Actionable Stock Portfolio Mining by Using Genetic Algorithms

陳俊豪(CHUN-HAO CHEN) ； 謝晶宇(CHING-YU HSIEH)

《Journal of Information Science and Engineering》 32卷6期 (2016/11) Pp. 1657-1678

摘要

Financial markets have many financial instruments and derivatives, including stocks, futures, and options. Investors thus have many choices when creating a portfolio. For stock portfolio selection, many approaches that focus on optimizing the weights of assets using evolutionary algorithms have been proposed. Since investors may have various requests, an approach that takes these requests into consideration is needed. Based on the domain-driven data mining concept, this paper proposes a domain-driven stock portfolio optimization approach that can satisfy an investor's requests for mining an actionable stock portfolio. A set of stocks are first encoded into a chromosome. Two real numbers that represent whether to buy a stock and the number of purchased units, respectively, are utilized to represent each stock. In the fitness evaluation, each chromosome is evaluated in terms of the investor's objective and subjective interestingness. Objective interestingness includes return on investment and value at risk. Subjective interestingness contains a portfolio penalty and an investment capital penalty, which reflect the satisfactions of the investor's requests. Experiments on real datasets are conducted to show the effectiveness of the proposed approach.

關鍵字

data mining ； domain-driven data mining ； genetic algorithms ； minimum transaction lots ； stock portfolio optimization