Slides from a one-day training held for 11 Caribbean and Latin American central banks at Bank of Jamaica, Kingston.

1. WORKSHOP ON PAYMENT SYSTEMS OVERSIGHT
CEMLA - Bank of Jamaica
Kingston, Jamaica, December 5-7, 2012
Tools for Quantitative Oversight of
Market Infrastructures
Dr. Kimmo Soramäki
Founder and CEO
FNA, www.fna.fi

2. Agenda
Part I: Visualization
• Financial Cartography
• Getting Started with FNA Platform
• Tutorial 1 – Loading Networks into FNA
• Tutorial 2 – Managing Data in FNA
• Tutorial 6 – Network Visualization
Part II: Analytics
• Oversight Metrics
• Oversight Monitor -application
• Tutorial 3 – Network Summary Measures
• Tutorial 5 – Connectedness and Components
• Tutorial 4 – Centrality Measures
Part III: Simulations
• Payment System Simulations and Stress Tests
• Payment System Simulator -application
• Tutorial 8 – Payment System Simulations
2

3. Part I: Visualization
Financial Cartography

4. ―When the crisis came, the serious limitations of existing
economic and financial models immediately became apparent.
[...]
As a policy-maker during the crisis, I found the available
models of limited help. In fact, I would go further: in the face of
the crisis, we felt abandoned by conventional tools.‖
in a Speech by Jean-Claude Trichet, President of the
European Central Bank, Frankfurt, 18 November 2010
4

5. We did not have maps …
5

6. Eratosthenes' map of the known world6
c. 194 BC

7. … but what are maps
―A set of points, lines, and areas all defined both by position with
reference to a coordinate system and by their non-spatial attributes‖
Data is encoded as size, shape, value, texture or pattern, color and
orientation of the points, lines and areas – everything has a meaning
Cartographer selects only the information that is essential to fulfill the
purpose of the map
Maps reduce multidimensional data into a two dimensional space that
is better understood by humans
Maps are intelligence amplification, they aid in decision making and
build intuition
7

8. I. Mapping II. Mapping
Systemic Risk Financial Markets
8

9. Systemic risk ≠ systematic risk
News articles mentioning ―systemic risk‖, Source: trends.google.com
The risk that a system composed of many interacting
parts fails (due to a shock to some of its parts).
In Finance, the risk that a disturbance in the financial
system propagates and makes the system unable to
perform its function – i.e. allocate capital efficiently.
Not:
Domino effects, cascading failures, financial
interlinkages, … -> i.e. a process in the
financial network
9

10. First Maps Fedwire Interbank Payment
Network, Fall 2001
Around 8000 banks, 66 banks
comprise 75% of value,25 banks
completely connected
Similar to other socio-
technological networks
Soramäki, Bech, Beyeler, Glass and Arnold (2007), M. Boss, H. Elsinger, M. Summer, S. Thurner, The
Physica A, Vol. 379, pp 317-333. network topology of the interbank market, Santa
See: www.fna.fi/papers/physa2007sbagb.pdf Fe Institute Working Paper 03-
10
10-054, 2003.

11. This is still shocking …
―In 2006, the Federal Reserve invited a group of researchers to
study the connections between banks by analyzing data from the
Fedwire system, which the banks use to back one another up.
What they discovered was shocking: Just sixty-six banks — out of
thousands — accounted for 75 percent of all the transfers. And
twenty five of these were completely interconnected to one
another, including a firm you may have heard of called Lehman
Brothers.‖
Want to Build Resilience? Kill the Complexity
Harvard Business Review Blogs, 9/2012
11

12. Interbank payment networks
Becher, Millard and Soramäki (2008). Agnes Lubloy (2006). Topology of the Hungarian
The network topology of CHAPS large-value transfer system. Magyar Nemzeti Bank
Sterling. BoE Working Paper No. 355. Occasional Papers
Embree and Roberts (2009). Network
Analysis and Canada's Large Value Transfer
SystemBoC Discussion Paper 2009-13
12

