Nerds on Wall Street

Artificial Intelligence and Intelligence Amplification in Financial Markets

Securities Markets are Machinery Now.

This raises the question of how to best participate in the world’s new wired markets. People who use information technology most effectively will be rewarded.

Artificial intelligence (AI) as an academic discipline began at the famous 1955 Dartmouth conference organized by John McCarthy from Stanford University and Marvin Minsky from MIT. The goal of the AI pioneers was to create a mind, a human in silicon. One key idea was that the brain was a biological computer so all the researchers had to do was figure out what the brain was doing and put it into an actual computer and they’d be done. This was something that people thought in the 1950s might take 10 years to accomplish, maybe 15 with long lunches.

So far, it hasn’t exactly worked out. In fiction we have the example of HAL, from 2001: A Space Odyssey. Letting the computer do the thinking turned out badly in that case. HAL discovered the lie in the first reel, and quickly moved on to become a paranoid serial killer.(1)

In financial circles there was a lot of sort of irrational technology exuberance as well. As seen in Chapter 2, there were some inspired magazine covers from the magazine Wall Street Computer Review(2) that showed the unrealistic expectations for AI. One, from 1987, depicts Socrates on the steps of the stock exchange surrounded by a horde of PCs, and touts: “Teaching Computers to Emulate Great Thinkers.”

Source: Wall Street Computer Review (now Wall Street & Technology), June 1987.

Computers “thinking” like a person didn’t really work out as well as people had hoped. In the academic AI world, the artificial sentients were persistent no-shows. Artificial intelligence got kind of a bad rep on Wall Street as well. To get into the game, you had to buy a LISP (list processing) machine, a $100,000 machine that ran the elegant but arcane AI language of choice, LISP, and related expert system tools with names like Automated Reasoning Tool (ART) and Knowledge Engineering Environment (KEE).

You could prove a theorem or two with this stuff right out of the box, but getting a price for IBM on the screen proved to be a prodigious amount of work.

As recounted in the Introduction to this book, I worked for one of the companies that made these machines—there were two, LISP Machines and Symbolics—and came to the realization that building from the high-concept top down was a poor idea. Here is a picture I used back then to show the user reaction to AI investment systems, showing a mix of fear and rage.

Chapter 7 – A Little AI Goes a Long Way on Wall Street, describes a successful application, built in the early 1990’s, which was much more modest in scope than bringing back the great thinkers. The goal was to get computers to be decent users of other computer systems, which were overloading their users. Recently, this idea has been called intelligence amplification (IA). The systems here were market data and execution systems, and the combination was an early version of algorithmic trading.

The AI/IA-flavored approach to algo trading and market surveillance used in QuantEx and MarketMind, described in Chapter 7, was very effective in many contexts—equity strategies, quant option trading, and monitoring liquidity— but, like spreadsheets, they were clever ways to get the computer to do what you said, and didn’t have anything in the box that would let the machine learn what you should have said and adjust its behavior or models to better meet your needs.

One of the early enthusiasts for these technologies was Henry Lichstein, an MIT graduate then serving as technology adviser to John Reed, chairman of Citicorp. Henry was also on the board of the Santa Fe Institute (SFI),(3) a new high-powered research institute interested in complex interactions of complex entities. SFIdid a great deal of computer experimentation in artificial life (ALife).

The ALife people had much more modest goals than AI; all they wanted to do was build software entities that displayed lifelike behavior. And their efforts were met with early success. A bunch of random ALife birds might fl y in random directions, and not act much like birds. But giving them a few simple rules, like “fl y toward the closest bird,” “go with the flow,” and “don’t hit other birds,” could give rise to distinctly birdlike behavior.

The original ALife flock is Craig Reynolds’ “Boids,” done at Symbolics in 1986.(4) Simulated herding and fl ocking turn out to be of some commercial interest. Those massive stampedes in Disney cartoons, with thousands of , and all those schools of talking fish are descendants of the Boids. Hollywood showed its appreciation by giving Reynolds an Academy Award in 1998.

