250 likes | 443 Views
Dynamics of the FX Market: A Minimal Spanning Tree Approach. Omer Suleman OCCF and Department of Physics University of Oxford Collaborators: N F Johnson, M McDonald, S Williams, S Howison . Networks. Yeast Proteins. World Wide Web. Stock Market. High School Dating. Networks.
E N D
Dynamics of the FX Market:A Minimal Spanning Tree Approach Omer Suleman OCCF and Department of Physics University of Oxford Collaborators: N F Johnson, M McDonald, S Williams, S Howison
Networks Yeast Proteins World Wide Web Stock Market High School Dating
Networks Cyclic Network Tree: Acyclic Network Fully Connected Network
Networks of Financial Time Series • Correlation Based Networks • Entities generating financial time series (stocks, indices, hedge funds or currencies) are represented by nodes. • Weighted edges between nodes represent the correlation between the time series generated by these entities. • this gives us a fully connected network with ½[n(n-1)] edges where n is the number of nodes.
Filtering the Connections • The fully connected network contains too many connections, each with a range of possible weights, and hence too much information for it to be useful. • A filter has to be applied to this network in order to extract the most important links between the nodes thus clustering them. • Any scheme to do this will need a measure of distance or dissimilarity between nodes.
Distance • The weights of the links between nodes are based on the correlation between them. • The most intuitive measure of distance is the Euclidian distance between the time series: • This is a non-linear transformation of the correlation which gives a metric distance between nodes.
Metric Space: d(x,x) = 0 d(x,y) = d(y,x) d(x,z) ≤ d(x,y) + d(y,z) Ultrametric Space: u(x,x) = 0 u(x,y) = u(y,x) u(x,z) ≤ max{ u(x,y) , u(y,z) } Ultrametricity Ultrametric distance is a measure of distance found useful for data classification. • Many different Ultrametrics are possible on a space • Out of all Ultrametrics such that: u(x,y) ≤ d(x,y) the greatest is called the Subdominant Ultrametric which is unique and can be determined by a Minimal Spanning Tree.
Minimal Spanning Tree Tree: A connected graph without cycles is called a tree. Spanning Tree: A subgraph that is a tree and reaches out to all vertices of the original graph is called a spanning tree of the graph. Minimal Spanning Tree: Out of all possible spanning trees of a graph the one with minimum total edge weight is called the Minimal Spanning Tree of the graph.
MST in Finance – Equity Market Mantegna, J-P Onnela et. al.
MST in Finance – Hedge Funds Miceli and Susinno
MST and FX Market • Hedge fund profits and stock market returns can be measured in a single currency. • Nothing in the currency market is absolute. • Prices for a currency are quoted relative to another, usually USD. • How do we build the tree without missing out any currency?
Data Description • We look at XAU and 10 currencies USD, CAD, GBP, DEM, CHF, SEK, NOK, AUD, NZD and JPY from Jan 1993 to Dec 1994. • Thus we have hourly data points for 10 time series of the form USD/X. • We expand this set to all time series Xi/Xj possible in this group. • This gives us 110 different time series, with every currency represented in the network.
Spurious Correlations? Triangle Effect Correlation of returns:
Comparison of Real and Random Trees Intersection of real and random MST for 1993-94 Currency MST for 1993-94
1.00 0.90 0.82 100 200 300 400 500 0 Stability of MST Single step survival ratio dt
Multi-step Survival of Links 1.0 0 3000 4000 4500 1000 1500 2000 2500 500 0
Work in progress • We are currently applying this analysis to higher frequency data (5 min, tick data). We hope this will give us a real time picture of the market and indicate the currencies “in play”. • We are also investigating the effect of market news, both expected and unexpected, on the currency trees.