My Life as a Quant (22 page)

Goldman's role was that of custom tailor. The funds could have sold listed interest-rate options and bond futures contracts, available on the Chicago exchanges, to make their bets. But these listed instruments provided only a limited number of standardized strikes and expirations, too constrained to match the particular strikes and expirations of the bonds actually owned by a fund. To accommodate them, Goldman bought over-the-counter (OTC) call options from each fund, each option a privately negotiated contract written on the specific bonds held by that fund. The options contacts were chosen to expire close to the dates on which the funds had to report their quarterly earnings. If interest rates did not decline and the funds won their bet, they kept the premium and enhanced their reported return.

The enterprise of selling bond options, like many early-stage businesses, was particularly profitable. A trading desk could charge clients an extra fee for an option that was custom-tailored to fit a fund's exact needs. The bond options desk at Goldman offset the risk of the tailored option they had sold by buying a bundle of cheaper listed options and futures that approximately matched its payoff. There was some risk involved in hedging tailored options with listed ones; the mismatch in strikes and expirations made the hedge approximate rather than exact. You could think of the fee as our desk's compensation for being willing to accept and manage this risk.

To hedge one option with another, firms like Goldman needed a model that told you each option's value and its sensitivity to changes in interest rates. The renowned Black-Scholes model for stock options, which Ravi had told me to learn, did not strictly apply to bond options. Stocks are relatively simple; they guarantee no future dividend payments and have no natural termination date, so their future prices are unconstrained. Treasury bonds are much more intricate: Because they promise to repay their principal when they mature, their price on that date is constrained to be par. Furthermore, since all Treasury bonds can be decomposed into a sum of more primitive zero-coupon bonds of varying maturities, they are all interrelated.

My new boss Ravi had heuristically modified the Black-Scholes stock option model to make it work, at least approximately, for short-dated Treasury bond options. He had written a computer program to implement it, and the bond options desk now priced and hedged their options by means of it. As they got more experienced at using it, Peter Freund's desk discovered that Ravi's model was fine for short-term options but questionable for longer-term ones; it suffered from a variety of theoretical inconsistencies stemming from its inadequate modeling of the long-term behavior of bond prices. It was an ingenious first cut, quickly created to catapult the desk into doing business, but now both the model and its computer interface needed work. A few days after I arrived, Ravi directed me to extend the model and the program. It was a feet-first introduction to working with traders, and much of what I know about the need to be pragmatic and business-oriented as a quant I learned in those first few months working with Ravi and the desk.

From the desk's point of view, the greatest hindrance to exploiting the model for business was not the theory, but rather the lack of a graphical user interface. Each time a salesperson needed to value an option for a potential trade with a client, he or she had to type in, on one line after another, the bond's current price, maturity, and coupon as well as the option's expiration and strike; then the salesperson had to enter the current short-term interest rate and the bond's assumed future yield volatility. One more tap on the return key and the program computed the model's theoretical price and told you how to hedge it with the underlying Treasury bond. If you wanted to compute the option value for a variety of volatilities, expirations or strikes, you had to repeat the same sequence, entering items and hitting return keys all over again.

Setting up a trade could take several days—a typical client might get a quote from Goldman, hang up, call another dealer to get their price, ponder a while, and then call us back the next day to continue the discussion. At that point someone on our desk would have to start running our model again by reentering all the terms of the deal. This slowed down the interaction with clients, and was far too viscous for a growing business.

The software engineering was deficient, too. Ravi's program had been written in FORTRAN, but the scientific and financial worlds were rapidly moving towards C, which provided much better tools for collaborative software development. Ravi instructed me to learn options theory, rewrite the model in C, and construct a friendlier front end. It was a perfect assignment since it immediately exposed me to theory, implementation, and interaction with the business. I spent the next few days making a rapid first pass through the theory of stock options as described in the original Cox-Ross-Rubinstein binomial paper. Then, I studied the FORTRAN version of Ravi's bond option model and set about rewriting it in C on FSG's VAX computer.