Like our early algos, the SFIartificial life was dumb, just obeying a few simple handwritten rules, without any ability to learn from mistakes. But the ALife researchers had a good answer—evolving intelligent behavior by mimicking natural evolution. Represent the programs as digital chromosomes, and simulate crossover and mutation, to breed better programs. The original version was called the genetic algorithm (GA), which later evolved into evolutionary computation (EC). Henry Lichstein suggested this approach for use in trading around 1990. The results for ALife had been striking. All sorts of tasks could be learned by the programs operating in an electronic world, which sure sounded like quant investment and trading. ALife programs evolved behavior to solve problems, avoid obstacles, find rewards, and cooperate. These are remarkable to watch.(5)

Chapter 8 – Perils and Promise of Evolutionary Computation on Wall Street, is one of the few detailed descriptions of the use of the GA and EC in finance. Actual chromosomes and fitness functions are included. If that last sentence made your eyes glaze over, skip to Chapter 9. It was satisfying research, and the power of evolutionary search truly teaches you to be careful what you ask for. Evolutionary approaches could be given rules to avoid many sins of data mining, and had many appealing features. As much as I would like to see this work, we don’t see a torrent of EC scientists fl owing toward greater Wall Street. We don’t even see a trickle. Olsen & Associates was a Zurich-based currency trading firm, founded in 1992, heavy with physicists from the Eidgenössische Technische Hochschule (ETH), Einstein’s alma mater. They were keen on genetically adaptive strategies and well funded, but vanished, and few of the principals are still keen on genetic algorithms.

After sending the GA to the back of the breakthrough line in the previous chapter, in Chapter 9 – The Text Frontier, we get to using IA, natural language processing, and Web technologies to extract and make sense of qualitative written information from news and a variety of disintermediated sources.

In Chapter 6 – Stupid Data Miner Tricks, we saw how you could fool yourself with data. When you collect data that people have put on the Web, they can try to fool you as well.

Chapter 10 – Collective Intelligence, Social Media, and Web Market Monitors and Chapter 11 – Three Hundred Years of Stock Market Manipulations, include some remarkable and egregious examples.

>>>>>> READ MORE HERE < <<<<<<

Notes for this section about Financial Artificial Intelligence, Intelligence Amplification, Genetic Algorithms, and Evolutionary Computation in Securities Markets:

1) Sci-fi buffs have a rich amount of material to consider in this context. Vernor Vinge, a computer scientist who has also won five Hugo awards, deals with the topic in much of his work, including his latest novel, Rainbow’s End. If we construct an artificial super-intelligent entity, what will it think of us?

2) Source: Wall Street Computer Review, June 1987. Since renamed Wall Street & Technology, and a useful resource for nerds on Wall Street at www.wallstreetandtech.com.

3) “The Santa Fe Institute is devoted to creating a new kind of scientific research community, one emphasizing multidisciplinary collaboration in pursuit of understanding the common themes that arise in natural, artificial, and social systems.” It was founded in 1984 (www.santafe.edu/).

4) … and still flocking after all these years at www.red3d.com/cwr/boids.

5) Until they start barcoding these into books, printed URLs are annoying, but it is actually worth typing these in. Karl Sim’s MIT video is here: www.youtube.com/watch?v=F0OHycypSG8.

Passive Investing – Active Investing – Alpha Returns

Index funds are passive investments; their goal is to deliver a return that matches a benchmark index. The Old Testament of indexing is Burton Malkiel’s classic A Random Walk Down Wall Street, first published in 1973 by W.W. Norton and now in its ninth edition. For typical individual investors, without special access to information, it offers what is likely the best financial advice they will ever get: It is hard to consistently beat the market, especially after fees. A passive strategy will do better in the long run.

Of course, no one thinks of oneself as a typical individual investor. That might be your brother-in-law or the guy across the hall. And index funds are just not as much fun as picking stocks. It’s called passive investing for a reason. Alpha, outperforming a passive benchmark, is the goal of active investing. Even Malkiel has admitted to actively managing some his own money.(*) Recent additions to the Forbes 400 list include more than a few people who seem unusually adept at finding alpha, and keeping a piece of it.

