|
- Random Matrix Theory Filters and Currency Portfolio Optimisation - Multiscaled Cross-Correlation Dynamics in Complex Dynamical Systems
Random matrix theory filters and currency portfolio optimisation PhD Student: Justin DALY Co supervised by Dr. Martin Crane and Prof. Heather Ruskin To see Justin's poster, on his work up to January 2009, follow the link: Justin Daly's POSTER 09. Project description: Random matrix theory (RMT) filters, applied to covariance matrices of financial returns, have recently been shown to offer improvements to the optimisation of financial portfolios. This work studies the effect of three RMT filters on realised portfolio risk, using bootstrap analysis and out-of-sample testing, in the case of a typical foreign exchange and commodity portfolio, weighted towards foreign exchange, and consisting of N = 39 assets. This is intended to test the limits of RMT filtering, which is more obviously applicable to portfolios with large numbers of assets. Taken as a whole, our results suggest that RMT filtering can provide risk reduction for foreign exchange portfolios involving suffcient numbers of assets. Moreover, RMT filters uncovered different uses of models than were possible with unfiltered analysis, namely ones that reacted quickly to market events. Without filtering these features, which utilise very recent data, were found to be masked by noise. Multiscaled Cross-Correlation Dynamics in Complex Dynamical Systems PhD Student: Tom CONLON Co supervised by Prof. Heather Ruskin and Dr. Martin Crane To see Tom's poster, presented at Sci-Sym Center first meeting, January 2009, follow the link: Tom's Conlon POSTER 09 Project description: The equal-time cross-correlation matrix was explored as a mechanism to detect frequency dependent changes in the synchronisation structure between components of multivariate Complex Systems. To this end, we examined both small and large eigenvalues dynamics of the correlation matrix for time series from EEG seizures and the Stock Marke, with contrasting results. For EEG epilectic seizure data, the method was shown to detecgt subtle changes in the synchronisation tructure across frquencies. However, for Equity returns, the results were less clear with covariation across frequencies. It is possible, however, that non-linear synchronisation methods, such as Mutual Information, would have better predictive ability in case of stock market returns |
