Constructing of Optimal Portfolio on Russian Stock Market Using Nonparametric Method - Artificial Neural Network

Abstract

In this paper were used nonparametric method Artificial Neural Network (ANN) for analyzing 50 stocks, which are included in calculation base of MICEX stock index. Using of this method allowed us to use not only macroeconomic and technical factors, but also factors with limited data set (factors of fundamental analysis). As a result we constructed optimal portfolio with average return on 8% higher than market portfolio with the same risk during period Jan. 2015 - Jan. 2016.

ANN method also allows conducting a comparative analysis of influence of the factors on stock return. As the result we showed that Russian stock market has features of speculative market, because the most important factors for stock return of Russian stocks are momentum, bid - ask spread and oil price. Significantly, that the same factors were determined in research dedicated to the problem of constructing optimal portfolio on Russian stock market using classification and regression tree (CART) method. Potential investors take into account oil price as key determinant of economic environment in Russia and select stocks with high momentum and high liquidity.

In this research ANN method was compared with another nonparametric method (CART) by solving utility maximization problem of investors with different value of coefficient of risk aversion. As the result, ANN method shows a strictly higher return than CART during analyzing period. This fact could be explained by the logic of ANN method: ANN doesn't require more observations if the number of independent variables. On the other hand CART method requires more observations because of minimal number of observation in one node. More variables are in the structure of CART, more nodes will be in the tree, hence, more observations are required.