Information Technology Reference
In-Depth Information
An Improved Multi-Objective Genetic Model
for Stock Selection with Domain Knowledge
Shin-Shou Chen 1 , Chien-Feng Huang 2 , and Tzung-Pei Hong 1, 2
1 Department of Computer Science and Engineering,
National Sun Yat-Sen University, Kaohsiung, 804, Taiwan
2 Department of Computer Science and Information Engineering,
National University of Kaohsiung, Kaohsiung, 811, Taiwan
cronus4619@hotmail.com, {cfhuang15,tphong}@nuk.edu.tw
Abstract. In the past, we employed a multi-objective genetic algorithm
(MOGA) for optimization of model parameters and feature selection, and then
devised a stock scoring mechanism to rank and select stocks for forming a port-
folio. With each chromosome representing a feasible portfolio, that adopted
multi-objective genetic algorithm (MOGA) model thus decided good portfolios
by considering their return and risk. In this paper, we further improve upon the
MOGA model using financial knowledge to help selection of beneficial portfo-
lios. Especially, we refine the evaluation criteria with the assistance of relevant
domain knowledge from investment. Based on the promising results, we expect
this improved MOGA methodology to advance the current state of research in
soft computing for real-world stock selection applications.
Keywords: stock selection, genetic algorithms, feature selection, multi-
objective optimization.
1
Introduction
Stock selection has been an important and challenging task in the area of investment.
Researchers from the field of artificial intelligence have made several attempts to
tackle this task. One of the most known methods is genetic algorithm [6], which simu-
lates the process of biological evolution to solve optimization problems. In the early
works, the study of portfolio optimization typically centered around one single objec-
tive, i.e., the return of portfolios. In this context, the goal of stock selection may aims
at maximizing the expected return of individual stocks in order to obtain the maximal
actual return of portfolio. To improve the performance of the single-objective GA-
based models, multi-objective methods that consider two competitive goals of return
and risks simultaneously have been developed.
In this study, we extend our previous multi-objective GA-based method [2] for in-
creasing investment return and reducing the risk. In the previous approach, we used
the non-dominated sorting to find non-dominated solutions. In that treatment, the
portfolios with low risk but negative return might be generated, this might lead to
bankruptcy over the course of investment. In order to mitigate this problem, in this
 
Search WWH ::




Custom Search