Derivation of a Dynamic Market Risk Signal Using Kernel PCA and Machine Learning

Abstract

Kernel principal component analysis (PCA) is an extension of the conventional PCA method that employs a kernel transformation whereby hidden patterns in possibly multidimensional data may be detected and extracted more explicitly. In this article, the author applies the method of kernel PCA to a currency prediction case study and derives an aggregate market signal. It is observed that this signal has desirable information-compression properties and may be used as a predictive risk indicator in the return prediction models. Used alongside common drivers of exchange rates, a kernel PCA signal enhances in-sample and out-of-sample risk-adjusted performance across a range of machine learning strategies. In particular, the author observes that a kernel PCA signal remains robust and predictive during volatile market conditions. The kernel PCA signal may be used as a machine learning feature to inform and support data-driven risk management strategies. TOPICS:Big data/machine learning, derivatives Key Findings • This article develops a risk signal based on the method of kernel PCA. • Kernel PCA conveys desirable information on market dynamics. • Currency machine learning strategies benefit from this kernel PCA feature.