13. Overnight lending networks
Federal funds
Bech, M.L. and Atalay, E. (2008), “The Topology of
the Federal Funds Market”. ECB Working Paper No. 986.
Italian money market
Iori G, G de Masi, O Precup, G Gabbi and G
Caldarelli (2008): “A network analysis of the Italian
overnight money market”, Journal of Economic
Dynamics and Control, vol. 32(1), pages 259-278
Unsecured Sterling
money market
Wetherilt, A. P. Zimmerman, and K. Soramäki
(2008), “The sterling unsecured loan market
during 2006–2008: insights from network
topology“, in Leinonen (ed), BoF Scientific
monographs, E 42
Minoiu, Camelia and Reyes, Javier A. (2010). A network analysis of global 13
Cross-border bank lending banking:1978-2009. IMF Working Paper WP/11/74.

14. Network Theory can be to Financial Maps
what Cartography is to Geographic Maps
Main premise of network theory:
Structure of links between nodes
matters
To understand the behavior of one
node, one must analyze the
behavior of nodes that may be
several links apart in the network
Topics: Centrality, Communities,
Layouts, Spreading and generation
processes, Path finding, etc.
14

15. Centrality Measures for
Financial Systems
• Traditional
– Degree, Closeness, Betweenness
centrality, PageRank, etc.
• DebtRank
– Battiston et al, Nature Science
Reports, 2012
– Feedback-centrality
– Solvency cascade
• SinkRank
– Soramäki and Cook, Kiel
Economics DP, 2012
– Transfer along walks
– Liquidity absorption
15

16. I. Mapping II. Mapping
Systemic Risk Financial Markets
16

17. Outline
Purpose of the maps
– Identify market dynamics
– Reduce complexity
– Spot anomalies
– Build intuition
The maps: Heat Maps, Trees, Networks
and Sammon‘s Projections
Based on asset correlations or tail
dependence
These methods are showcased for
visualizing markets around the collapse
of Lehman brothers
17

18. The Case
Lehman was the fourth largest investment bank in the US (behind
Goldman Sachs, Morgan Stanley, and Merrill Lynch) with 26.000
employees
At bankruptcy Lehman had $750 billion debt and $639 billion assets
Collapse was due to losses in subprime holdings and inability to find
funding due to extreme market conditions
Is seen as a divisive point in the 2007-2009 financial crisis
We create 3 visualization of a 5 month period around the failure (15
September 2008) from asset price data
18

19. The Data
Pairwise correlations of
return on 141 global
assets in 5 asset classes
9870 data points per
time interval
5 intervals, 2 months
before and 3 months
after Lehman collapse
19

20. i) Heat Maps
2004-2007
Corporate
Bonds
CDS on
Government
Debt
FX Rates
Government
Bond Yields
Correlation
-1
Stock
Exchange 0
Indices
+1
20

21. Collapse of Lehman, t=month
2004-2007 t-2 t-1
t+1 t+2 t+3

22. ii) Asset Trees
Originally proposed by Rosario Mantegna in 1999
Used currently by some major financial institutions
for market analysis and portfolio optimization and
visualization
Methodology in a nutshell MST
1. Calculate (daily) asset returns
2. Calculate pairwise Pearson correlations of returns
3. Convert correlations to distances
4. Extract Minimum Spanning Tree (MST)
5. Visualize (as phylogenetic trees)
22

23. Demo
Click here for interactive visualization 23

24. Correlation filtering PMFG
Balance between too much and too little
information
One of many methods to create networks
from correlation/distance matrices
– PMFGs, Partial Correlation Networks,
Influence Networks, Granger Causality, Influence Network
NETS, etc.
New graph, information-theory, economics
& statistics -based models are being
actively developed
24

25. iii) NETS
• Network Estimation for Time-
Series
• Forthcoming paper by Barigozzi
and Brownlees
• Estimates an unknown network
structure from multivariate data
• Captures both comtemporenous
and serial dependence (partial
correlations and lead/lag effects)
25

26. iv) Sammon‘s Projection
Proposed by John W. Sammon in IEEE Transactions on Computers 18: 401–409
(1969)
A nonlinear projection method to map a
high dimensional space onto a space of
lower dimensionality. Example:
Iris Setosa
Iris Versicolor
Iris Virginica
26

