Stock price prediction based on singular spectrum analysis and support vector machines

The prediction of stock market price direction serves as a primary financial distress warning framework for long-term investors, and as early reference framework for short-term investors. The most important factor in selecting a forecasting method is accuracy. The suitable stock selection appropriate for investment, is a tough task. The key aspect for each stockholder is to earn maximum revenues on investments. In the research reported in the thesis, singular spectrum analysis and support vector machines were used. Singular spectrum analysis is an effective noise reduction algorithm, which eliminates noise by choosing informative terms and reconstructing a filtered time series which is nonstationary or noisy. Support vector machines is a detailed classifier prediction model using kernel functions and capacity control of the decision function. In the proposed approach for stock prediction, singular spectrum analysis is used as a preprocessing tool for producing a reconstructed time series and its characteristic features. These features with additional parameters, such as technical indicators, are used to train the support vector machines for forecasting stock prices. The proposed approach has been implemented in R language and applied on two years of data retrieved from the Yahoo Finance website. The results indicated that the proposed approach was effective in predicting stock prices with accuracy, better than previously proposed approaches