Presentation held at the 5th Risk Summit organized by Center for Risk Studies at the University of Cambridge. See http://www.risk.jbs.cam.ac.uk/news/events/risksummits/risksummit2014.html
Nirala Nagar / Cheap Call Girls In Lucknow Phone No 9548273370 Elite Escort S...
Financial Cartography
1. FINANCIAL CARTOGRAPHY
1
23 June 2014
!
Risk Summit at Center for Risk Studies
The Pulse of Risk: From Big Data to Business Value
!
!
Dr. Kimmo Soramäki
Founder and CEO, FNA Ltd.
10. 10
Correlation Maps
Difficult to understand large-scale
correlation or other dependence
structures.!
!
Especially time series.!
!
How to filter signal from noise?!
!
How to put the correlations and
their changes in context with
changes/returns and volatility?!
!
!
Objective is to efficiently
represent a complex system!
…"
11. 11
Significant Correlations
Common method to visualize
large correlation matrices is
with heat maps.!
!
!
!
!
!
If we only keep statistically
significant correlations with
95% confidence level, the
resulting matrix is sparse (with
short time periods).
All correlations
(last 100 days)!
Statistically
significant
correlations
(last 100 days)!
13. 13
Correlation Networks
A sparse matrix is often well
represented as a network. !
!
We encode correlations as links
between the correlated nodes/
assets.!
!
!
Red link = negative correlation
Black link = positive correlation!
!
!
Absence of link marks that
asset is not significantly
correlated.!
14. 14
Dimensionality Reduction & Filtering
Next, we identify the Minimum
Spanning Tree (MST) and filter
out other correlations.!
!
Rosario Mantegna (1999)
‘Hierarchical Structure in
Financial Markets’
!
This shows us the backbone
correlation structure where
each asset is connected with
the asset with which its
correlation is strongest.
15. 15
Coordinate System
We use a radial tree layout
algorithm (Bachmaier &
Brandes 2005) that places the
assets so that:!
!
• Shorter links in the tree
indicate higher correlations!
!
• Longer links indicate lower
correlations!
!
As a result, we also see how
the assets cluster (analogous
to single linkage clustering).!
16. 16
Encoding non-spatial data
Node color indicates last daily
return!
!
Green = positive!
!
Red = negative!
!
Node size indicates magnitude
of return!
!
!
17. 17
“Here be Dragons”
Sornette’s Dragon King:
“Extreme events can be
predicted”!
!
Mandelbrot’s Volatility
Clustering: “Large changes tend
to be followed by large changes”!
!
!
-> Identify VaR exceptions
(return outside 95% VaR
bounds)!
!
-> Map them as bright green or
red nodes!
Track the number of outliers
each day
Highlight outliers in their context
18. 18
Correlations Maps
!
Dimensionality Reduction & Filtering!
-> Minimum Spanning Tree!
!
!
Coordinate System!
-> Radial Tree layout algorithm (correlation
distances)!
!
!
System for visual encoding of (non-spatial) data!
-> Returns and Outliers!
!
!