This paper propose the rough set approach for prediction and trading in stock index futures and we experiment it with KOSPI 200 futures price. And the comparison of discretizational methods as data preprocessing is included in the area of this research.
The rough set approach is useful to extracting trading rules in stock market. It can handle the imprecise data and need no additional information about data. And it can construct the rule sets composed of essential variables (attributes) for catching the decision signal. The most important characteristic in rough set is as enabler of instance selection. The patterns of all the day are not needed, for we do not trade everyday due to transaction cost. So the rough set approach is appropriate for detecting stock market timing because this approach does not generate the signal for trade when the pattern of market is the one with uncertainty. This is due to the reduction of objects(cases, rules) as well as of attributes(variables). The experiment results are encouraging with respect to profitability in the comparison with 'buy and hold strategy'.
The other part of this researches is the comparison of discretizational methods. Four methods are compared. (domain knowledge based cutting, equal frequency cutting, minimum entropy cutting, and naive and boolean reasoning cutting) The most profitable method in training sample is the naive and boolean reasoning cutting but in validation sample is the domain knowledge based cutting. The domain knowledge cutting shows the consistent result in profitability.