The Problem with Algorithms and Money (and Life?)

By Kevin Farnham (108 posts) on February 9, 2008 at 1:25 pm

Day 2 of the Money:Tech Conference included an interview / conversation with Richard Bookstaber, author of the book "A Demon of our own Design". The book was included in the bag of goodies that was given out to conference attendees.

One thing I learned at Money:Tech is that the financial markets have become increasingly dominated by automation. Larry Tabb's excellent presentation on Wednesday, "Search, Dark Pools, and Disappearing Traders: A Financial Technology Roadmap", pointed out the incredible extent to which this has occurred, and also pointed out how much of the activity involved in modern trading strategy and trade execution happens in an almost "lights out" manner, with little or no observation of what's happening by humans. For example, no human can watch and monitor the system for problems when a trading sequence consists of hundreds of near-instantaneous bid and bid cancel sequences, where the bidding price itself has been specified by an algorithm.

At the conference, I also learned that something bad happened in August 2007, involving "quant funds," hedge funds that trade exclusively based on algorithms. Richard Bookstaber mentioned this in his session as yet another example of automated trading gone awry. The problem is that algorithms make assumptions, and these assumptions are correct almost all of the time. But what happens when reality ceases to match the assumptions the models are applying? Do they suddenly wake up and say "hey, something's wrong here"? Of course not, they're automatons.

One assumption most of the trading models make is that they assume it will be possible to actually execute the trades the model specifies. Stated another way, the assumption is that the market will always have adequate liquidity, there will always be a buyer if the model wants to sell, always a seller when the model wants to buy. Unfortunately, this is not a valid assumption!

Another assumption is that the activities of the model occur in isolation, the external marketplace is a great big ocean, and the model can just swim along, pushing and shoving in whatever direction it likes, without that having any significant and enduring effect on the other players in the market. Again, under normal conditions, this may seem to be the case, and it's apparently true often enough for the trading strategies to work. But in unusual circumstances, the actions of the model become coordinated with the actions of other models and non-algorithmic market participants, creating a kind of tidal wave where everyone is pushing in the same direction -- surely a recipe for disaster.

The August 21, 2007 Washington Post article "For Wall Street's Math Brains, Miscalculations" talked about the effect of assumptions modelers at Goldman Sachs had made. The assumptions are true 99.x percent of the time, but...

Last week, Goldman Sachs said its Global Alpha quant fund had lost 27 percent of its value this year because its computers failed to anticipate what the firm called "25 percent standard deviation moves" or events so rare Goldman had seen them only twice before in the firm's history.

Richard Bookstaber talked about the 1987 U.S. stock market crash, which I witnessed up close, since 1987 was the start of my own algorithm-based trading (note: I was out of the market, intentionally, when the crash happened). In that event, two different mechanized trading methods ("portfolio insurance" and "program trading") combined to reinforce each other's strategy of selling stock index futures and real stocks -- what both models said was the correct, safest thing to do under the circumstances -- leading to a 22.6% decline on October 19. Stated simply, here's what happened:

You can easily see the combined effect: a vicious circle of relentless automated selling pressure that produced what came to be known as "Black Monday." Yet, both strategies were doing exactly what they had been programmed to do, portfolio "insurance" striving to preserve past profits, and program trading seeing unusually superb opportunities for new profits.

Conclusion

The automation of the markets would seem to have increased the likelihood of financial Black Swan events, that is, rare, hard-to-predict, high-impact events. The coupling of automation in both trading strategies and trade execution means that when the assumptions of models turn out not to be true (which rarely happens), the models can turn a somewhat random anomaly into a high-impact event, turning what might have been a unusual ripple (if humans were doing the trading) into a tsunami.

The Money:Tech conference showed how integrated automation has come to be in the global world of finance. But computers and software are certainly becoming increasingly prevalent and integrated into our daily lives. Multicore computers powered by multithreading technologies such as Threading Building Blocks can only be expected to increase this trend, making the integration of hardware and software all the more pervasive in daily human life.

This is a good thing, certainly. Life gets better. But, as Richard Bookstaber says, automation is always based on assumptions, and there will always be possibilities that aren't programmed into the automation mechanism. What happens then?

Kevin Farnham, O'Reilly Media, TBB Open Source Community, Freenode IRC #tbb, TBB Mailing Lists

Download TBB

Categories: Multicore, Open Source, Threading Building Blocks

Comments (3) Comments RSS Feed

By Ray Zed Blog on February 9th, 2008 at 9:35 pm
links from Technoratiunknown: Day 2 of the Money:Tech Conference included an interview / conversation with Richard Bookstaber, author of the book “A Demon of our own Design”. The book was included in the bag of goodies that was given out to conference attendees.

By Stocks Technical Analysis Options Trading Strategies, Automated Stock Trading Online on February 10th, 2008 at 9:37 pm
links from TechnoratiLIVE Trading Room Recap Jumpstarting Your Business With Free Stock Trade Old Dominion Freight Chugs Along Indicators And Oscillators Are Wonderful If You Use Them ProperlyThe Problem with Algorithms and Money (and Life?)

By Блоги Intel® Software Network » Top 5 ISN Blogs в феврале on March 7th, 2008 at 4:45 am
[...] популярным в феврале был пост Кевина Фарнама (Kevin Farnham) "Проблемы алгоритмов и денег". Кевин приостановил свои попытки интегрировать TBB в [...]


What do you think?

Name (required)

Email (required; will not be displayed on this page)

Your URL (optional)

Comments (required)