مدل ترکیبی بر اساس شبکه تطبیقی ​​مبتنی بر سیستم استنتاج فازی و الگوریتم ژنتیک تطبیقی برای پیش بینی بورس اوراق بهادار سرمایه شاخص سهام وزنی تایوان

Technical analysis is one of the useful forecasting methods to predict the future stock prices. For professional stock analysts and fund managers, how to select necessary technical indicators to forecast stock trends is important. Traditionally, stock analysts have used linear time series models for stock forecasting. However, the results would be in doubt when the forecasting problems are nonlinear. Further, stock market investors usually make short-term decisions based on recent price fluctuations, but most time series models only use the last period of stock prices in forecasting. In this paper, the proposed hybrid model utilizes an adaptive expectation genetic algorithm to optimize adaptive network-based fuzzy inference system (ANFIS) for predicting stock price trends, and four proposed procedures are included in the hybrid model for forecasting: (1) select essential technical indicators from popular indicators by a cited paper (Cheng et al., 2010); (2) use subtractive clustering to partition technical indicator values into linguistic values based on an objective data discretization method; (3) employ fuzzy inference system (FIS) to build linguistic rules from the linguistic technical indicator dataset and optimize the FIS parameters by adaptive network; and (4) refine the proposed model using the adaptive expectation model, which optimizes parameter by genetic algorithm. The effectiveness of the proposed model is verified with performance evaluations and root mean squared error (RMSE), and a 6-year period of TAIEX (Taiwan Stock Exchange Capitalization Weighted Stock Index) is selected as the experimental datasets. The experimental results have shown that the proposed model is superior to the three listing forecasting models (Chen's model, Yu's model, and Cheng et al.'s model) in terms of RMSE.

In stock markets, there are complex factors that will influence stock markets and nonlinear relationships, which are contained among different periods of stock prices for investors to forecast the future stock trends with difficulty. Therefore, many forecasting methods have been employed in predicting stock prices since the first stock market was opened. Further, financial analysts and stock fund managers attempt to predict price activity in the stock market on the basis of either their professional knowledge or with the assistance of stock analyzing tools. If more accurate predictions are given, myriad profit will be made. Therefore, stock analysts have perennially strived to discover ways to predict stock prices accurately. However, forecasting stock returns is difficult, because market volatility needs to be captured in a used and implemented model. Accurate modeling requires, among other factors, consideration of phenomena that are characterized. Many conventional numeric forecasting models by financial researchers have been proposed, such as Engle's (1982) autoregressive conditional heteroscedasticity (ARCH) model, Bollerslev's (1986) generalized ARCH (GARCH) model, Box and Jenkins's (1976) autoregressive moving average (ARMA) model, and the autoregressive integrated moving average model (ARIMA). Further, many researchers have been focusing on technical analysis to improve the investment return (Azo, 1994, Chi et al., 2003 and Kimoto et al., 1990), because technical analysis methods are one of the major analysis approaches for investors to make investment decisions. A technical analysis method is one of the major analysis techniques in stock markets that has the ability to forecast the future price direction by studying past market data, primarily stock price and volume. The technical analysis method has assumed that stock price and volume are the two most relevant factors in determining the future direction and behavior of a particular stock or market, and the technical indicators, coming from the mathematic formulas based on stock price and volume, can be applied to predict the future price fluctuation and also provide for investors to determine the timing of buying or selling stocks (Chi et al., 2003). In the evolution of time series models, many researchers have applied data mining techniques in financial analysis. In 1990, Kimoto et al. (1990) developed a prediction system for stock markets by using neural networks. The following researchers, Nikolopoulos and Fellrath (1994), have combined genetic algorithms (GAs) and neural networks to develop a hybrid expert system for investment decisions. Kim and Han (2000) have proposed a genetic algorithm approach to feature discretization and the determination of connection weights for artificial neural networks (ANNs) to predict the stock price index. Huarng and Yu (2006) have applied backpropagation neural networks to establish fuzzy relationships in fuzzy time series for forecasting stock price. Roh (2007) has integrated neural networks and time series models for forecasting the volatility of a stock price index. From the literature above, there are four major drawbacks found in these forecasting methods and models: (1) stock market analysts and fund managers apply various technical indicators to forecast stock markets based on their personal experiences, which might give wrong judgments on market signals; (2) for most statistical methods, there are some assumptions about the variables used in the analysis, and they cannot be applied to these datasets, which do not follow the statistical distributions; (3) artificial neural network (ANN) is a black-box method, and the rules mined from the methods are not easily understandable; and (4) stock market investors usually make short-term decisions based on recent price fluctuations, but most time series models use only the last period of stock price in forecasting. To improve the past forecasting models, a thoughtful model should be able to overcome these drawbacks contained in past models and offer good methodology to be used and realized easily by investors. In time series research, the adaptive expectation model (Kmenta 1986) is a reasonable forecast model. Stock market investors usually make their short-term decisions based on recent stock information, such as the latest market news or price fluctuations. Furthermore, Chen et al. (2008) proved that the price patterns in the Taiwan and Hong Kong stock markets are short-term. Therefore, for stock price forecasting, we claim that the linear relationships between recent periods of stock prices should be included in forecasting models (Chen et al., 2007 and Cheng et al., 2006). To consider the linear relationships between recent periods, we apply an adaptive expectation model to the proposed model for enhanced forecasting performance. Further, genetic algorithms, which are usually the preferred solution to the optimization problems, perform genetic operations, such as selection, crossover, and mutation. Genetic algorithms provide near-optimal solutions for an evaluation (fitness) function in optimization problems. For this reason, this paper uses a GA to optimize the parameter of the adaptive expectation model. Therefore, this paper proposes a hybrid forecasting model to refine past models in stock price forecasting, and there are four processes provided in the forecasting model: it (1) selects essential technical indicators from a citation paper (Cheng et al., 2010); (2) uses subtractive clustering to discretize condition features (technical indicators); (3) applies a fuzzy inference system (FIS) to produce rules from the linguistic values of technical indicators and employs an adaptive network to optimize FIS parameters to improve forecasting accuracy; and (4) refines the forecasting accuracy of the proposed model by the adaptive expectation model, which optimizes parameters using a genetic algorithm. In an empirical study, this paper employs a stock index as the experimental datasets. From the model verification, it is shown that the proposed processes are effective in improving forecasting accuracy, and based on the evidence, stock analysts or investors can employ the refined processes proposed in this paper to improve their forecasting tools or models. The rest of this paper is organized as follows. Section 2 introduces the related works. Section 3 demonstrates the proposed model and algorithm. Section 4 describes the model verification. Conclusions are finally drawn in Section 5, as well as findings, conclusions, and recommendations for future research.

