With higher average growth rates and a significant consumer base that is fueling economic expansion, the economic successes of several Asian economies and their increasingly important roles in the global financial market have motivated several studies on Asian stock markets (Ané et al., 2008, Kim and Shamsuddin, 2008 and Mazouz et al., 2009). While Japanese and China’s stock markets are the biggest two in Asia, they respectively represent the two types of stock markets. Japan’s stock market represents the advanced financial market and has already attracted many researches (Atsalakis and Valavanis, 2009, Huang et al., 2005, Lee and Chen, 2002, Lu, 2010 and Lu et al., 2009). On the contrary, China’s stock market is a typically emerging market. Few studies are reported on the prediction of China’s stock market (Atsalakis and Valavanis, 2009 and Cao et al., 2005). But as one of the so-called BRICs, China’s stock market is now the second largest in Asia following only Japan, and it is speculated with the potential to rank among the top four or five in the world within the coming decade. In 27th February 2007 which is called “Black Tuesday”, with China’s stock slumping near to 9%, Global stock marked suffered a sharp fall. It shows that China’s stock market begin to play an important role for the world economy. So, China’s stock market has recently attracted some interest in literatures. These includes the cross section of expected stock returns (Wang & Iorio, 2007), market efficiency (Fifield & Jetty, 2008), turning points of stock price (Yan, Powell, Shi, & Xu, 2007), and stock price forecasting (Cao et al., 2005).
A large number of successful applications have shown that neural networks can be a useful technique for stock price/index prediction since they do not require strong model assumptions and can map any nonlinear function without a priori assumption about the properties of the data (Atsalakis and Valavanis, 2009, Vellido et al., 1999 and Zhang et al., 1998). In the most existing prediction methods using neural networks, the observed original values of prediction variables are usually directly used for building prediction models (Vellido et al., 1999 and Zhang et al., 1998). However, a stock price may be affected by some underlying factors like seasonal variations or economic events (Back and Weigend, 1997, Górriz et al., 2003, Kiviluoto and Oja, 1998, Malaroiu et al., 2000 and Mok et al., 2004). For example, in Mok et al. (2004), the observed daily stock returns reflected the stock market reaction to a few factors such as release of various economic indicators, company news, psychological factors, government intervention, or political issues. Kiviluoto and Oja (1998) showed that the underlying factors affected the cash flow time series of several stores belonging to the same retail chain were trends, holiday effects, seasonal variation and competition. Back and Weigend (1997) used ICA to exact the features of the daily returns of the 28 largest Japanese stocks. The results showed that the dominant ICs can reveal more underlying structure and information of the stock prices than principal component analysis. Therefore, if the underlying/interesting information that can’t be observed directly from the observed original data can be revealed through transforming the input space into the feature space with suitable feature extraction methods, the prediction performance will be improved by using the features to build prediction model.
Independent component analysis (ICA) is a novel feature extraction technique. It aims at recovering independent sources from their mixtures, without knowing the mixing procedure or any specific knowledge of the sources (Hyvärinen, Karhunen, & Oja, 2001). The independent sources, called independent components (ICs), are hidden information of the observable data. ICA model can be classified as linear ICA (LICA) and nonlinear ICA (NLICA) models. Linear ICA model is mainly based on the assumption that the observed mixture data are generated by a linear mixing from ICs (Hyvärinen et al., 2001). Conversely, the nonlinear ICA model assumes that the observations are a nonlinear combination of ICs (Almeida, 2003, Hyvärinen and Pajunen, 1999 and Valpola, 2000).
Linear ICA has been employed successfully in various fields, such as signal processing, face recognition and time series forecasting (Górriz et al., 2003, Hyvärinen and Oja, 2000, Lu, 2010, Lu et al., 2009, Malaroiu et al., 2000 and Popescu et al., 2003). Oja, Kiviluoto, and Malaroiu (2000) applied LICA in foreign exchange rate time series prediction. They firstly used LICA to estimate independent components and mixing matrix from the observed time series data and filtered the independent components (ICs) to reduce the effects of noise through the linear and nonlinear smoothing techniques. Then, the autoregression (AR) modeling was employed to predict the smoothed independent components. Finally, they combined the predictions of each smoothed IC by using mixing matrix and thus obtained the predications for the original observed time series. Cao and Chong (2002) employed LICA as a feature extraction tool in developing a support vector machine (SVM) forecaster. The ICs were considered as features of the forecasting data and used to build the SVM forecasting model. Lu et al. (2009) proposed a two-stage forecasting model by integrating LICA and support vector regression (SVR) for financial time series. They first used LICA to the forecasting variables for generating the ICs. After identifying and removing the ICs containing the noise, the rest of the ICs were then used to reconstruct the forecasting variables which contain less noise and served as the input variables of the SVR forecasting model. Lu (2010) utilized LICA to generate a denoising scheme for stock price data. The scheme was then used as preprocessing tool before building a stock price prediction model.
Linear ICA is suitable when it is reasonable to assume that the mixing mechanism is linear or the observations are linear mixture data. However, in many realistic cases/problems, the observed data (for example, stock index data) might be a nonlinear mixture of latent source signals. Asian stock markets are highly dynamic and exhibit wide variation since a lot of global investors are attracted to Asian stock markets. This makes Asian stock market indexes can be realistically and practically assumed to be the nonlinear mixture data. Therefore, NLICA has the potential to become rather powerful tools in the prediction of Asian stock market indexes. Moreover, unlike LICA model, NLICA remains an active field of research. Few applications of NLICA are proposed in literatures, such as image denoising (Haritopoulos, Yin, & Allinson, 2002) and feature extraction (Almeida, 2003 and Zhang and Chan, 2007).
In this study, a stock index prediction model by integrating NLICA and neural network is proposed for two major Asian stock markets-China and Japan. The proposed approach first uses NLICA on the input space composed of original forecasting variables into the feature space consisting of independent components representing underlying/hidden information of the original data. The hidden information of the original data could be discovered in these ICs. The ICs are then used as the input variables of the neural network for building the prediction model. In order to evaluate the performance of the proposed approach, the Shanghai B-Share stock index closing price and Nikkei 225 closing cash index are used as the illustrative examples.
The rest of this paper is organized as follows. Section 2 gives brief overviews of nonlinear ICA and neural network. The proposed model is described in Section 3. Section 4 presents the experimental results and this paper is concluded in Section 5.
