The previous researches in stock market predictions using Artificial Intelligence Techniques, such as Neural Networks, Case-Based Reasoning, etc., were focused mainly on spot market index prediction. Finally on May 3, 1996, Korea launched trading in index futures market(KOSPI 200), and then more and more people became attracted to the market. However, lack of the previous researches for the index futures market seemed to discourage people's interest. Thus, this research was intended to predict the daily fluctuating direction of KOSPI 200 index futures price to meet this recent surge of interest. The forecasting methodologies employed in this research were Genetic Algorithm/Neural Network Integrated Method(GANN) and Genetic Algorithm/Case-Based Reasoning Integrated Method(GACBR). Genetic Algorithm was used to select relevant input variables.
As for the data preprocessing method, this research divided that into typical data preprocessing and expert's knowledge-based categorical data preprocessing. The experimental results of each forecasting methods with each data preprocessing method were cautiously compared and statistically tested.
In the end, Neural Network and Case-Based Reasoning methods with best performance were integrated. Out-of-the-Model Integration and In-Model Integration were presented as the integration methodology. This research analyzed the experimental results of the integrated model and its practical meaning.
The research outcomes were as follows,
First, Genetic Algorithm was a useful method to select input variables in AI techniques. Second, the results of the experiment with categorical data preprocessing significantly outperformed that with typical data preprocessing in forecasting fluctuating direction of index futures price. Third, Genetic Algorithm/Case-Based Reasoning Integrated Method outperformed Genetic Algorithm/Neural Network Integrated Method in experimental results. Fourth, Genetic Algorithm, Case-Based Reasoning, Neural Network Integrated Model revealed similar or somewhat worse prediction accuracy than GACBR, but they bore higher return than GACBR and GANN from a practical view.
