1. Prediction of nonlinear nonstationary time series data using a digital filter and support vector regression
- Author
-
Premanode, Bhusana, Meade, Nigel, and Toumazou, Christofer
- Subjects
621.3 - Abstract
Volatility is a key parameter when measuring the size of the errors made in modelling returns and other nonlinear nonstationary time series data. The Autoregressive Integrated Moving- Average (ARIMA) model is a linear process in time series; whilst in the nonlinear system, the Generalised Autoregressive Conditional Heteroskedasticity (GARCH) and Markov Switching GARCH (MS-GARCH) models have been widely applied. In statistical learning theory, Support Vector Regression (SVR) plays an important role in predicting nonlinear and nonstationary time series data. We propose a new class model comprised of a combination of a novel derivative Empirical Mode Decomposition (EMD), averaging intrinsic mode function (aIMF) and a novel of multiclass SVR using mean reversion and coefficient of variance (CV) to predict financial data i.e. EUR-USD exchange rates. The proposed novel aIMF is capable of smoothing and reducing noise, whereas the novel of multiclass SVR model can predict exchange rates. Our simulation results show that our model significantly outperforms simulations by state-of-art ARIMA, GARCH, Markov Switching generalised Autoregressive conditional Heteroskedasticity (MS-GARCH), Markov Switching Regression (MSR) models and Markov chain Monte Carlo (MCMC) regression.
- Published
- 2014
- Full Text
- View/download PDF