This paper has proposed a new hybrid model based on technical indicators, subtractive clustering, ANFIS, and an adaptive expectation genetic algorithm to promote forecasting performance for forecasting the stock market. There are two conclusions: (1) the experimental results (see Table 3) show that the proposed model could efficiently describe the fluctuation trends in the TAIEX; and (2) few forecasting rules are generated (only two rules); the number of rules is the same as the number of linguistic intervals by subtractive clustering to forecast electricity loads. From the verification in Section 4, we can see that the proposed model significantly outperforms the listing models. From the experimental results, there are two findings in this paper to show why the proposed model surpasses Chen's (1996) model and Yu's (2005) model as follows: (1) Chen's (1996) and Yu's (2005) models only use stock price as input data. But, the proposed model takes into account the causality of technical indicators with ANFIS learning and incorporates an adaptive equation for TAIEX forecasting. This is the main reason why the proposed model outperforms Chen's (1996) and Yu's (2005) models. (2) GA is suitable for parameterized optimization problems, and the proposed method can achieve the minimal RMSE at which parameter α within the adaptive expectation equation is generated by the genetic algorithm. The proposed model utilizes an adaptive expectation genetic algorithm to refine forecasts. However, Chen's (1996) and Yu's (2005) models did not apply the GA method to their models, and these two models cannot surpass the proposed model. Additionally, from the experiment, two advantages for the proposed model are discovered: (1) the more reasonable and understandable “if-then” rules produced by ANFIS can model the qualitative aspects of human knowledge; and (2) the proposed model produces forecasting rules based on objective stock data rather than subjective human judgments, providing stock investors' objective suggestions (forecasts) to make investment decisions in stock markets. For subsequent research, we can use other stock datasets, such as China, Japan, and USA, to further validate the proposed model; there are two suggested approaches to refine the proposed model to improve forecasting performance: (1) validate the generated rules by an expert group to improve accuracy; and (2) other data mining methods can be employed in the proposed model.