27. Demo
Click here for interactive visualization 27

28. Extensions
• Correlation is a linear dependence. The same visual maps can be extended
to non-linear dependences.
• Instead of correlation, links and positions measure similarity of distances to
tail losses
• Instead of returns, links can be based on bank balance sheet item, portfolio,
etc. co-movements
Tail Tree Tail Sammon
(Click here for interactive visualization) (click here for interactive visualization)28

29. ―In the absence of clear guidance from existing analytical
frameworks, policy-makers had to place particular reliance on
our experience. Judgment and experience inevitably played a
key role.‖
in a Speech by Jean-Claude Trichet, President of the
European Central Bank, Frankfurt, 18 November 2010
29

30. Part II: Analytics
Network Metrics for Oversight

31. Standard reporting
• System turnover (value, • Intraday statistics
– Payments value over day
volume)
– Intraday pattern/throughput
– Daily (value/volume)
– Monthly peak/low/average – Delays due to lack of liquidity
– Yearly total
– Unsettled payments • Technical
– Distribution – Processing times
– Settlement mode (by algorithm)
• Individual payments • Static information
– Average/min/max value – Number/types of participants
– Value distribution – Opening/closing balances
– Intraday credit limits
– Payment type breakdown
– Bilateral limits
(interbank, ancillary, cb
operations, etc)
• Incident reports
– Priority (urgent, normal)
– Breakdown by bank
31

32. Monitoring Indicators for Intraday Liquidity
Management
Consultative report by BCBS. Final document expected 1Q 2013
―.. the indicators are also likely to be of benefit to overseers of payment
and settlement systems. Close cooperation between banking supervisors
and the overseers is envisaged.―
Most indicators can be calculated from payments data
(i) Daily maximum liquidity requirement usage √
(ii) Available intraday liquidity √ (partly possible)
(iii) Total payments (sent and received) √
(iv) Time-specific and other critical obligations √ (partly possible)
(v) Value of customer payments made on behalf √ (partly possible)
of financial institution customers
(vi) Intraday credit lines extended to financial X
institution customers
(vii) Timing of intraday payments √
(viii) Intraday throughput √
32

33. (i) Daily maximum liquidity requirement usage
33

34. Network Metrics
Payment Systems are ―Complex Adaptive Systems‖
A bank‘s ability to settle payments (its liquidity risk) depends on its
available liquidity and other banks ability to settle payments, which
depend …
Galbiati and Soramäki (2011), An Agent based Model of
Payment Systems. Journal of Economic Dynamics and
Control, Vol. 35, Iss. 6, pp 859-875

35. Network Theory
Financial
Network Analysis
Social Network
Network Science
Analysis
NETWORK
THEORY
Graph & Matrix Computer
Theory Science
Biological
Network Analysis

36. Network basics
• Terminology
– node/vertex -> Bank/banking group, Asset
– link/tie/edge/arc -> Financial interlinkages, bilateral positions, exposures
– directed vs undirected
– weighed vs unweighted
– graph + properties = network 2
1
• Algorithms/measures
3 4
– Centrality -> Systemic importance
– Flow -> Liquidity
– Community/pattern identification -> Core-Periphery (Craig – Von Peter)
– Distance, shortest paths -> Liquidity absorption, SinkRank
– Connectivity, clustering
– Cascades, epidemic spreading -> Contagion
36

37. Network analysis for Oversight
• Network maps: intuitive, provide a deeper understanding of the
system via anomaly explanation and visualization
• Centrality metrics: such as Pagerank and SinkRank can be used as a
proxy for systemic importance, contagious links
• Monitor over time: build reference data, detect and understand
gradual change
• Tied to availability of data: enables ―Analytics based policy‖, i.e. the
application of computer technology, operational research, and
statistics to solve regulatory problems

38. Research
• A growing body of empirical research on financial networks
• Interbank payment flows
– Soramäki et al (2006), Becher et al. (2008), Boss et al. (2008), Pröpper et al. (2009),
Embree and Roberts (2009), Akram and Christophersen (2010) …
• Overnight loans networks
– Atalay and Bech (2008), Bech and Bonde (2009), Wetherilt et al. (2009), Iori et al. (2008)
and Heijmans et al. (2010), Craig & von Peter (2010) …
• Flow of funds, Credit registry, Stock trading, Markets, …
– Castren and Kavonius (2009), Bastos e Santos and Cont (2010), Garrett et al. 2011, Minoiu
and Reyes (2011), (Adamic et al. 2009, Jiang and Zhou 2011), Langfield, Liu and Ota (2012)
• More at www.fna.fi/blog