The basic fee structure in the hedge fund world is “2 and 20.” Managers are paid 2 percent of assets and 20 percent of alpha. Similar arrangements are also used for performance paid to institutional managers, blurring the distinction between these types of buy-side firms. To see how this works, consider a $100 million portfolio, benchmarked against Treasury bills. If the manager produced a return equal to the T-bills, the alpha would be zero, and the manager’s fee would be $2 million, all from the asset-based portion. Unless the firm gave really good parties or had a great story, it would probably be replaced, since the client would end up earning the T-bill rate minus 2 percent, or something like a passbook savings account.

With a skilled, lucky, or skilled and lucky manager, the situation could be quite different. If the T-bills returned 3 percent that year and the hedge portfolio returned 28 percent, then the manager’s alpha is 25 percent, $25 million on the original investment. Under the 2-and-20 plan the firm would get to keep 20 percent of that, another $5 million on top of the $2 million in asset-based fees. The client keeps $18 million, substantially more than the meager few percent the client would have gotten in Treasuries.

A $100 million portfolio is small as hedge funds go.

It costs money to do the research or proprietary trading to produce that 25 percent alpha, so by the time all the bills are paid, that $7 million the manager takes is seriously pared down. But when the fund gets larger, the economies of scale kick-in in a major way. Investment strategies don’t scale to the sky, but it is (approximately) true that the cost to run a $1 billion portfolio is not that much more than for $100 million. In that case, the manager on the 2-and-20 plan takes home $70 million with performance as in the example. On $10 billion, the manager takes home $700 million, which begins to look like serious coin—even on the right side of the tracks in Greenwich, Connecticut. Deliver this kind of performance consistently, and you can raise the rates to 4 percent of assets and 40 percent of alpha, which would pay the $10 billion manager $1.4 billion with the same performance scenario.

This is where those billion-dollar paydays for hedge fund managers we read about in Institutional Investor and Parade magazine come from, and why people with what seem like good, solid $5 million annual paychecks at places like Goldman Sachs leave to start their own hedge funds. The whole alpha ecosystem depends on, and is a creature of, technology. Before computers, it was sufficiently tedious to compute the alpha of a portfolio that no one did it.

Comparing one stock to another is easy. Real portfolios are much messier. They have cash flows in from additional investments, and cash flows out from payments or withdrawals. There are dividends paid in from long positions, and dividends paid out from shorts. Stocks split, companies merge, symbols change. International investments’ returns are subject to currency variations to the extent that they are not hedged, and if they are, there are costs associated with those hedge positions.

Bill Fouse, who started the world’s first index fund, tells a story about the early days of performance measurement. In the 1950s and 1960s the reporting from investment managers to clients was almost anecdotal. The manager would invite the clients up to the lavishly decorated dark wood-paneled office and show them a list of stocks in their portfolio, with the prices paid and the recent prices.

Nothing would be said about cash flows, holding periods, or dividends, and nothing about closed positions. It was easy to pretty up the report by cleaning out the losers. Everyone would sit around the conference table to review the list of holdings, and enjoy a fine n-martini lunch.

In 1968 A.G. Becker, a brokerage firm, changed the game by using computers to keep accurate annualized scores for clients’ accounts, and by comparing the results with index benchmarks. This was possible only because the firm had acquired one of the early mainframe computers, a room-filling behemoth like the IBM System 360. The news wasn’t pretty. Any asset managers were much better at telling a good story and coming up with a good lunch than they were at managing assets. As Fouse tells it, managers resisted the idea of quantitative performance measurement.

They sent out the word, “Hire them, and you can’t hire us.” Some of their objections were valid; a simple performance measurement doesn’t consider the risks that a manager is allowed to take. Other measures—like the Sharpe, Jensen, and Treynor ratios(**) — refined the idea, but the alpha industry was born and has been growing ever since.

