| Hi-index | 0.00 |
With the growing importance of the role of equities to institutional and individual investors, the selection of attractive stocks to reach a higher return performance is a complex challenge. A reliable tool that helps identify the top performing stocks with minimum effort in the selection process can be of great assistance to investors. This dissertation proposes a computation based forecasting system that integrates the generalized regression neural network and two supporting technologies, namely information gain and principal component analysis, to manage stock portfolios. The integrated forecasting system takes advantage of the synergy among these technologies to collect the relevant data from various financial variables, transform the data so that they will be uncorrelated, and estimate the daily expected stock return performance from the underlying data. As a test case, thirty stocks listed in Dow Jones Industrial Average Index (DJIA) were selected in this dissertation. In a simulation with out-of-sample data, two trading systems that were designed to form a portfolio of stocks were investigated. For the first trading system, portfolios were constructed from stocks that achieve expected returns of at least ±2% performance, representing a reasonable opportunity to take a position. The portfolios with the positive performance estimations, the negative performance estimations, and a combination of each were constructed. For the second trading system, the 30 individual stocks were ranked according to their expected return estimations. The four stocks with the best ranking, the worst ranking, and a combination of each were put into portfolios. The buy-and-hold strategy and the ARIMA model were used as benchmarks for both trading systems. These benchmarks helped evaluate whether an adoption of the systematic trading directed by the forecasting system leads to profitability improvement when constructing stock portfolios. Experimentations with different computational parameters of the forecasting system, as well as extended trading of the individual stocks and stock portfolios, were investigated to validate system robustness and performance. The sensitivity results of the stock portfolios constructed with different expected return performances and with a different number of stocks were also reported. The integrated forecasting and trading systems gave promising results for all stock portfolios evaluated in the research.