A model of portfolio optimization using time adapting genetic network programming

https://doi.org/10.1016/j.cor.2009.12.003Get rights and content

Abstract

This paper describes a decision-making model of dynamic portfolio optimization for adapting to the change of stock prices based on an evolutionary computation method named genetic network programming (GNP). The proposed model, making use of the information from technical indices and candlestick chart, is trained to generate portfolio investment advice. Experimental results on the Japanese stock market show that the decision-making model using time adapting genetic network programming (TA-GNP) method outperforms other traditional models in terms of both accuracy and efficiency. A comprehensive analysis of the results is provided, and it is clarified that the TA-GNP method is effective on the portfolio optimization problem.

Introduction

This paper describes a decision-making model of stock portfolio optimization with the application of an evolutionary computation named genetic network programming (GNP) [1]. Using technical indices and candlestick chart as judgment functions, the computational intelligence system determines the distribution of initial capital to each brand in the portfolio, and also creates trading rules to buy and sell stocks on a regular basis. Rather than fixing these rules created by GNP throughout the period, the trading rules can adapt to the changes of stock prices, leading to a model of dynamic portfolio optimization that changes with time based on time adapting genetic network programming (TA-GNP).

The foundation of portfolio optimization was laid by Harry Markowitz in paper [2], as well as in his book published 7 years later [3]. He suggests that investors should decide the allocation of their investment on the basis of a trade-off between risk and expected returns based on mean-variance analysis. The mean-variance framework is so intuitive and strong that it has been widely applied to different areas within finance and risk management. In the case of linear constraints, Stein et al. [4] solve the problem efficiently by parametric quadratic programming. However, there are many real-world nonlinear constraints which limit the number of different assets in a portfolio. As a consequence, evolutionary computation methods were developed to calculate the optimal portfolio in the financial market. In this paper, how to allocate the given capital to a certain number of fixed brands is discussed, which is different from the classical mean-variance optimization models.

This paper contributes to the existing computing and finance literature in several ways. From a financial perspective, the use of learning model to build a stock portfolio based on GNP is not well developed and not comprehensively examined. Current emphasis is usually placed on developing buy and sell trading rules for individual stocks or indices, not a whole portfolio of stocks. This paper addresses the portfolio issue. To confirm the efficiency of the proposed method, comprehensive comparison and evaluation are done to give a detailed assessment of the model's performances.

The rest of this paper is organized as follows. Section 2 provides background information and a literature review. Section 3 explains the TA-GNP approach to be studied in this paper. In Section 4, we explain the portfolio optimization algorithm. Section 5 presents experimental environments, conditions and results using TA-GNP method. The trading profits are presented and compared with the traditional methods and Buy&Hold method. Finally, Section 6 concludes this paper.

Section snippets

Background

Over the last few decades, various approaches have been applied to financial modeling, especially for stock market activities. In this section, we survey some of this work.

Generally speaking, these approaches can be separated into two categories: statistical method and artificial intelligence (AI). The statistical methods are widely used to predict the stocks based on the past time series data. For example, Box and Jenkins [5] firstly proposed the autoregressive moving average model (ARMA) for

Time adapting genetic network programming

In this section, TA-GNP method is explained briefly. Basically, GNP is an extension of GP in terms of gene structures. The directed graph structure of GNP has some inherent characteristics such as compact structures and an implicit memory function that contribute to creating effective action rules.

Technical indices and candlestick chart

In our proposed portfolio model, technical indices and candlestick chart are used as judgment functions. Concretely speaking, each judgment node uses one of the following technical indices for its judgment: rate of deviation from moving average (ROD), relative strength index (RSI), rate of change (ROC), volume ratio (VR), rank correlation index (RCI), stochastics, golden/dead cross and moving average convergence and divergence (MACD).

As a judgment function, a candlestick chart is also used in

Simulations

In order to confirm the effectiveness of TA-GNP for the portfolio investment strategy, we carried out the trading simulations using 10 brands selected from the companies listed in the first section of Tokyo stock market in Japan. The simulation period is divided into two periods: one is used for training and the other is used for testing. We use the data of the stock market from January 4, 2001 to December 30, 2004 in the simulations. Moreover, we suppose that the initial funds, i.e., Initial(t)

Conclusions

This paper has provided a decision-making model of dynamic portfolio optimization based on TA-GNP, using the information of technical indices and candlestick chart patterns. Compared to conventional GNP, the proposed method has two advantages, first, TA-GNP has many control nodes, and each group of control nodes is assigned to each stock brand to create effective portfolio strategy. Second, TA-GNP can adapt to the change of stock prices with real-time updating data windows. Since TA-GNP can

References (38)

  • J. Potvin et al.

