedge fund. There are complicated issues involving the quality of the data and whether the patterns discovered are genuine. But there is clearly a powerful connection between speech recognition and investing that Renaissance is exploiting to the hilt.

A clue to the importance of speech recognition to Renaissance's broader makeup is that Brown and Mercer were named co-CEOs of Renaissance Technologies after Simons stepped down in late 2009.

"It's a statistical game," said Nick Patterson, a former Renaissance a.n.a.lyst and trader who'd previously done work as a cryptographer for the British and U.S. governments. "You discern phenomena in the market. Are they for real? That's the key question. You must make sure it's not model error or just noise."

If the phenomenon is "for real," capitalizing on it can be an even tougher challenge. How much leverage should be used? How much cash can be tossed at the strategy before it vanishes into thin air? The deep thinkers at Renaissance considered all of these issues and more. "Our edge was quite small, but it's like being the house player at a casino," Patterson added. "You have a small edge on every bet, and you have to know how to handle that."

A common thread runs through voice recognition technology and cryptography: information theory. Indeed, information theory sprouted in part from the government's efforts to crack codes during World War II. In financial markets, cryptographers try to discover hidden patterns that will recur in the future.

Medallion may tweak its models more than outsiders believe. One person familiar with the fund says it adjusts models for market conditions far more frequently than most quant operations. The switches are based on complex market signals discerned by Medallion's powerful computers. Since its trades are processed so rapidly and Medallion trades in so many markets, this gives the fund more flexibility to s.h.i.+ft its focus than most one-trick-pony quant funds.

Perhaps no one is more astounded at the Medallion fund's two-decades-and-running streak than Simons himself. Throughout the 1990s, employees at Renaissance braced themselves for an end to the spectacular, lotterylike success. In 1992, the senior staff held a meeting to discuss the prospects for the fund over the next decade. Most expected to be in a different line of work in ten years. Simons was known for constantly saying, "The wolf is at the door."

So paranoid is Simons about the threat of employees leaving the fund, taking its special sauce elsewhere, that he's more than willing to ruin the careers of such apostates. In December 2003, Renaissance sued two employees, Alexander Belopolsky and Pavel Volfbeyn, who'd left the firm to join New York hedge fund giant Millennium Partners. The suit accused the two former MIT physicists of misappropriating trade secrets. In his defense, Volfbeyn accused Renaissance of asking him to devise methods to "defraud investors trading through the Portfolio System for Inst.i.tutional Trading, or POSIT," referring to a dark pool of liquidity-essentially an electronic market that matches buy and sell orders for stocks out of the public eye. Volfbeyn said he was instructed to create a code that would "reveal information that POSIT intended to keep confidential," according to an article by Bloomberg, and that he refused to partic.i.p.ate in the scheme, as well as others, because he believed they were against the law. The suit also hinted at a nefarious swaps deal that he described as a "ma.s.sive scam," but didn't explain the deal in detail.

Nothing ever came of the allegations, and the two parties eventually settled their differences. But the message to Renaissance's employees had been sent.

Insiders say the pressure to succeed at Renaissance can be brutal. One mathematician at the fund may have succ.u.mbed to the pressure on March 1, 2006. That's when Alexander Astashkevich, a thirty-seven-year-old MIT graduate who worked at Renaissance, shot and killed his estranged wife in the small town of Port Jefferson, Long Island, before turning the shotgun on himself. He left behind a six-year-old son named Arthur.

Perhaps the intense pressure explains why Simons was known to burn through three packs of Merit cigarettes a day. One day Patterson came into Simons's office to discuss a management issue. After some time, he noticed that Simons, puffing away at a Merit, wasn't listening-he was transfixed by the flitting numbers on his screen: numbers showing big losses in the Medallion fund. Even though Medallion always seemed to claw back from such dips, which were part and parcel of running a fund, only to rack up more gains, they caused Simons's stomach to churn every time. Robert Frey, who left Renaissance in 2004, said one of the biggest reasons he quit was that he couldn't take the gut-wrenching day-to-day volatility anymore. Despite Medallion's success, it always seemed ephemeral, as if one day the magic would go away, vanish like a genie into its bottle. As if one day the Truth wouldn't be the Truth anymore.

