School of Computer Science
Computer Science Department
Graduate School of Industrial Administration
Carnegie Mellon University
Pittsburgh, PA 15213
Thesis Committee:
Katia Sycara, Chair
Andrew Moore
Bryan Routledge
Blake LeBaron
Submitted in partial ful¯llment of the requirements
for the degree of Doctor of Philosophy.
AI has long been applied to the problem of predicting ¯nancial mar-
kets. While AI researchers see ¯nancial forecasting as a fascinating chal-
lenge, predicting markets has powerful implications for ¯nancial economics
{ in particular the study of market e±ciency. Recently economists have
turned to AI for tools, using genetic algorithms to build trading strategies,
and exploring the returns those strategies generate of evidence of market
ine±ciency.
The primary aim of this thesis is to take this basic approach, and
put the arti¯cial intelligence techniques used on a ¯rm footing, in two
ways: ¯rst, by adapting AI techniques to the stunning amount of noise in
¯nancial data; second, by introducing a new source of data untapped by
traditional forecasting methods: news.
I start with practitioner-developed technical analysis constructs, sys-
tematically examining their ability to generate trading rules pro¯table on
a large universe of stocks. Then, I use these technical analysis constructs
as the underlying representation for a simple trading rule leaner, with
close attention paid to limiting search and representation to ¯ght over-
¯tting. In addition, I explore the use of ensemble methods to improve
performance. Finally, I introduce the use of textual data from internet
message boards and news stories, studying their use both in isolation as
well as augmenting numerical trading strategies.
发表于 2010-11-26 13:50
| 
