Financial time series forecasting using Simulated Annealing and Support Vector Regression

Abstract

The behavior of stock exchanges around the world has an important role in the financial development of countries. The Mexican Stock Exchange (BMV, for its acronym in Spanish) is the financial entity in Mexico where investors build and manage investment portfolios to generate profit. Financial time series forecasting is an important problem for investment portfolios creation and optimization because it allows them to have a prediction of the value of investment assets over time, therefore reducing at some extent the uncertainty of these operations. Classical models, such as ARIMA, artificial neural networks and support vector machines, are usually used as forecasting tools. In this paper, we forecast financial time series of the BMV using a hybrid parallel algorithm that employs both simulated annealing (SA) and support vector machines for regression (SVR). Finally, the resulting Mean Absolute Percent Error (MAPE) of this hybrid algorithm is compared with that of an ARIMA model.