Machine Learning Approach for Investment Portfolio Management

It is not a surprise that humankind has a tendency to structure, systematize and regulate all. Capital markets are not an exception. Notions of chaos and order constantly accompany finance industry researches, that, in turn, get investment portfolio management, a crucial realm of finances, a much puzzled – over activity indeed. A linear paradigm, used for capital market depicting for long years, has been undermined and exposed to a subverting process. Moreover, modern finance has become global and highly interconnected that produces sparse, noisy and restricted market data. Using traditional risk measures and return forecasting in high-dimensional settings poses significant risk with respect to portfolio optimization. Therefore, an acute need for methods, enabling to reveal hidden nonlinear interconnections in massive finance data and, thereby, deriving a profit, has reached its pick. In parallel «Advanced Analytics», «Big Data», «Machine learning» have become buzzwords in many industries. Despite being long standing, these terms did not attract any attention, but their potential has bursted in the scientific and business worlds until quite recently. Aforementioned facts have served as the reason of writing the paper. A multistage approach to investment portfolio building was proposed within which some strategies of its construction were tested. A core concept of portfolio has become Markowitz model but with alternative interpretation based on machine learning methods and AI. Cluster analysis, neural network were applied for stocks’ selection, finding expected returns, asset risk and portfolio risk measures. Proposed strategies are directed at speculative investment portfolio construction focused on profit from price fluctuations under lower risks that determines significance of the paper. As a result, it has turned out that in ~80% of cases machine learning approach is effective allowing to achieve portfolio return exceeding index of blue chips.