In between developing the most successful trading programs in the world, Renaissance's wealthy band of quants found time to relax in the exclusive environs of East Setauket and Port Jefferson. Simons and Laufer, the fund's "chief scientist," owned mansions perched on the Long Island Sound, just a few minutes' drive from the firm's headquarters. Simons loved to take his staff sailing on his luxury yacht or jet off to exclusive resorts such as Atlantis in the Bahamas.

Rival quants such as Peter Muller and Cliff Asness, meanwhile, looked upon Medallion's chart-crus.h.i.+ng success, year after year, with awe. None had any idea how Simons had done it. No matter what the market was doing, Medallion cranked out billions in profits. Many wondered: had Simons and his band of reclusive quants out in the woods of Long Island discovered the holy grail, the philosopher's stone-the secret mythical Truth of the financial markets? Perhaps, they thought with envy, Simons really had cracked the code.

One thing was certain: Simons wasn't talking.

THE MONEY GRID

By the late 1990s, Ken Griffin was swapping convertible bonds from a high tower in Chicago. Jim Simons was building his quant empire in East Setauket. Boaz Weinstein was scouring computer screens to trade derivatives for Deutsche Bank. Peter Muller was trading stocks at Morgan Stanley. Cliff Asness was measuring value and momentum at AQR. They were all making more money than they'd ever dreamed possible. late 1990s, Ken Griffin was swapping convertible bonds from a high tower in Chicago. Jim Simons was building his quant empire in East Setauket. Boaz Weinstein was scouring computer screens to trade derivatives for Deutsche Bank. Peter Muller was trading stocks at Morgan Stanley. Cliff Asness was measuring value and momentum at AQR. They were all making more money than they'd ever dreamed possible.

And each was becoming part of and helping to create a ma.s.sive electronic network, a digitized, computerized money-trading machine that could s.h.i.+ft billions around the globe in the blink of an eye, at the click of a mouse.

This machine has no name. But it is one of the most revolutionary technological developments of modern times. It is vast, its octopuslike tentacles reaching to the farthest corners of civilization, yet it is also practically invisible. Call it the Money Grid.

Innovators such as Ed Thorp, Fischer Black, Robert Merton, Barr Rosenberg, and many others had been early architects of the Money Grid, designing computerized trading strategies that could make money in markets around the world, from Baghdad to Bombay, Shanghai to Singapore. Michael Bloomberg, a former stock trader at Salomon Brothers and eventual mayor of New York City, designed a machine that would allow users to get data on virtually any security in the world in seconds, turning its creator into a billionaire. The Nasdaq Stock Market, which provided entirely electronic transactions, as opposed to the lumbering humans at the New York Stock Exchange, made it quicker and cheaper to buy and sell stocks around the globe. The entire global financial system became synced into a push-b.u.t.ton electronic matrix of unfathomable complexity. Money turned digital.

Few were as well placed to take advantage of the Money Grid as the Floridian boy wonder Ken Griffin.

[image]GRIFFIN[image]

Griffin's fortress for money, Citadel Investment Group, started trading on November 1, 1990, with $4.6 million in capital. The fund, like Princeton/Newport Partners, specialized in using mathematical models to discover deals in the opaque market for convertible bonds.

In its first year, Citadel earned a whopping 43 percent. It raked in 41 percent in its second, and 24 percent in its third.

One of Citadel's early trades that caught the Street's eye concerned an electronic home security provider called ADT Security Services. The company had issued a convertible bond that contained a stipulation that if a holder converted the bond into stock, he wouldn't be eligible for the next dividend payment. That meant that the bond traded at a slight discount to its conversion value, because the holder wouldn't receive the next dividend.

