Algorithmic Trading Model for Manifold Learning in FX

This paper will give a brief overview of the work of introducing machine learning intelligence in the Kineta e-markets system, to facilitate auto-hedging, smart price engine algorithms and proprietary automatic positioning within the foreign exchange market. In this paper we will give a brief overview of the steps taken in the project. A number of quantitative techniques have been implemented in the system and evaluated. As of late we have investigated the use of manifold learning; a class of geometrically motivated nonlinear data mining methods, to predict movements in the foreign exchange market. Financial time series are often correlated over time; and may contain valuable customer specific proprietary information. In principle, such relationships may be exploited for forecasting. However, they may be noisy, nonlinear and changing over time, making this a challenging task. Hence, robust methods for detection and exploitation of such correlations are of high interest for model trading and quantitative strategies. To this end, we study the application of a proposed method for nonlinear regression on manifolds. The approach involves dimensionality reduction through Laplacian Eigenmaps and optimization of cross-covariance operators in the kernel feature space induced by the normalized graph Laplacian