Fuzzy GA-SVR for Mexican Stock Exchange's Financial Time Series Forecast with Online Parameter Tuning
Time series play a major role in a large number of economic problems such as predicting the volatility of price indices and market rates on the stock exchanges. Volatility reflects the behavior of fluctuations in the prices of assets, allowing to measure the risk of portfolios. However, volatility is not directly observable, is not constant throughout the period, is dependent on the past and does not follow Gaussian normality. In the literature, there have been many studies using classic methods for the Mexican Stock Exchange. New heuristic methods of supervised learning for forecasting financial time series are presented in this paper. These methods are structured in such a way that, knowing only the time series, the values of the dependent variable several periods ahead are estimated with low errors. This work presents a hybrid method based on support vector machines, fuzzy controllers, evolutionary algorithms and classic methods applied to the asset portfolios optimization of the Mexican Stock Exchange. New heuristic methods of supervised learning for forecasting financial time series are presented in this paper. Besides, a fuzzy logic approach with a Genetic Support Vector Regression (Fuzzy GA-SVR) algorithm for the prediction of time series with a broad spectrum of application is presented. This algorithm uses an online tuning process to produce superior efficacy than the classic forecasting methods. Experimentation presented in the paper shows that Fuzzy GA-SVR outperforms both, the classic ARIMA methods and the individual use of SVR.