Griffin and his small band of researchers figured out that in the United Kingdom, the dividend was technically not a dividend but a "scrip issue"-which meant that a buyer of the bond in the United Kingdom would be paid the dividend. In other words, the bond was cheaper than it should be.

Citadel bought as many of these bonds as it could buy. It was a trade that a number of the large dealers had missed, and it was a trade that put Citadel on the map as a shop that was on top of its game.

By then Griffin, still a boy-faced whiz kid in his mid-twenties, was juggling nearly $200 million with sixty people working for him in a three-thousand-square-foot office in Chicago's Loop district.

Then he lost money. A great deal of money. In 1994, Alan Greenspan and the Federal Reserve shocked the market with a surprise interest rate increase. The bottom fell out of the rate-sensitive convertible bond market. Citadel dropped 4.3 percent, and a.s.sets under management fell to $120 million (part of the decline came from worried investors pulling money out of the fund). Until 2008, it was the only year Citadel's flags.h.i.+p Kensington fund lost money.

Used to unstinting success, Griffin was stunned, and h.e.l.l-bent on making sure his financial battlements couldn't be breached in the future.

"We're not going to let this happen again," he told his patron, Frank Meyer. Citadel began crafting plans to fortify its structure, inst.i.tuting changes that may have saved it from a complete collapse fourteen years later. When investors had seen the bond market crumbling, they had called up Griffin in a panic and demanded a refund. Griffin knew that the market was eventually going to bounce back, but there was little he could do. The solution: lock up investors for years at a time. He slowly began negotiating the new terms with his partners, eventually getting them to agree to keep their investment in Citadel for at least two years (and at the end of each two-year period agreeing to another two-year lockup). A long lockup meant that when times got tough, Griffin could remain calm, knowing that fidgety, fleet-footed investors couldn't cut and run at a moment's notice. By July 1998, the new model was in place-and just in time.

Later that year, Long-Term Capital crashed. As other hedge funds sold indiscriminately in a broad, brutal deleveraging, Citadel snapped up bargains. Its Kensington fund gained 31 percent that year. By then, Citadel had more than $1 billion under management. The fund was diving into nearly every trading strategy known to man. In the early 1990s, it had thrived on convertible bonds and a boom in j.a.panese warrants. In 1994, it launched a "merger arbitrage" group that made bets on the shares of companies in merger deals. The same year, encouraged by Ed Thorp's success at Ridgeline Partners, the statistical arbitrage fund he'd started up after shutting down Princeton/Newport, it launched its own stat arb fund. Citadel started dabbling in mortgage-backed securities in 1999, and plunged into the reinsurance business a few years later. Griffin created an internal marketmaking operation for stocks that would let it enter trades that flew below Wall Street's radar, always a bonus to the secrecy-obsessed fund manager.

As Griffin's bank account expanded to eye-watering proportions, he began to enjoy the perks of great wealth. Following a well-trodden path among the rich, he indulged his interest in owning great works of art. In 1999, he snapped up Paul Cezanne's Curtain Jug and Fruit Bowl Curtain Jug and Fruit Bowl for $60.5 million. Later that year, he became enamored of an Edgar Degas sculpture called for $60.5 million. Later that year, he became enamored of an Edgar Degas sculpture called Little Dancer, Aged Little Dancer, Aged 14, that he chanced upon in Sotheby's auction house in New York. He later bought a version of the sculpture, as well as a Degas pastel called 14, that he chanced upon in Sotheby's auction house in New York. He later bought a version of the sculpture, as well as a Degas pastel called Green Dancer Green Dancer. Meanwhile, in 2000, he sh.e.l.led out $6.9 million for a two-story penthouse in a ritzy art deco building on North Michigan Avenue in Chicago, an opulent stretch of properties known as the Magnificent Mile.

