الفهرس | Only 14 pages are availabe for public view |
Abstract The primary purpose of trading in stock markets is to profit from buying and selling listed stocks. Therefore, forecasting stock prices is crucial for successful investment in financial markets. However, numerous factors can influence the stock prices, such as the company’s present financial situation, news, rumor, macroeconomics, psychological, economic, political, and geopolitical factors. Consequently, tremendous challenges already exist in predicting noisy stock prices. Recently, artificial neural networks (ANNs) attained remarkable outcomes in predicting stock markets. Nevertheless, the nonstationarity and the interaction between hidden features of the price time series lessen forecasting accuracy. Consequently, data preprocessing techniques such as discrete wavelet transform (DWT) and singular spectrum analysis (SSA) are required to improve the prediction accuracy of ANNs by reducing the noise and extracting hidden characteristics of the time series. This thesis proposes hybrid models that integrate data preprocessing techniques, i.e., DWT and SSA, with the nonlinear autoregressive neural network (NARNN) to predict stock prices in the Egyptian Exchange. The NARNN involves a time delay line (TDL) in the input layer representing the memory that recognizes subsequent or changing patterns over time and fades the short-term volatility. The DWT-NARNN and SSA-NARNN models first divide the stock prices into training and testing sets. Then the training set is decomposed using data preprocessing techniques to reduce the noise and lessen the data’s nonlinearity. Afterward, each extracted component is utilized for training a separate NARNN. To predict the future components, the model decomposes the preceding available prices at each time step in the testing set and utilizes the latest points as input to the NARNNs. Eventually, the outputs from the NARNNs are aggregated to provide the final predicted prices. The weekly closing prices for twenty-four stocks from the Egyptian Exchange (EGX-30) are used to verify the proposed models’ performance. The DWT-NARNN and SSA-NARNN models are compared with other methods, including the backpropagation neural network (BPNN), NARNN, DWT-BPNN, and SSA-BPNN models. The empirical results reveal that the proposed models perform better than the other s, indicating the data preprocessing techniques’ ability to extract hidden information and reduce the noise effect of the original time series. Moreover, this thesis proves that the old approach of decomposing the entire dataset and partitio then ning it into training and testing sets is unrealistic. The unrealistic approach causes the testing set to inherit the stock’s future performance, leading to optimistic deceptive results. In contrast to the old method, our point simul by point decomposition ates the actual trading process, and the validation process is reliabl |