39. Common centrality metrics
Centrality metrics aim to summarize some notion of importance
Degree: number of links
Closeness: distance from/to other
nodes via shortest paths
Betweenness: number of shortest
paths going through the node
Eigenvector: nodes that are linked by
other important nodes are more central,
visiting probability of a random process

40. Eigenvector Centrality
Problem: EVC can be (meaningfully) calculated only for ―Giant
Strongly Connected Component‖ (GSCC)
Solution: PageRank

41. PageRank
• Algorithm used by Google to rank web pages. Random surfer model.
• Solves the problem of dead-ends with a ―Damping factor‖ which is
used to modify the adjacency matrix (S)
– Gi,j= Si,j
• Effectively allowing the random process out of
dead-ends (dangling nodes), but at the cost of
introducing error
• Effect of
– Centrality of each node is 1/N
– Eigenvector Centrality
– Commonly is used

42. Which measure to calculate?
Depends on process that takes place in the network!
Trajectory Transmission
– Geodesic paths (shortest paths) – Parallel duplication
– Any path (visit no node twice) – Serial duplication
– Trails (visit no link twice) – Transfer
– Walks (free movement)
Source: Borgatti (2004)

43. Systemic Risk in Payment Systems
• Credit risk has been virtually eliminated
by system design (real-time gross
settlement)
• Liquidity risk remains
– ―Congestion‖
– ―Liquidity Dislocation‖
• Trigger may be
– Operational/IT event
– Liquidity event
– Solvency event
• Time scale is intraday, spillovers possible

44. SinkRank
SinkRanks on unweighed
networks
• Soramäki and Cook (2012),
―Algorithm for identifying
systemically important banks in
payment systems‖
• Measures how big of a ―sink‖ a bank is
in a payment system
• Based on theory of absorbing markov
chains: average transfer distance to a
node via (weighted) walks from other
nodes
• Provides a baseline scenario of no
behavioral changes by banks
• Allows also the identification of most
vulnerable banks

45. Distance to Sink
Absorbing Markov Chains give distances:
From B 1
To A
From C 2
(66.6%) (100%) From A
To B
From C 1
(33.3%)
From A
To C
From B
(100%)

46. SinkRank
SinkRank is the average distance of a unit of liquidity to the sink
Actual liquidity distribution can be used in calculating SinkRank
Uniform PageRank “Real”
(A,B,C: 33.3% ) (A: 37.5% B: 37.5% C:25%) (A: 5% B: 90% C:5%)
Note: Node sizes scale with 1/SinkRank

47. How good is it? Experiments:
• Design issues
– Real vs artificial networks?
– Real vs simulated failures?
– How to measure disruption?
• Approach taken
1. Create artificial data with close resemblance to the US Fedwire
system (BA-type, Soramäki et al 2007)
2. Simulate failure of a bank: the bank can only receive but not send
any payments for the whole day
3. Measure ―Liquidity Dislocation‖ and ―Congestion‖ by non-failing
banks
4. Correlate 3. (the ―Disruption‖) with SinkRank of the failing bank

48. SinkRank vs Disruption
Relationship between
SinkRank and Disruption
Highest disruption by
banks who absorb
liquidity quickly from the
system (low SinkRank)

49. Distance from Sink vs Disruption
Relationship between
Failure Distance and
Disruption when the most
central bank fails
Highest disruption to
banks whose liquidity is
absorbed first (low
Distance to Sink)
Distance to Sink

50. To sum up
• Existing centrality measure do not accurately reflect the
process of payment systems
• SinkRank accurately predicts the magnitude of disruption
caused by the failure of a bank in a payment system and
identifies banks most affected by the failure.
• SinkRank is based on absorbing Markov chains, which are
well-suited to model liquidity dynamics in payment systems.
• We find that the failing bank‘s SinkRank is highly correlated
with the resulting disruption in the system overall
• Finally, we present software that implements SinkRank in a
payment system simulation environment