Citadel's returns had become the envy of the hedge fund world, nearly matching the gains put up by Renaissance. It posted a gain of 25 percent in 1998, 40 percent in 1999, 46 percent in 2000, and 19 percent in 2001, when the dot-com bubble burst, proving it could earn money in good markets and bad. Ken Griffin, clearly, had alpha.

By then, Griffin's fund was sitting on top of a cool $6 billion, ranking it among the six largest hedge funds in the world. Among his top lieutenants were Alec Litowitz, who ran the firm's merger arbitrage desk, and David Bunning, head of global credit. In a few years, both Litowitz and Bunning would leave the fund. In 2005, Litowitz launched a $2 billion hedge fund called Magnetar Capital that would play a starring role in the global credit crisis several years later. A magnetar is a neutron star with a strong magnetic field, and Litowitz's hedge fund turned out to have a strong attraction for a fast-growing crop of subprime mortgages.

Citadel, meanwhile, was quickly becoming one of the most powerful money machines on earth, fast-moving, extremely confident, and muscle-bound with money. It had turned into a hedge fund factory, training new managers such as Litowitz who would break off and grow new funds. Ed Thorp's progeny were spreading like weeds. And Griffin, just thirty-three years old, was still the most successful of them all.

The collapse of Enron in 2001 gave him a chance to flex his muscles. In December 2001, a day after the corrupt energy-trading firm declared bankruptcy, Griffin hopped on a plane to start recruiting energy traders from around the country. Back in Chicago, a team of quants started building commodity-pricing models to ramp up the fund's trading operation. The fund also signed up a number of meteorologists to help keep track of supply-and-demand issues that could impact energy prices. Soon Citadel sported one of the largest energy-trading operations in the industry.

As his fund grew, Griffin's personal wealth soared into the stratosphere. He was the youngest self-made member of the Forbes 400 in 2002. The following year, he was number ten on Fortune's Fortune's list of the richest people in America under forty years old, with an estimated net worth of $725 million, a hair behind Dan Snyder, owner of the Was.h.i.+ngton Redskins. list of the richest people in America under forty years old, with an estimated net worth of $725 million, a hair behind Dan Snyder, owner of the Was.h.i.+ngton Redskins.

He'd reached a level of success few mortals can contemplate. To celebrate that year, he got married-at the Palace of Versailles, playground of Louis XIV, the Sun King. Griffin exchanged vows with Anne Dias, who also managed a hedge fund (though a much, much smaller one). The reception for the two-day affair was held in the Hameau de la Reine, or "Hamlet of the Queen," where Marie Antoinette lived out Jean-Jacques Rousseau's back-to-nature peasant idyll in an eighteenth-century faux village.

The Canadian acrobat squad Cirque du Soleil performed. Disco diva Donna Summer sang. Guests dangled from helium balloons. The party in Paris included festivities at the Louvre and a rehearsal dinner at the Musee d'Orsay.

It was good to be Ken Griffin. Perhaps too too good. good.

[image]MULLER[image]

Just as Griffin was starting up Citadel in Chicago, Peter Muller was hard at work at Morgan Stanley in New York trying to put together his own quant.i.tative trading outfit using the models he'd devised at BARRA. In 1991, he pulled the trigger, flipping on the computers.

It was a nightmare. Nothing worked. The sophisticated trading models he'd developed at BARRA were brilliant in theory. But when Muller actually traded with them, he ran into all sorts of problems. The execution wasn't fast enough. Trading costs were lethal. Small bugs in a program could screw up an order.