Within a few weeks I was learning on many fronts and adjusting to new social and computing environments. When I ran into hurdles, Ravi grew impatient and irritable. One day, only three weeks or so after I had arrived, he suggested that, since I was taking longer than he had expected, he should perhaps give the work to someone else to do. Upon returning home that evening I took some perverse pleasure in reporting my first taste of Wall Street brutality to Eva, and got some shocked sympathy. I wanted to show her that, with my passage from the civil groves of physics to the harsh engine rooms of capitalism, I had now entered a world in which people ten or more years younger than me found it necessary to crack the whip when I didn't trot fast enough. But in truth, it wasn't that bad—Wall Street has been always meritocratically discourteous. It never seemed to me that great an indignity. In the nineteen years that have elapsed since then, I have been ordered about by young traders, had my back patted and my upper arm encouragingly squeezed in elevators by newly minted partners, and, once, been pushed across the trading floor and cursed in full view of everyone there by an angry and foul-mouthed saleswoman. There is a general lack of respect for age in all of this that makes you disregard your own age, and I like it.

I soon finished programming the model and began to think about attaching a friendly front end to it. In those pre-Macintosh/pre-Windows days, windowing packages were a rarity. Brought up on the UNIX philosophy, I began building my own generic toolkit for data entry and display. I hurriedly force-marched myself through the manual for the UNIX “curses” library that allowed you to flexibly read and write 80-character-by-24-line screens of text, and designed an interface. In a month or so I had created
Bosco
, a new calculator named after an affectionate abbreviation for my son Joshua.

Though the new model was better, it was my new user interface that had the biggest impact. It made negotiating with clients easy. All the model's input and output were visible on one screen, with one field for each item of data to be entered (the bond coupon, maturity, and so on), and one field for each answer the model provided (the option price, its hedge ratio, and so on). There were also fields for storing information about the client and the trade. To run the model and obtain an option price, you simply hit the “Calculate!” key. To change an input field value, you moved the cursor to the relevant field, entered the data, and hit “Calculate!”again. This saved countless keystrokes compared to the command-line interface that drove the previous FORTRAN version. Best of all, at any instant you could save all the details of a potential trade to a computer file for future retrieval. If you used the calculator to value an option during a preliminary conversation with a client, you could store all the information in the file and then resume the discussion the next day exactly where you had left off.

Though primitive by today's standards, this was astonishingly better than what the desk had used before, and the traders and salespeople were overjoyed. By creating and saving the most common types of option trades as templates in files at the start of each day, they could respond rapidly to clients, accommodating many more requests much more efficiently. Ever since, it's been impossible for me to overlook the difference that a simple and well-designed piece of software can make to a business. Despite the genuine glories of quantitative modeling, quant groups often have the most dramatic effect by improving the ergonomics of trading and sales.

When I became head of the Quantitative Strategies Group in the Equities Division a few years later, I always tried to recreate for anyone new in the group my first fortuitous experience with a trading desk. I set them a problem whose solution was useful to traders, and which required both theoretical analysis and software implementation. In that way, I hoped, they would develop a relationship with their end users in the trading division, learn the jargon and style of the business, and combine both theory and practicality.

Like many organizations, Goldman's Financial Strategies Group into which Ravi inducted me that December was rife with politics. Stan Diller, who had left Goldman shortly before I arrived, had reportedly often kidded that his was the only second-rate mind on Wall Street. To replace him and grow the business, the partners in charge of the Fixed Income Division had brought in two finance professors from the midwest. One was tall and lean, fast-talking and confident; the other was much shorter and spoke slowly. Their arrival didn't displace Stan on the mind-ranking charts.

The new structure they imposed was very different from what I had seen of Diller's. Diller appeared to have led FSG autocratically, but with a vision of trading as science. He had accentuated a quantitative approach to financial research, with an emphasis on software development and trading systems. He had sought out PhDs—engineers, physicists, computer scientists, and mathematicians—most of whom arrived ignorant of finance and then learned theory and business on the job. He sought people who could simultaneously do finance, math, and computer science, and had an interdisciplinary view of the world to which I'm still partial.

