Clustering and Dynamics of Foreign Exchange Markets
This event is part of the Preliminary Oral Exam.
H.E. Stanley Irena Vodenska Ophelia Tsui Kevin Black
Abstract: Financial markets are highly complex and dynamic systems. The challenges posed by non-linearity and non-stationarity have attracted interdisciplinary approaches to investigate new perspectives on this complexity. We study 12 of the most actively traded currencies from 2005 to 2014 with intraday resolution of 10 minutes, yielding more than 300,000 observation points. Currencies are quoted in pairs, such as USDEUR and USDAUD, which gives rise to a triangular structure in the foreign exchange market. This calls for devising a novel way to study individual currencies and their characteristics. We introduce the symbolic performance to describe the roles currencies play within the global market, removing their pairwise dependence. This approach allows us to classify currencies as individual entities. In order to study the dynamics of currency roles, we employ machine learning algorithms, such as k-means to find clusters of characteristic behaviors. Interestingly, we find that the euro shows features of a reference currency more so than the US dollar, which is evidenced by their respective distributions of the symbolic performance across the entire time scale. We also demonstrate how our approach allows us to uncover currency movements hidden in the pairwise dependencies.