Stock market trading rule discovery using two-layer bias decision tree
Introduction
Accurately forecasting stock prices has been extensively studied. According to the literature, stock market price movements are highly nonlinear and dynamic (Hiemstra & Jones, 1994). To solve the nonlinear problem and improve stock price evaluation, many researchers have focused on technical analysis and using advanced maths and science. Extensive attention has been devoted to the application of Artificial intelligence techniques like artificial neural networks (ANN), genetic algorithms and pattern recognition to this area (Brownstone, 1996, Gencay, 1998, Kamijo, 1990, Kim, 1998, Kim, 2000, Leigh, 2002, Refenes, 1997, Saad, 1998, Trippi, 1993, Tsaih and Y, 1998).
Owing to a trading rule system being a classification system that matches moments in time with trading recommendations (Leigh, Purvis, & Ragusa, 2002), research recently has developed ensemble combination methods and ensemble classifiers, generally, improve purchasing accuracy more than do their constituent classifiers alone (Bauer & Kohave, 1999). According to the literature, multiple classifier systems can be organized as ‘conditional’, ‘serial’, ‘hybrid’, or ‘parallel’ combinations of pattern recognition, neural network, and/or other classifier methods. For example, to improve ANN forecasting quality, most of these systems combined neural network with genetic algorithms or expert systems to solve the problems of local minimization and learning speed (Dourra, 2002, Kim, 2000). Leigh et al. (2002) provides an excellent survey of the multiple classifier methods and implications. Unlike previous methodologies, this study applies the concept of serial topology and designs a new decision system, namely the two-layer bias decision tree, for stock price prediction. Regarding the serial topology, classifiers are applied successively, with the output of one classifier used as the input to the next. Please refer to Kim and Noh (1997) for further details.
The methodology developed here differs from other studies in two respects. First, to reduce the classification error, the decision model was modified into a bias decision model. Second, a two-layer bias decision tree is used to improve purchasing accuracy. The empirical results indicate that the presented decision model not only produces excellent purchasing accuracy, but also significantly outperforms than random purchase.
The rest of the paper is organized as follows. Section 2 describes the methodologies used. Section 3 then describes the research design. Next, Section 4 summarizes and discusses the empirical results. Finally, Section 5 presents some conclusions.
Section snippets
Bias decision model
Suppose c possible pattern classes exist, ω1,ω2,…,ωc, and an arbitrary pattern belongs to class ωi with a priori probability, p(ωi), p(ωi)≥0, . Patterns are d-component measurement vectors or feature-vectors. The subject approach assumes pattern x to be a random vector taking value in d-dimensional feature space and governed by a multivariate probability density function p(x|ωi) when pattern x is known to belong to class ωi, i=1,…,c.
An arbitrary pattern x of unknown class ω normally
Research design
Technical analysis attempts to predict future stock price movements based on past price behavior. The concern in technical analysis is historical price movements and the forces of supply and demand affecting those prices (Edwards, 1966, Levy, 1966, Murphy, 1986). Technical analysis relies on chart/indictors and seeks particular configurations with predictive values.
Accordingly, this study uses daily stock price to assess stock market purchasing opportunities. The proposed two-layer bias
Empirical results
This study presents an improved approach for developing a new decision support system, namely two-layer bias decision tree, which can implement stock purchasing strategies.
Conclusion
Trading rule systems are classification systems that match moments in time with trading recommendations (Leigh, Purvis, & Ragusa, 2002). This study presents a proposal to use a two-layer bias decision tree with technical indicators to create a decision rule that makes buy or not-buy recommendations in the stock market. A novel method designed for using two-layer bias decision tree to improve purchasing accuracy. Comparison with random purchases, the results indicate the system presented here
References (18)
Using Percentage Accuracy to Measure Neural Network Predictions in Stock Market Movements
Neurocomputing
(1996)- et al.
Investment Using Technical Analysis and Fuzzy Logic, Fuzzy sets and systems
(2002) The Predictability of Security Returns With Simple Technical Trading Rules
Journal of Empirical Finance
(1998)- et al.
Predictability of Interest Rates Using Data Mining Tools: a Comparative Analysis of Korea and the US
Expert Systems with Applications
(1997) - et al.
Graded Forecasting Using an Array of Bipolar Predictions: Application of Probabilistic Neural Networks to a Stock Market Index, International Journal of Forecasting
(1998) - et al.
Genetic Algorithms Approach to Feature Discretization in Artificial Neural Networks for the Prediction of Stock Price Index, Expert Systems with Applications
(2000) - et al.
Forecasting S&P 500 Stock Index Futures With a Hybrid AI System, Decision Support Systems
(1998) - et al.
Kohave An Empirical Comparison of Voting Classification Algorithms: Bagging, Boosting, and Variants, Machine Learning
(1999) - Edwards, R., J. Magee (1966). Technical Analysis of Stock Trends (pp. 86-87), Boston: Magee John,...