۱٫ Introduction

۲٫ Dataset

۳٫ Research method and data analysis

۴٫ Discussion

۵٫ Conclusion

Acknowledgments

Appendix.

References

Abstract

This research examines the foreign exchange (FX) market from the perspective of currency network. We construct the network based on correlations between exchange rates of 37 currencies from 2006 to 2012. The minimum spanning tree (MST) is used to generate a simplified network and bootstrap technique is employed to test the reliability of links. The full correlation matrices are further analyzed to support and test the robustness of the results from the MSTs. Specifically, we compare the results in the pre-crisis period (2006-2007) and the post-crisis period (2011-2012) to show the impact of the 2008 global financial crisis on the FX market. We have the following findings: (a) the currency network is more scattered in the post-crisis period; (b) the connections between currencies tend to be more internal within the geographic region after the crisis; (c) the European currencies maintain strong connections and keep their clustering feature stable.

Introduction

The process of globalization is closely linking financial markets worldwide, leading to the formation of economic networks. As one of the most important markets in the global financial system, the foreign exchange (FX) market has experienced a large increase in recent decades. According to the triennial survey conducted by Bank for International Settlements (BIS), global FX turnover averaged $5.1 trillion per day in 2016, down from $5.4 trillion in 2013 for the appreciation of the US dollar, but up with a 27.5% increase by comparison with $4.0 trillion in 2010 [1].

The FX market has received attention from several disciplines such as economics, physics and systems science. The network approach from systems science and econophysics seems to provide a new paradigm to touch the complexity of the FX market with the interaction of world currencies. From the network perspective of the FX market, the currencies (or exchange rates) are considered as nodes and the pairwise correlations between them as the links, and thus a similarity based network can be constructed