He'd set up shop on the thirty-third floor of Morgan's headquarters inside the Exxon Building at 1251 Avenue of the Americas, the same skysc.r.a.per that had housed Bamberger and Tartaglia's stat-arb experiment, with several Unix workstations, high-end computers designed for technical applications and complex graphics. His first hire was Kim Elsesser, a programmer with a master's degree in operations research from MIT. Elsesser was thin, tall, blond, and blue-eyed: a perfect target for the testosterone-soaked Morgan traders. She was also a highly gifted mathematician and computer programmer. She'd first joined Morgan in January 1987 before leaving for grad school in Cambridge, then returned to the bank in 1992. Within a few months, she signed up with Muller. He dubbed his new trading outfit Process Driven Trading, PDT for short. "Process-driven" was essentially shorthand for the use of complex mathematical algorithms that only a few thousand people in the world understood at the time.

Muller and Elsesser built the operation from scratch. They wrote trading models in computer code and hooked up their Unix workstations to Morgan's mainframe infrastructure, which was plugged into major exchanges around the world. Muller designed the models, and Elsesser, familiar with Morgan's system, did most of the programming. They started trading in the United States, then added j.a.pan, followed by London and Paris. They would trade once a day, based on their models. They worked crazy hours, but it all seemed for naught.

Muller was able to glean tidbits of information from other fledgling groups of mathematicians who were trying to crack the market's code. In 1993, he paid a visit to a little-known group of physicists and scientists running a cutting-edge computerized trading outfit from a small building in Santa Fe, New Mexico. They called themselves Prediction Company, and they were reaching out to Wall Street firms, including Morgan Stanley, for seed capital. Muller's job was to check them out.

A founder of Prediction Company was Doyne Farmer, a tall, ropy physicist and early innovator in an obscure science called chaos theory. Given more to tie-dyed T-s.h.i.+rts and flip-flops than the standard-issue Wall Street suit and tie, Farmer had followed in Ed Thorp's footsteps in the 1980s, creating a system to predict roulette using cutting-edge computers wedged into elaborate "magic" shoes. Also like Thorp, Farmer moved on from gambling in casinos to making money using mathematics and computers in financial markets around the world.

Muller and Farmer met at the company's office on 123 Griffin Street in Santa Fe, otherwise known as the "Science Hut." Muller's questions came quick and fast. When Farmer would ask for information in return, Muller, poker player that he was, held his cards close to his vest. Eventually Farmer had enough.

"We had to kick him out," Farmer later recalled. "If you give someone a piece of information that they can use, you expect to get something in return that you can use. It makes sense. But Pete didn't give us anything."

Farmer didn't realize that Muller didn't have much of anything to give. Not yet.

Later that year, Morgan's management was looking to trim the fat. PDT was in the crosshairs. The firm had paid a lot of money to Muller, and he wasn't delivering. John Mack, the bond trader who'd recently been named president, called a meeting to hear managers defend their businesses.

Muller wore a suit to the meeting. His hair was oiled and combed, rather than in its usual floppy-banged tangle. A team of tight-lipped Morgan execs sat around a long table in a warm, dusky conference room. Muller had to wait as several managers made their survival pitches. Their desperation was evident. Muller made a mental note: Stay calm, look cool, be confident Stay calm, look cool, be confident. When his turn came, he flatly admitted that PDT hadn't succeeded yet. But it was on the edge of great things. Computerized trading was the future. He just needed more time.

As he stopped speaking, he looked at Mack, who gave him a confident nod back. Mack had bought in.

Perseverance paid off, and soon there were signs PDT was beginning to grasp the Truth, or at least a small corner of it-they turned a profit. The day they made their first million dollars, Muller and Elsesser tossed themselves a party (consisting of cheap wine in plastic cups). In short order, a million would be a sleepy morning yawn, a blink of the eye.

In early 1994, Muller put together his dream team of math and computer aces: Mike Reed, soft-spoken geophysicist with a Ph.D. in electrical engineering from Princeton; Ken Nickerson, the ultimate number cruncher, a tall, brooding math expert with a Ph.D. in operations research from Stanford; Shakil Ahmed, a skin-and-bones computer programming whiz from Princeton; and Amy Wong, who sported a master's in electrical engineering from MIT. This small group would form the core of what soon became one of the most profitable, and little known, trading operations in the world.

