CMStatistics 2018: Start Registration
View Submission - CFE
Title: Decomposition of high frequency Forex signals for copula based pairs trading strategy with support vector regression Authors:  Carlin Chu - The Open University of Hong Kong (Hong Kong) [presenting]
Po Kin Chan - Quantitative Research Team - Legend Arb Trading Limited (Hong Kong)
Abstract: The microstructure noise inherited in high frequency trading signal is a nuisance for effective modeling of short-term trends. The typical approach to tackle this issue is to remove the noisy data part by filtering or smoothing methods. Oscillating zig-zag patterns are considered as noises and smoothed out. However, this approach does not consider the possibility of utilizing the information hidden in the noise data. A proper usage of decomposition method to exploit the noise information is investigated. Empirical Mode Decomposition (EMD) is proposed to decompose the non-stationary trading signal into intrinsic mode functions (IMFs) and residual series for building a prediction model. The aim is to extend the use of copula-based Mispriced Index (MI) and Support Vector Regression (SVR) for pairs trading by incorporating IMFs on the model building stage. Properties of IMFs, selection of bivariate copulas and suitability of different SVR kernels are examined. The empirical results indicated that the use of IMFs can significantly improve the model accuracy in almost all settings. The subtle relationships among the EMD, copula functions and hyperparamters of SVR are discussed.