| Under the highly unstable financial markets, investment is full of uncertainty. The stock market is affected by the unpredictability of financial markets seriously, therefore holding a diverse portfolio with minimum risks is no doubt all the expectations of investors. The effective prediction of price fluctuation can highly influence the decision of the portfolio investment strategy.In this paper, according to the economic data of stock, the samples are divided into some groups by high dimensional clustering method, to find out the relative portfolio. Then, the portfolio is used to model and predict closing price.Financial Time Series usually are the multivariable time series. The effect of traditional time series model to fit such time series and predict are not very ideal. Therefore, in this paper, using the phase space reconstruction theory, combined with the genetic algorithm to the SVR(support vector regression) to estimate parameters, can find the optimal delay time and embedding dimension, to predict the 15-days seven indexes which influence the closing price. Then the LASSO algorithm is used to predict the closing price of above indexes. Finally, through the forecast of the closing price, its tendency can determine the investment limit.The empirical results show that, compared with the single variable time series clustering, the rationality of multivariate time series clustering is obvious. By comparing the multivariable prediction with single variable prediction based on phase space reconstruction, it shows that the former prediction effect is better than the latter, less error and high precision. Prediction of the closing price in terms of multivariable, sparse LASSO is high prediction accuracy, more conducive to explain the model than the SVR and ARIMA. It can capture the fluctuation trend of time series more easily than SVR and ARIMA. To determine the portfolio trading strategy, investors can reduce risk and get high profit rate. |
PDF Full Text Request