Aside from deep pockets, Muller had another advantage in working for a giant investment bank. Other trading outfits, such as hedge funds, funneled their trades to exchanges such as the NYSE through regulated broker dealers, including Morgan. One hedge fund that used Morgan as its brokerage for stocks was a trading group at Renaissance Technologies called Nova, run by Robert Frey, the mathematician who'd worked under Nunzio Tartaglia at Morgan Stanley.

In the mid 1990s, the Nova fund had a bad stretch. PDT took the positions off Renaissance's hands and folded them into its own fund. It worked out quite well, as the positions eventually became profitable and also gave Muller a rare glimpse inside Renaissance's secret architecture. Renaissance, for its part, retooled Nova into a profit-generating machine.

[image]

By 1994, the stage was set. Muller had the money and the talent to go to work. They didn't have much time. Mack would slam the door shut in a heartbeat if he thought the group wasn't delivering. 1994, the stage was set. Muller had the money and the talent to go to work. They didn't have much time. Mack would slam the door shut in a heartbeat if he thought the group wasn't delivering.

Working late hours and weekends, PDT's dream team built an automatic trading machine, a robot for making money. They called their robot Midas-as if everything it touched would turn to gold. Nickerson and Ahmed did the fine-grained number crunching, searching for hidden signals in the market that would tell the computer which stocks to buy and sell. Nickerson focused on the U.S. market, Ahmed overseas. Reed built up the supercomputer infrastructure, mainlining it into financial markets around the world. The strategy was statistical arbitrage-the same strategy Bamberger had devised at Morgan Stanley in the 1980s. PDT's quants had largely discovered how to implement the strategy on their own, but there's little question that by the time Midas was up and running, the idea of stat arb was in the air. Doyne Farmer's Prediction Company was running a stat arb book in Sante Fe, as were D. E. Shaw, Renaissance, and a number of other funds. Over the years, however, few stat arb funds would do nearly as well as PDT, which in time became the most successful proprietary trading desk on Wall Street in terms of consistency, longevity, and profitability.

Midas focused on specific industries: oil drillers such as Exxon and Chevron, or airline stocks such as American Airlines and United. If four airline companies were going up and three were going down, Midas would short the stocks going up and buy the stocks going down, exiting the position in a matter of days or even hours. The tricky part was determining exactly when to buy and when to sell. Midas could do these trades automatically and continuously throughout the day. Better yet, Midas didn't ask for a fat bonus at year's end.

By the fourth quarter of 1994, the money started piling up. Midas was king. Just flip on the switch and zzip-zip-zip ... zap ... zap ... zooing ... bapbapbapbap ... zing ... zing ... zap! zzip-zip-zip ... zap ... zap ... zooing ... bapbapbapbap ... zing ... zing ... zap! The digitized computerized trades popped off like firecrackers, an electronic gold mine captured in upward-flying digits on PDT's computer screens as the money rolled in like magic. The digitized computerized trades popped off like firecrackers, an electronic gold mine captured in upward-flying digits on PDT's computer screens as the money rolled in like magic.

It was amazing, exhilarating, and at times terrifying. One night Elsesser was riding home in a taxicab, exhausted after a long day's work. The buildings and lights of the city flashed by in a blue blur. The driver's radio was an annoying fuzz in the background. A piece of news broke through the static: a radio announcer was describing unusual trading activity that was wreaking havoc in Tokyo's markets.

Elsesser's ears p.r.i.c.ked up. Could that be us? s.h.i.+t Could that be us? s.h.i.+t.

Frantic, she ordered the taxi to take her back to Morgan's headquarters. She was always worried that some glitch in their computer program could unleash tsunamis of buy or sell orders. You never knew if the system would go haywire like some kind of computerized Frankenstein. PDT wasn't responsible for the chaos in Tokyo that day, but the possibility always lurked in the background. It could be hard to sleep at night with the computers whirring away.