The application of chemometrics derived pattern recognition methods to futures market analysis

Cycle-theory-based market analysis is the main focus of this thesis, in which I try to find systematic methods to recognize and utilize market patterns, especially cyclic ones, to obtain a correct understanding of market movements. The KNN algorithm, a pattern recognition method extensively used in chemometrics, has been employed to recognize similarities of current market movements and historical markets to permit market forecasts. Bayesian analysis, another pattern recognition method, has been used to infer longer-term market trends based on observable shorter-term market behaviors and to improve the real-time application of the KNN algorithm. An artificial neural network method, an example of a non-linear information processing system, has also been applied in this research to combine cycle-relative information to market behavior modeling. The promising overall results show that there exists a correlation between current and historical price movements and shows the possibilities of utilizing pattern recognition methods to obtain correct market forecasts. Also, a novel use of moving averages, especially suitable for an oscillating market, has been introduced and successfully applied in market prediction. In this thesis, the S&P 500 futures market has been chosen as the market to study.