Stock market time-series behavior predictability and profitability

This dissertation presents an investigation of the behavior of stock market time series in terms of "randomness" and "predictability" from three different perspectives.; First, some of the established theoretical assumptions and tools used in past studies were tested. The use of autocorrelation as a tool to judge randomness of stock market time series was put in question because of nonlinearity issues, and empirical results contradicted the assumptions of the theoretical Random Walk model. Shannon's entropy and short-term patterns were used as additional tools to see if they add any knowledge about the behavior of stock market time series. It was found that these techniques can increase the overall understanding of price fluctuations.; Subsequently, the importance, in terms of predictability and profitability, of the reoccurring short-term chart patterns in stock market time series was substantiated. And, the generality of the Efficient Market Hypothesis was refuted via counter example. Specifically, the Random Walk model was rejected as a null. In addition, it was shown that predictability based on a heuristically selected pattern can lead to profitability as long as transaction costs are reasonable and the funds invested are adequately high.; The last part of this study focused on the notion of entropy and its adequacy as a measure of randomness in the stock market where there is a need to objectively classify time series based on their randomness level. Three different methods based on Shannon's entropy were proposed and used to measure randomness for stock market time series for both direction and magnitude. Using the proposed methods, entropy was found to be a useful tool to compare and rank stock market time series based on their randomness and to detect local areas where randomness may be low.