The new hegemony placed more emphasis on management. They had been given
carte blanche
to grow and they went on a hiring spree that rapidly grew the Strategies group from fifteen or twenty ex-scientists to about a hundred people, many of them professional managers who claimed they knew how to “talk to traders.” They certainly knew enough to realize that they were on to a good thing and wisely made the most of it while it lasted. Over time, the modeling and programming groups became inverted pyramids: one or two technically skilled people at the bottom, who could write programs or build models, supporting a larger number of human conduits above them who passed the results up to the trading desk and then passed the subsequent responses back down again. Having a PhD and being good at research or being able to program well was not an advantage in this structure.

David Garbasz, with his usual accurate eye and sharp tongue, took to calling the two new heads of FSG Mutt and Jeff, and sarcastically referred to the organization they now headed as the “Financial Tragedies Group.” In 1986 Goldman was still getting used to managing quants.

Despite the politics, I loved Goldman and FSG, and made many new friends among the programmers and quants. Most of them were much younger than me—it was my second career and their first. Working away in my undistinguished cubicle—for the first year I had no seat in an office—I did occasionally feel incongruous. One day, as I mindlessly whistled a Beatles song to myself while I programmed my bond option model, I heard the 23-year old kid in the next cube turn around and exclaim astonishedly, “How do
you
know
that
?” In fact, though, age didn't matter that much if you had ability, and the turbulence in financial markets since then—the crashes of 1987 and 1989, the collapse of Long Term Capital Management and Russia's default in 1998—has made the appearance of maturity an advantage.

Among the new people I met was Roscoe, the amiable, cheerily disillusioned leader of a group of good-humored programmer malcontents who occupied the cubicles on what he called Dissident Row. Everyone on Dissident Row ate lunch early and then took a constitutional up to the Brooklyn Bridge and back again. Roscoe's real name was William Dumas and he was rumored to be related to Alexandre Dumas,
père
. He had little respect for the new regime in FSG, and referred caustically to one of the new MIS managers who circulated ancient corporate memos on good programming style from his previous employer as the “Master of the Do-Loop.” Roscoe had a minor genius for creating inventively allusive nicknames for new people in FSG. His method was Cockney rhyming, an associative slang in which, for example, the number five is referred to as “Lady” because Lady is short for Lady Godiva, and Godiva rhymes with “fiver,” the colloquial term for the old British 5-pound note (as in “Can you lend me a fiver?”). In this spirit he dubbed a newly hired Pakistani programmer “Mander,” because his true name was Salah, reminiscent of Salamander. He christened me “E-man,” which I liked, because it was almost identical to the shortened version of “Emanuel” that my family and friends in South Africa had always called me since childhood, and, like Roscoe's other christenings, it stuck with me for the rest of my time at Goldman. After the 1994 fixed-income market slump, Roscoe left Goldman to work for Iris, a financial software company in San Francisco run by an old friend of mine, ex-Wells Fargo quant Jeremy Evnine.

Another new hire at Goldman was a fellow South African, Jonathan Berk, who had come to work at Goldman as an analyst with an undergraduate degree. He was wildly fired up about finance and markets and I soon got my first glimpse of capitalist thinking. Shortly after I arrived, the Challenger space shuttle exploded, and Jonathan, young and naive and enthusiastic, rushed off to call his Goldman stockbroker to buy puts on Morton Thiokol, hoping to profit from a drop in the price of the company that designed the leaky seals in the booster. Witnessing his quick response to the shuttle disaster, I thought Jonathan was destined for the business world, but I was wrong. Stimulated by financial theory, he left Goldman a few years later to get a PhD in finance and is now a professor at Berkeley. I met him again 15 years later, in November 2000, at a meeting of the sponsors of the new Berkeley degree in financial engineering, where I represented Goldman. Fervent as ever, he told me that in 1986 he had expected finance to become the theoretical physics of the twenty-first century. We laughed a little at the mismatch between fantasy and reality.