Analyzing the Dynamics of Stock Networks for Recommending Stock Portfolio

摘要

Traditional approaches to portfolio management and optimization often rely on the certain statistical properties, such as expected return and price variance. But these properties generally represent the local behavior of the stocks and are thus not able to represent the stock characteristics in terms of the whole stock market. This paper considers the stock market as a complex system, where stocks affect one another due to unknown forces. It may not possible to measure the actual magnitudes and directions of these forces, but it is possible to observe their effects using the external observations such as the correlation between stocks. To forecast the dynamics, this paper proposes a seminal measure, the cohesion of the stock market network induced by the correlation of stocks. The important observations we obtained from the analyses of the real market (S&P500 and KOSPI200) in the past thirteen years are two folds: (1) the cohesion tends to increase more in a bear market than in a bull market, and (2) the cohesion of the stock market Granger causes the stock returns. Based on these observations, we implemented a stock portfolio recommending system, namely StoPoR. To evaluate the effectiveness of StoPoR system, we conducted the simulated investment based on the portfolio recommended by the StoPoR. The result shows that the monthly returns of StoPoR portfolio (1.44%) is bigger than that of Markowitz efficient portfolio (1.07%) and further bigger than that of the S&P500 index (0.50%). The similar result also holds for the Korean stock market; the monthly return of the StoPoR suggested portfolio (1.83%) gets far bigger than that of Markowitz efficient portfolio (1.61%) and further bigger than that of the KOSPI200 index (0.81%). This result indicates that the characteristics of the stock network are related to the stock returns and its dynamics can be used for constructing a stock portfolio and the prediction of the changes in the stock markets.