Elsevier

Information Sciences

Volume 170, Issue 1, 18 February 2005, Pages 3-33
Information Sciences

A hybrid genetic-neural architecture for stock indexes forecasting

https://doi.org/10.1016/j.ins.2003.03.023Get rights and content

Abstract

In this paper, a new approach for time series forecasting is presented. The forecasting activity results from the interaction of a population of experts, each integrating genetic and neural technologies. An expert of this kind embodies a genetic classifier designed to control the activation of a feedforward artificial neural network for performing a locally scoped forecasting activity. Genetic and neural components are supplied with different information: The former deal with inputs encoding information retrieved from technical analysis, whereas the latter process other relevant inputs, in particular past stock prices. To investigate the performance of the proposed approach in response to real data, a stock market forecasting system has been implemented and tested on two stock market indexes, allowing for account realistic trading commissions. The results pointed to the good forecasting capability of the approach, which repeatedly outperformed the “Buy and Hold” strategy.

Introduction

It is widely acknowledged that financial time series modeling and forecasting is an arduous task. These time series behave very much like a random walk process and several studies have concluded that their serial correlation is economically and statistically insignificant [11]. The same studies seem to confirm the efficient market hypothesis (EMH) [8], which maintains that the current market price of a stock fully reflects––at any time––the available information assimilated by traders. As new information enters the system, the imbalance is immediately detected and promptly redressed by a counteracting change in market price. Depending on the type of information examined, three forms of EMH exist: weak, semi-strong, and strong. We are particularly concerned with the weak EMH, which only takes into account past stock price data. In this case, the underlying assumption is that no predictions can be made based on stock price data alone, as they follow a random walk in which successive changes have zero correlation. This hypothesis implies that future changes in stock market prices cannot be predicted from information about past prices. Notwithstanding these difficulties, most stock market investors seem convinced that they can statistically predict price trends and make a profit. This is done by exploiting technical or fundamental analysis rules, as well as “momentum strategies” (i.e., buying when the market is bullish and selling when it is bearish). For these reasons, many attempts have been made to model and forecast financial markets, using all the computational tools available for studying time series and complex systems: linear auto-regressive models, principal component analysis, artificial neural networks (ANNs), genetic algorithms (GAs), and others (for an interesting review see [12]). In this paper, we present a hybrid approach to stock market forecasting that integrates both GAs [13] and ANNs [29] and cooperatively exploits them to forecast the next-day price of stock market indexes. In particular, we use an extended classifier system (XCS) [36], which relies on technical-analysis indicators (see, for example, [1]) to determine the current market status, in conjunction with feedforward ANNs explicitly designed for financial time series prediction. To our knowledge, no previous work has been done on hybrid systems that integrate XCSs with feedforward ANNs. We have called the proposed approach NXCS, standing for neural XCS, and customized it for financial time series prediction. A forecasting system based on an NXCS module has been tested on financial time series showing the trend of some major stock market indexes on a fairly large observational window. In particular, about 9 years of data of the COMIT1 and S&P500 stock market indexes have been used to train and test the system, being very careful to avoid any form of data snooping. In both cases, the first 1000 data were used to train and tune the NXCS module, and the resulting system was tested on the subsequent 1000 data, leaving its overall configuration unchanged. We compared the system's forecasting capabilities with the “Buy and Hold” (B&H) strategy, considering realistic transaction costs. Further comparisons with a system based on neural networks technology have also been performed. The results are encouraging, demonstrating the validity of the approach. The remainder of the paper is organized as follows: in Section 2 a short introduction to the state-of-the-art in financial time series forecasting with artificial intelligence (AI) techniques is given; in Section 3 the novel approach is described, first from a general perspective and then with all the customizations designed to handle the problem of financial time series forecasting; in Section 4, after briefly outlining the overall architecture of the proposed system, experimental results are discussed; and lastly the conclusions are drawn in Section 5.

Section snippets

AI techniques for financial time series forecasting

In recent years, advances in both analytical and computational methods have led to a number of interesting new approaches to financial time series forecasting, based on non-linear and non-stationary models. In the following, we focus the review of previous work on GAs and ANNs applied to stock market prediction, as our proposal is based on the integration of such techniques. In addition, as the resulting framework is also an implementation of the general concepts known as mixture of experts,

A hybrid approach for dealing with stock market forecasting

In this section, the hybrid approach previously summarized is described with more detail, from both a conceptual and a technical perspective. As for conceptual issues, a novel kind of model identification is introduced, originated by the need of dealing with multistationary processes. In addition, the idea of partitioning the input space starting from suitable technical-analysis domain knowledge is illustrated and framed in a multiple-experts perspective. To this end, the underlying framework

Experimental results

Assessing the performance of a forecasting system is not an easy task for a number of reasons. First, as data are non-stationary, the significance of the results obtained in a test period is not easy to quantify. Furthermore, variables traditionally handled by learning algorithms, such as mean square error or percent of correct classifications, do not have a direct economic relevance. Finally, stock markets have a number of constraints and costs that cannot be overlooked. To test our approach,

Conclusions and future work

In this paper, a novel approach to perform stock market forecasting has been presented and described from the conceptual and technical perspectives. Conceptually, predictions are obtained by following a divide-and-conquer strategy, based on the idea that partitioning the input space according to a significant amount of domain knowledge would facilitate the prediction task. To let the reader better understand the motivations of this choice, the non-stationary characteristics of financial time

References (39)

  • T. Bollerslev

    Generalized autoregressive conditional heteroskedaticicty

    Journal of Econometrics

    (1986)
  • G. Hawawini et al.

    On the predictability of common stock returns: world-wide evidence

  • J.H. Holland

    Adaptation

  • P.J.B. Hancock

    Pruning neural nets by genetic algorithm

  • S.B. Achelis

    Technical Analysis from A to Z

    (1995)
  • J.Y. Campbell et al.

    The Econometrics of Financial Markets

    (1997)
  • P. Campolucci et al.

    On-line learning algorithms for locally recurrent neural networks

    IEEE Transactions on Neural Networks

    (1999)
  • F. Castiglione

    Forecasting price increments using an artificial neural network

    Advances in Complex Systems

    (2001)
  • R.E. Dorsey et al.

    The use of parsimonious neural networks for forecasting financial time series

    Journal of Computational Intelligence in Finance

    (1998)
  • S.E. Fahlmann, C. Lebiere, The cascade-correlation learning architecture, Technical Report CMU-CS-90-100, Carnegie...
  • E.F. Fama

    The behavior of stock market prices

    The Journal of Business

    (1965)
  • C.L. Giles, S. Lawrence, A. Chung Tsoi, Rule inference for financial prediction using recurrent neural networks, in:...
  • D.E. Goldberg

    Genetic Algorithms in Search, Optimization and Machine Learning

    (1989)
  • T. Hellström, K. Holmström, Predicting the stock market, Technical Report IMa-TOM-1997-07, Department of Mathematics...
  • R.A. Jacobs et al.

    Adaptive mixtures of local experts

    Neural Computation

    (1991)
  • F. Kai, X, Wenhua, Training neural network with genetic algorithms for forecasting the stock price index, in:...
  • T. Kovacs, Evolving optimal populations with XCS classifier systems, MSc. Dissertation, University of Birmingham, UK,...
  • T. Kovacs, Strength or accuracy? A comparison of two approaches to fitness, in: A.S. Wu (Ed.), Second International...
  • P.L. Lanzi

    A study of the generalization capabilities of XCS

  • Cited by (0)

    View full text