Preface
Introduction
Computational intelligence techniques, including machine learning and data mining,
have significantly reshaped the financial investment community over recent decades.
Examples include high-frequency trading and algorithmic trading. This book studies a
fundamental problem in computational finance, or online portfolio selection (OLPS),
which aims to sequentially determine optimal allocations across a set of assets. This
book investigates this problem by conducting a comprehensive survey on existing
principles and presenting a family of new strategies using machine-learning tech-
niques. A back-test system using historical data has been developed to evaluate the
performance of trading strategies.
Our goal in writing this monograph is to present a self-contained text to a wide
range of audiences, including graduate students in finance, computer science, and
statistics, as well as researchers and engineers who are interested in computational
investment. The readers are encouraged to visit our project website for more updates:
http://olps.stevenhoi.org
Organization
Part I introduces the OLPS problem. Chapter 1 introduces the background and sum-
marizes the contributions of this book. Chapter 2 formally formulates OLPS as a
sequential decision task.
Part II presents some key principles for this task. Chapter 3 summarizes three
benchmarks: the Buy-and-Hold strategy, Best Stock strategy, and Constant Rebal-
anced Portfolios. Chapter 4 presents the principle of Follow the Winner, which moves
weights from winning assets to losing assets. Chapter 5 presents an opposite princi-
ple called Follow the Loser, which moves weights from losers to winners. Chapter 6
demonstrates the principle of Pattern Matching, which exploits similar patterns among
historical markets. Chapter 7 talks about Meta-Learning, which views the strategies
as assets, and thus hyperstrategies.
Part III designs four novel algorithms to solve the OLPS problem. All algo-
rithms apply the state-of-the-art machine-learning techniques to the task. Chapter 8
designs a new strategy named CORrelation-driven Nonparametric (CORN) learn-
ing, which overcomes the limitations of existing pattern matching–based strategies
using Euclidean distance to measure the similarity between two patterns. Chapter 9
develops Passive–Aggressive Mean Reversion (PAMR), which is based on the
first-order passive–aggressive online learning method, and Chapter 10 designs
xv
T&F Cat #K23731 — K23731_C000 — page xv — 10/13/2015 — 16:23