ON FINANCIAL TIME SERIES DATA MINING

Abstract

The portfolio selection problem haH a venerable hi::;tory. I\•1arkowitz {1952)) one of the creators of the modern portfolio theory, formulates the problem as a trade-off be­ tween the expected return and the expected risk of a portfolio. For his path breaking work that has revolutionized investment practice, he \VOn the Nobel Prize in 1990. In this Dissertation we propose two enhancements to the traditional portfolio selection problem. First we enhance the formulation of the problem by introducing four addi­ tional constraints that take into account the following: (a) the collinearity problem to decrease the portfolio risk, (b) the special preference to active stocks to incrca::;e the expected return and decrease the systematic risk, (c) the special preference to stocks with outstanding performance to increase the un-expected return, and (d) control the overall risk of the portfolio. Second, one of the common al orithms for solving the portfolio election optimiza­

tion problem is the Genetic Algorithm (GA), which is a stochastic search that starts with an initial solution and then allocates increasing trials to regions of the search space found to improve the objective function. This. algorithm can run into problems when the optimal solution is in a small region surrounded on all directions by regions of lmv value of the objective function. " 11'./e propose an enhancement to the GA that avoids this problem.