psychology e-book

This paper analyzes and examines the general ability of Support Vector Machine (SVM) models to correctly predict and trade daily EUR exchange rate directions. Seven models with varying kernel functions are considered. Each SVM model is benchmarked against traditional forecasting techniques in order to ascertain its potential value as out-of-sample forecasting and quantitative trading tool. It is found that hyperbolic SVMs perform well in terms of forecasting accuracy and trading results via a simulated strategy. This supports the idea that SVMs are promising learning systems for coping with nonlinear classification tasks in the field of financial time series applications.


Introduction
Support Vector Machines (SVMs) have proven to be a principled and very powerful supervised learning system that since its introduction (Cortes and Vapnik (1995)) has outperformed many systems in a variety of applications, such as text categorization (Joachims (1998)), image processing (Quinlan et al. (2004)), and bioinformatic problems (Brown et al. (1999)). Subsequent applications in time series prediction (M¨uller et al. (1999)) indicate the potential that SVMs have with respect to economics and finance. In predicting Australian foreign exchange rates, Kamruzzaman and Sarker (2003b) showed that a moving average-trained SVM has advantages over an Artificial Neural Network (ANN) based model, which was shown to have advantages over ARIMA models (2003a). Furthermore, Kamruzzaman et al. (2003) had a closer look at SVM regression and investigated how it performs with different standard kernel functions. It was found that Gaussian Radial Basis Function (RBF) and
polynomial kernels appear to be a better choice in forecasting the Australian foreign exchange market than linear or spline kernels. Although Gaussian kernels
are adequate measures of similarity when the representation dimension of the space remains small, they fail to reach their goal in high dimensional spaces (Francois et al. (2005)).We will examine the general ability of SVMs to correctly classify daily EUR/GBP, EUR/JPY and EUR/USD exchange rate directions. It is more useful for traders and risk managers to predict exchange rate fluctuations than their levels. To predict that the level of the EUR/USD, for instance, is close to the level today is trivial. On the contrary, to determine if the market will rise or fall is much more complex and interesting. Since SVM performance mostly depends on choosing the right kernel, we empirically verify the use of customized p-Gaussians by comparing them with a
range of standard kernels. The remainder is organized as follows: Section 2 outlines the procedure for obtaining an explanatory input dataset. Section 3 formulates the SVM as applied to exchange rate forecasting and presents the kernels used. Section 4 describes the benchmarks and trading metrics used for model evaluation. Section 5 gives the empirical results. The conclusion, as well as brief directions for future research, are given in Section 6.