Finance students and Wall Street sorts around the world yearn for knowledge that will let them find ever more alpha. This raises the simple question of Chapter 4 – Where does alpha come from? That question opens this part of the book. The chapter explains why the search for alpha is more than just a snipe hunt, and why the people who find it may be more than just plain lucky.

Chapter 5 – A Gentle Introduction to Computerized Investing, starts out with a description of indexing, the great granddaddy of all quant equity strategies, and how it is transformed into active quant strategies by adding information beyond knowledge of an index’s constituent stocks.

In the last of this part, Chapter 6 – Stupid Data Miner Tricks, we see how with the right mix of hubris, stupidity, and CPU cycles, it is possible to do some real damage to your financial health. In investing, as in the bomb squad, knowing what not to do is extremely worthwhile.

>>>>>> READ MORE HERE < <<<<<<

* This surprising admission came in a dinner speech at the Investment Management Network “Superbowl of Indexing” Conference (December 1996, Palm Springs, California). No performance figures were disclosed.

** The Sharpe ratio is a measure of management skill that adjusts pure alpha (value added) by the variability of that value added. The others (Jensen & Treynor) are refinements based on characteristics of the portfolio, such as beta. They are less commonly used. Details are here http://en.wikipedia.org/wiki/Sharpe_ratio.

Financial Markets – Electronic Markets

Not too long ago, going to a stock market meant you would meet lots of new people who were energetically shouting, running around, and making a mess with great quantities of paper. No more. Visiting a financial market now is more like visiting a telephone exchange. Computers and network gear hum in racks. Fans blow. Rows of tiny lights flicker. Occasionally someone shows up to replace a disk.

Technology did not suddenly transform our markets. It has been a gradual process, and understanding how we got here, and the simpler machines we used along the way, provides insight into today’s complex markets. In that spirit, the first chapter in this part, an illustrated history of market technology, gives an informative perspective on today’s wired markets.

Computers make a dramatic entrance into financial markets at the conclusion of Chapter 1 – An Illustrated History of Wired Markets.

“So how did that work out?” you might ask. Chapter 2 – Greatest Hits of Computation in Finance answers that question, surveying some of the greatest technological hits influencing the markets.

Electronic markets are at the top of our greatest hits list. They are about the mechanics of trading, that is, the implementation of investment decisions (in contrast to actually making those decisions). Chapter 3 – Algorithm Wars is a more in-depth view of one of the most dynamic areas in electronic markets.

>>>>>> READ MORE HERE < <<<<<<

Nerds on Wall Street Forward by Ted Aronson

Quantitative finance is not a topic usually associated with laughter. That is about to change with the publication of Nerds on Wall Street.

I was first exposed to Dave Leinweber’s wit when he delivered a speech entitled “Nerds on Wall Street.” I believe the event happened 20 or 25 years ago at a CFA Institute conclave. He was a dude obviously knowledgeable about the investment business, with impressive credentials (MIT, Harvard, RAND), alluding to technical aspects of numerical finance and getting into the minutiae of electronic trading. Certainly on Wall Street such qualifications aren’t unique. But, oh, how he delivered his message! His rapid-fire sense of humor was worthy of Henny Youngman or Shecky Greene.

Dave’s speech was augmented by equally hilarious visuals. (Unfortunately, some of Dave’s props are so rare they are no longer available. Rats!) On a particularly memorable occasion (one of ITG’s famous conferences), Dave did his shtick with a drummer punctuating his one-liners with rimshots. I kid you not! The crowd, loosened up by cocktails, was reduced to tears from laughter.

Now, decades later, I’ve heard Dave deliver numerous speeches and presentations with various titles. But they are always roughly the same subject — yup, you guessed it — “Nerds on Wall Street.” So sit back and be prepared to be educated by a master. The education will come with images, illustrations, and humor you will not soon forget. It will be love . . . at first sound bite!

Theodore R. Aronson
Managing Partner, Aronson_Johnson_Ortiz
Past Chairman, CFA Institute

>>>>>> READ MORE HERE < <<<<<<