Chemometrics-derived methods and statistical techniques to model and forecast futures markets

Based on the perceived internal structure of market price movements on appropriate time scales, a series of interrelated pattern recognition techniques were designed and applied to the cyclic analysis of financial futures markets. The research analysis is based on technical analysis which is used to forecast futures price movements by deriving price patterns from historical price movements in presumably analogous situations. Suitable chemometric methods such as the K-Nearest Neighbors method are applied to recognize such price pattern in two representative futures markets. The models for price pattern recognition described in this thesis are based on market cycle analysis which can predict futures market movements with a high degree of accuracy. Two types of futures, namely stock index futures (S&P 500) and foreign currency futures (Swiss franc) are chosen as model futures markets. Automatic cycle identification in real-time markets and the integration of different prediction signals are described in the discussion of my market modeling. Some statistical methods such as frequency distribution are applied to analyze and visualize the forecasting results. It is shown in an appendix that the analytical methods discussed can also be applied successfully to a different technical analysis model not involving cycle analysis.