Forecasting the future behaviour of market prices is an important area of research, which is exploited in asset allocation, risk management and algorithmic trading. Market behaviour is complex and influenced by many factors whose relationships are non-linear. The amount of financial data available for analysis is increasing substantially due to increased volumes of electronic trading, and market participants who are capable of extracting influential information from these massive amounts of data successfully benefit from using it in trading and investments. Computational intelligence techniques are recognised as capable of learning from large complex datasets and recently have been successfully applied to modelling capital markets. This thesis develops several financial forecasting systems based on computational intelligence that utilise financial time series and/or news articles to predict future stock price changes. The thesis starts with a comprehensive review of academic literature in market predictability and financial forecasting based on computational intelligence. As a result of the undertaken research, three main contributions are elaborated. First, a predictive system is developed that aims to predict future stock price changes based on technical indicators computed from financial time series of prices and volumes. The system is then used to study how its predictive performance depends on the choice of the input window length, which is a parameter used when calculating technical indicators, given a forecast horizon. The following pattern was discovered: for each horizon, the highest forecasting performance is achieved when the input window length is approximately equal to the horizon. The persistence of the pattern was successfully tested on fifty stocks using multiple performance measures and four well established machine learning techniques. Second, a news-based predictive system was developed and exploited to study how the simultaneous use of several categories of financial news can enhance its forecasting performance. News articles were allocated to categories based on their relevance to the target stock. To date, the simultaneous use of multiple news categories has not been investigated. The proposed approach was tested on stocks from the healthcare sector. The experiments reveal that pre-processing those categories independently and then using them simultaneously as input to different kernels of the Multiple Kernel Learning approach produces better results than approaches based on a single news category or a lower number of categories. This finding suggests that a proper inclusion of a wide range of news is advantageous in forecasting the market state. Moreover, the duration of the effects that news articles have on stock prices is explored; it is shown that predictive information relevant to future price changes for more than a month is contained in news articles. Third, the possibility of integrating the past prices-based and news-based predictive systems is explored. Primarily, prices and news are used simultaneously, however this integrated approach does not outperform the news-based system. It is concluded that, when important information is released in the form of a news article, this information influences the market and adding the analysis of past prices does not bring more knowledge about the current market state. Next, a combined approach is proposed where past prices and news are used interchangeably in a predictive system depending on the availability of news articles on the day of forecasting. The combined approach shows promising results; its main benefit is the ability to provide optimum support for everyday decision making processes in investment and trading. This thesis presents results of solid research work in financial forecasting based on news articles and historical prices. It also identifies potentially promising directions of future work.