Grant: Research Tools and Instruments - Category 1 (<$150,000)

Capital market refers to the complex and data intensive environment that supports the trading of financial products, such as stocks. It involves a very large amount of trading data far beyond the processing capability of the human brain. The practitioners of the capital market have increasingly recognized the importance of machine learning as a critical tool for technical analysis. Our research aims at discovering weak regularities from a tremendous quantity of noisy financial data to predict future trends of the stocks to form an optimal portfolio. We have designed a novel learning algorithm called Adaptive Prototype Kernel Regression for the task. Initial results are encouraging and promising. Our program is able to find weak trends from the historical stock data and form an optimal portfolio that consistently outperforms previous methods and the market indexes even after considering the trading cost. We have published our discoveries in two top conferences in the financial industry. Due to the significant amount of data to be mined, our algorithm is extremely hungry in computer speed and memory. In addition, we must frequently re-run the machine learning program on new data to accurately reflect the rapid changes in the capital market. The current computers in our lab or our department are vastly inadequate for this purpose. In addition, due to the sequential nature of the learning task and our algorithms, Sharcnet, a parallel computer network, is of little use to us. We propose to purchase a Sun Fire V40z server which is specifically designed for the tasks such as modeling to carry out the research. The new equipment is the key towards the success of our research, which will bring large impact not only to the machine learning community but also to financial companies in Canada.