I have recently been working with Jochen Papenbrock from Firamis on a new module called CANeM - Cross Asset Network Monitor. The objective of the work is to develop methods for quick identification of price driving themes and market dynamics, and the discovery of anomalies from any market data via network analysis.
The methodology involves first calculating pairwise correlations from the data and then using various techniques to filter out the most important links from the correlation matrix. These networks, which may form time series, are then visualized by means of layout algorithms that best exhibit the properties of the data. The ‘Correlation Networks with FNA – Tutorial‘ shows you in detail how to use FNA and CANeM to create them yourself.
For example, the picture below (see interactive time series) visualizes dependencies in a network of stocks in the German stock index DAX. Each network is constructed from correlations among the stocks during the last 100 trading days (the data is delayed by one trading day and updated daily).
(red) – financial stocks such as Allianz and Deutsche Bank
(green) – health care stocks such as Bayer and Fresenius
(blue) – materials and chemical stocks such as Linde and BASF
(gray) – energy stocks such as RWE and E.ON
(yellow) – automotive and consumables stocks such as Daimler
(purple) – consumables such as Adidas-Salomon
(orange) – transport and communications stocks such as Lufthansa and Deutsche Post
(magenta) – technology stocks such as SAP and Infineon
Negative correlations are displayed in red and positive in black. The width of the line scales with correlation: thin low vs thick high. Node size scales with its degree calculated from the network structure.
In interpreting the chart one can start by looking at the position of each stock in the network and at the number of links each stock has. Stocks in the outer branches are rather decoupled from the market dynamics whereas stocks in the network centre are dependency hubs. The more links a stock has, the more central role it plays in the dynamics. This is marked in the chart as a larger node size. The thickness of the links between nodes scales with correlation (a thick line represents either high positive or high negative correlation). The link lengths are defined by layout constraints and convey no economic meaning in this example.
The visualization is interactive. You can move back with the slider or the arrow buttons to see how the network looked on previous days. Nodes can be grabbed and moved around. Hovering over nodes and links shows their values and its posible to zoom in (with mouse wheel) to look at details.
The CANeM demo page includes more examples for interested readers.