51. Side note: Data generation process
Based on extending Barabasi–
Albert model of growth and
preferential attachment

52. Network Analysis Tools
• Pajek, Universty of Ljublana, Slovenia
– Focus on social network analysis of large networks
– pajek.imfm.si
• Gephi, Gephi Foundation, France
– Focus on graph visualisation ―Like Photoshop for graphs‖
– www.gephi.org
• FNA, Soramaki Networks, Finland
– Focus on Financial/Payment Networks/Simulation and interactive visualization
– www.fna.fi
• Many others
– Cytoscape, Graphviz, Network Workbench, NodeXL, ORA, Tableau, Ucinet,
Visone, etc.

53. Where are we today?
Regulatory response to recent financial crisis
was to strengthen macro-prudential
supervision with mandates for more
regulatory data
―Big data‖ and ―Complex Data‖-> Challenge
to understand, utilize and operationalize the
data
(network is fictional)
Promise of ―Analytics based policy and
regulation‖, i.e. the application of computer
technology, operations research, and Example: Oversight Monitor at Norges Bank
statistics to support human decision making
The monitor will allow the identification of
Growing body of empirical research, see systemically important banks and evaluation of
the impact of bank failures on the system
www.fna.fi/library
53

54. FNA Oversight Monitor
• Allow identification of systemically important banks
– Values/volumes
– SinkRank
• Allows a network view to payment systems
– Liquidity flows
– Throughput
• Allows visualization of network and other statistics
– Statistics can be calculated in real-time or at regular intervals directly
from raw payment data
– The default indicators consist of metrics proposed in the BIS/BCBS
report on "Monitoring indicators for intraday liquidity management―
• Can help in crisis management
54

55. Demo: FNA Oversight Monitor
Try it at www.fna.fi 55

56. Part III: Simulations
LSM and Stress Analysis

57. What are simulations?
• Methodology to understand complex systems – systems that are
large with many interacting elements and or non-linearities (such as
payment systems)
• In contrast to traditional statistical models, which attempt to
find analytical solutions
• Usually a special purpose computer program is used that takes
granular inputs, applies the simulation rules and generates outputs
• Take into account second rounds effects, third round, …
• Inputs can be stochastic or deterministic. Behavior can be static,
pre-programmed, evolving or co-learning

58. Short history of LVPS simulations
• 1997 : Bank of Finland
– Evaluate liquidity needs of banks when Finland‘s RTGS system was joined with TARGET
– See Koponen-Soramaki (1998) ―Liquidity needs in a modern interbank payment systems:
• 2000 : Bank of Japan and FRBNY
– Test features for BoJ-Net/Fedwire
• 2001 - : CLS approval process and ongoing oversight
– Test CLS risk management
– Evaluate settlement‘ members capacity for pay-ins
– Understand how the system works
• Since: Bank of Canada, Banque de France, Nederlandsche Bank, Norges Bank,
TARGET2, and many others
• 2010 - : Bank of England new CHAPS
– Evaluate alternative liquidity saving mechanisms
– Use as platform for discussions with banks
– Denby-McLafferty (2012) ―Liquidity Saving in CHAPS: A Simulation Study‖

59. Framework
Source: Koponen-Soramäki (1997). Intraday liquidity needs in a modern interbank payment system - a
Simulation Approach , Bank of Finland Studies in Economics and Finance 14.

60. Application Areas
Enhance
Evaluate alternative
understanding of
design features
system mechanics
Why
Simulate?
Stress testing and Platform for
liquidity needs communication
analysis among stakeholders

61. Data needs for simulations
• Historical transaction data
– From interbank payment systems
– At minimum: date, time, sender, receiver, value
– More data on type of payment, economic purpose,
second tier (if any), type of institution, etc. useful
• Artificial transaction data
– Based on aggregates (possible with Entropy
maximization methods)
– Based on a network model (defining bilateral flows)
– Assumptions
• Timing of payments
• Value distribution
• Correlations
– System stability (net flows over longer times)

