Abstract:
Capital markets are characterized by uncertainty. Investors intend to obtain sat-isfactory investment return; they need to employ a feasible tool-portfolio. Therefore, portfolio theory is an important part of modern finance. However, many traditional models derived from the mean-variance framework focus on the purpose of risk diversification and overlooked predictions of stock prices and market trends. Moreover, the portfolio problem is a non-linear optimization problem. But most traditional analysis methods and models are linear and single objective models. Nonlinear models that can provide external inputs and methods that can quickly search for Pareto optimality can solve the above problems. Fifty stocks from the S&P 500 are randomly selected for this study. The period of the collected samples is from January 1, 2000, to December 31, 2018. Stock prices, S&P 500 indexes, 13 macroeconomic factors and 9 micro factors are employed as inputs to forecasts. First, the influential factors strongly related to the stock price are found by the bivariate correlation analysis. Second, the factors and stock price are inputted to predict the future stock price by NARX model. Stocks with high forecast accuracy are used to form four portfolios. Finally, positive, and negative probabilities and stock returns are employed as objective functions. Pareto optimality of assets allocation is found by GA for multi-objective optimization. The method of combining nonlinear autoregressive exogenous model (NARX) with genetic algorithm (GA) is proved to be effective in making up for the shortcomings of traditional methods. The returns of four portfolios constructed by the methods are higher than the market returns, which is verified by the real data of the quarters of 2018. Moreover, it is worth noting that the univariate model that only enter one macro factor without considering micro factor has less error. In future research, the sample size can be expanded to further improve its effectiveness and reliability.