    Generating trading rules on the stock markets with genetic programming

    Computers & Operations Research

    (2004)
  • H. Etemadi et al.

    A genetic programming model for bankruptcy prediction: empirical evidence from Iran

    Expert Systems with Applications

    (2009)
  • S. Sette et al.

    Genetic programming: principles and applications

    Engineering Applications of Artificial Intelligence

    (2001)
  • A.S. Chen et al.

    Application of neural networks to an emerging financial market: forecasting and trading the Taiwan stock index

    Computers & Operations Research

    (2003)
  • P.C. Lin et al.

    Fuzzytree crossover for multi-valued stock valuation

    Information Sciences

    (2007)
  • S. Mabu et al.

    A graph-based evolutionary algorithm: Genetic network programming and its extension using reinforcement learning

    Evolutionary Computation

    (2007)
  • H. Markowitz

    Portfolio selection

    Journal of Finance

    (1952)
  • H. Markowitz

    Portfolio selection: Efficient diversification of investments

    (1959)
  • G.E.P. Box et al.

    Time series analysis: forecasting and control

    (1976)
  • Cited by (30)

    • A multi-objective evolutionary algorithm for a class of mean-variance portfolio selection problems

      2019, Expert Systems with Applications
      Citation Excerpt :

      Recently, Ruiz-Torrubiano and Suárez (2015) approached the multiperiod CCPSP with transaction costs and trading constraints. Chen, Mabu, and Hirasawa (2010) put forward a PSO for the PSP with fuzzy probabilities, whereas Noroozi, Mokhtari, Bagherpour, and Sadjadi (2016) introduced the application of a probabilistic model (CVX) to the classic version of the problem with random and uncertain return. In addition to the thorough review concerning deterministic models and applications for mean-variance portfolio optimization by Kalayci, Ertenlice, and Akbay (2019), the reader is also referred to the following surveys regarding specific type of methodologies proposed for PSPs: (i) MOEAs (Metaxiotis & Liagkouras, 2012; Ponsich et al., 2013); (ii) linear programming-based approaches (Mansini, Ogryczak, & Speranza, 2014); (iii) metaheuristics (Jarraya, 2013); and (iv) goal programming (Aouni, Colapinto, & Torre, 2014; Azmi & Tamiz, 2010).

    • A constrained portfolio selection model at considering risk-adjusted measure by using hybrid meta-heuristic algorithms

      2019, Applied Soft Computing Journal
      Citation Excerpt :

      Multi-method search Algorithm to portfolio selection introduced by Yuen et al. [121]. As there is a dynamic condition in the optimal stock selection, a new model was introduced that was named time adapting GNP (TA-GNP) [122]. In order to adapt to immediate and dynamic environmental changes, Online learning of GNP was investigated by Mabu et al. [123].

    • Stock portfolio selection using Dempster–Shafer evidence theory

      2018, Journal of King Saud University - Computer and Information Sciences
    • Application of evolutionary computation for rule discovery in stock algorithmic trading: A literature review

      2015, Applied Soft Computing Journal
      Citation Excerpt :

      However, by combining the information presented in Tables 1–3 and 5, we suggest that the phenomenon may be related to the input variable and the problem of liquidity in different market trends, which are discussed in Section 5. Periods for training and testing are also interesting factors that influence the return, both the long training period (about 10 years) with a short testing (1 year) period [32,101] and the short training period (180 days) with a shorter testing period (20 days) [28], are suggested by researchers. The periods in the remaining research works vary, but mainly in the middle in terms of training period (3–6 years) and with a short testing period (1 year).

    • MGP-INTACTSKY: Multitree Genetic Programming-based learning of INTerpretable and ACcurate TSK sYstems for dynamic portfolio trading

      2015, Applied Soft Computing Journal
      Citation Excerpt :

      Transaction cost and risk are the two factors that can affect the performance of the dynamic portfolio trading models significantly. However, they were not considered in [7,8]. The GNP model is extended to consider the risk of trading in terms of conditional value at risk [9,10].

    View all citing articles on Scopus

    This study is partially supported by JSPS and SUFE through Project 211 Phase III and Shanghai Leading Academic Discipline Project, Project Number: B803.

    View full text