62. Tools
• Bof-PSS2
– Bank of Finland, 1997- (BoF-PSS1)
– RTGS, RRGS, Net, many optimization methods
– www.bof.fi/sc/bof-pss
– Free, Support & training available, Annual
workshop
• FNA
– Soramaki Networks, 2009-
– RTGS, RRGS, many optimization methods,
visual exploration of results, network analysis
– www.fna.fi
– Free online, License, support & training
available
• Proprietary tools or general purpose programs
– Matlab, SAS, Excel, …

63. FNA Payment Simulator
• Allow testing of Liquidity saving mechanisms
– Queuing (FIFO + priorities + bypass)
– Bilateral limits, overdraft limits, opening balances
– Two-stream operation
– Bilateral offsetting (first, fifo, best)
– Queue optimization (Bech-Soramaki)
• Allow simulation of Stress scenarios, such as the scenarios in BCBS
document ―Monitoring Indicators for Intraday Liquidity Management‖
– (i) Own financial stress
– (ii) Counterparty stress
– (iii) Customer stress
– (iv) Market wide credit or liquidity stress
• Any functionality can be implemented in a custom Payment Simulator -
application
63

64. Demo: FNA Payment Simulator
64

65. FNA Platform
• Go to www.fna.fi
• Register account
(click login on top right)
• Watch ‗Getting started with
FNA‘ video
• More documentation available
at www.fna.fi/gettingstarted
65

66. Getting Started: Commands
• FNA operates on commands that are submitted to FNA server for
execution. Commands explore the database, alter it or create
visualizations from it
• Command syntax:
commandname –parameter1 value1 –parameter2 value2 …
e.g.
loada -file sample-arcs.csv -preserve false
(load arcs from sample-arcs.csv file and don‘t preserve any existing networks in database)
• Each command is on a single line. Character # marks a comment
line
• Commands can be bundled into scripts and executed in one go
66

67. Data Model
loada -file sample-arcs.csv -preserve false
sample-arcs.csv
network,source,target,value
2005-1Q,Australia,Austria,499
2005-1Q,Australia,Belgium,1135 Stores the data into
2005-1Q,Australia,Canada,1884 a graph database
... on FNA Server
net_id : 2005-1Q
arc_id : Australia-Austria vertex_id : Austria
value : 499
…
vertex_id : Australia vertex_id : Belgium
…
vertex_id : Canada
67

68. CSV files
Comment rows
# The data shows banking system exposures to particular countries. (1 and d 2)
# See http://www.bis.org/publ/qtrpdf/r_qt1006y.htm for details.
Empty row (3)
Header field (with
net_id,arc_id,from_id,to_id,value
value ‘net_id’)
2010-1Q,Australia-Austria,Australia,Austria,211 Header row (4)
2010-1Q,Australia-Belgium,Australia,Belgium,1128
2010-1Q,Australia-Canada,Australia,Canada,12231
2010-1Q,Australia-Chile,Australia,Chile,335
2010-1Q,Australia-France,Australia,France,8865
Field (with value
2010-1Q,Australia-Germany,Australia,Germany,11702
‘2010-1Q’)
2010-1Q,Australia-Greece,Australia,Greece,12
2010-1Q,Australia-India,Australia,India,2180
2010-1Q,Australia-Ireland,Australia,Ireland,3583 Data rows (5-)
2010-1Q,Australia-Italy,Australia,Italy,8657
2010-1Q,Australia-Japan,Australia,Japan,4035
2010-1Q,Australia-Netherlands,Australia,Netherlands,5970
2010-1Q,Australia-Portugal,Australia,Portugal,776
2010-1Q,Australia-Spain,Australia,Spain,2776
2010-1Q,Australia-Sweden,Australia,Sweden,773
2010-1Q,Australia-Switzerland,Australia,Switzerland,3366
2010-1Q,Australia-Turkey,Australia,Turkey,63
Column Delimiter (,)

69. Tutorial I – Loading Networks into FNA
Try:
loada -file sample-arcs.csv -preserve false
viz
69

70. Blog, Library and Demos at www.fna.fi
Dr. Kimmo Soramäki
kimmo@soramaki.net
Twitter: soramaki