Portfolio Selection Using Machine Learning Algorithms

Finding the best combination of buying weights of given portfolios at a certain time is the final goal in portfolio optimization problems. Markowitz showed that diversifying investment on portfolios reaches to a higher amount of return rather than one source of investment. This study focuses on predicting the future efficient pattern of portfolio and picking the final investment strategy. Using machine-learning algorithms for classifying the efficient and non-efficient portfolios and finally using the mean-variance model for obtaining the final weights of opted portfolios is the main purpose of this study. The implementation results showed that taking advantages of classifiers has a bright role in making the old and popular economical models for investment work more efficient. Two different classifiers called Support Vector Machine (SVM) and Extreme Learning Machine (ELM) are used and compared to each other. The use of classification has two major benefits first it classifies stock with the ability to reach a certain gain in the next day and second it reduces the quadratic programming search area by removing non-efficient stocks from available portfolios and leads the algorithm to find the global solution faster. Moreover, a supervised dimension reduction method called Negative Matrix Factorization (NMF) is used to see if some features are redundant or causing any conflict in classification. The results show that the model works better without NMF dimension reduction.

Keywords:

Portfolio Management , Support Vector Machine , Mean Variance Model , Stock Data , Extreme Learning machine , None Negative Matrix