ACM Talk Friday, Jan 12th: Evolutionary Software Design for the O ptions Market: Towards Fully Autonomous Trading Agents

[Benjamin Blair]

Speaker: Ben Blair (BS '00, Senior Programmer PEAK6 Capital Management)
Time: Friday, Jan. 12th at 3PM
Location: Ryerson Room 276
Topic: Evolutionary Software Design for the Options Market: Towards Fully
Autonomous Trading Agents

An option is a contract that provides the holder of the option with the
right to buy or sell an underlying commodity at a specified price on or
before a specified date. Options, then, have a fairly well defined price,
based on the current price of the underlyer, the time until the option
expires, and the volatility in the price of the underlyer.  Compare this to
the price of a stock, which can be based on factors as indeterminate as the
perceived mood of the Fed or the current weather. Since nearly all the
information required to determine the price of an option is well defined and
readily available, it is possible to design software systems that can watch
the market and make decisions about which options are mis-priced and
therefore should be traded. Present inefficiencies in the market allow the
development of software that is significantly less than perfect at valuing
options, yet, with some human oversight, is extremely successful at finding
lucrative trades. These revenues can then be re-invested to further improve
the software. As such software systems are deployed, their actions change
the market, in most cases making the market more efficient and forcing the
software systems to be further improved. This feedback will become ever more
pronounced as competing software systems begin to react to each other's
actions in the market. We believe that intelligent agents that are not
designed through interaction and co-evolution with a live data environment
such as the options market are not likely to succeed. We currently have
autonomous hedging systems that average around $20MM in trades each day.
There are a number of significant challenges that must be overcome before
software can be allowed to trade completely autonomously, but we are
confident we will be one of the first to achieve this goal. We believe that
with our current approach, we can have fully autonomous trading agents
trading on electronic exchanges in 2 - 5 years. The techniques developed
during this process should be widely applicable to other intelligent agent
applications.