Workplan

The macroscopic behaviour of financial markets is determined by the microscopic behaviour of individual agents within the
system. The emergence of dynamical behaviour which leads to market crashes at a macro level may therefore be detected at
a micro level through changes in the network structure of the market. Our research examines the direct causal connections
between the price movements of stocks from 8 different sectors of the US economy in the lead up to the 2008 stock market
crash. Using a mutual information measure called Partial Mutual Information at Mixed Embedding, which can detect both linear
and nonlinear causal connections, we detect a decrease in network density beginning as early as 2006 and continuing
throughout the 2008 crash. Moreover, observing changes in the density of connections from each sector to the rest of the
network we can detect distinct differences between sectors. In particular we note a reduction in strength of the causal
connections from the Financial, Consumer Goods and Industrial sectors following a peak in early 2006.
Micro level the individual elements of many complex systems are known to cluster into groups or “herds”. It is often the
dynamical behaviour and interactions of these herds which lead to the large scale macro behaviour of the system. The results
of our previous research indicate that stocks in different sectors of the U.S. economy may display heterogeneous behaviour in the lead up to the 2008 stock market crash. We propose to develop this result by testing for clustering of individual stocks in
the market. Furthermore we will use the results of our empirical testing to inform an agent based model of financial markets
which can capture this clustering behaviour and the dynamics which emerge at a macro level including financial market
crashes caused by phase transitions.
Specific attention will be given to the visual presentation of models and their results.