The multilayer structure of the financial system

Richard M. Bookstabera,1 , Dror Y. Kenettb,2,∗

a
Office of the Chief Investment Officer, Regents of the University of California
b
U.S. Department of Treasury, Office of Financial Research

Abstract
We introduce a multilayer network as a framework for analyzing the emer-
gence and propagation of risk within the financial system. The layers of the
network encompass assets, funding, and collateral. Various entities in the
financial system occupy these layers. For example, asset managers occupy
the asset layer, and central counterparties occupy the collateral layer. Some
entities span layers. For example, leveraged managers such as hedge funds
span the asset and funding layers. Banks are notable in spanning all three
layers, and thus are central to the spreading of risks. We show that a multi-
layer network structure presents risk characteristics that differ from those of
a single layer network, and can generate more extensive vulnerabilities and
more abrupt, multi-stage cascades than appear in a single-layer network.
Keywords: Financial system, Multilayer network, Financial stability
JEL: G01, G02, C02

1. Introduction
The financial system is built upon a complex set of interdependencies
spanning funding flows, counterparty and credit relationships, and asset price
dynamics, all passing through many distinct and heterogeneous institutions.
A stress to one part of the system can spread to others, often threatening the

∗
Corresponding author email: [Link]@[Link]
1
This paper was produced while Richard Bookstaber was employed by the Office of
Financial Research.
2
The views and opinions expressed are those of the individual authors and do not
necessarily represent official positions or policy of the Office of Financial Research or the
U.S. Treasury.

Preprint submitted to Journal October 25, 2015

stability of the entire financial system. What is more, a stress in one function
of the system can spread to affect another function. Stresses in funding
might affect assets, and stresses in assets might in turn affect collateral and
counterparty risk. Stresses in a Bank/Dealer might spread to affect a hedge
fund, and stresses in the hedge fund might affect a central counterparty.
For example, a Bank/Dealer facing a shortfall in funding might reduce the
lending it provides to hedge funds, and to control their risk the hedge funds
might respond by liquidating asset positions, with the resulting drop in prices
compromising collateral values. Thus the flow of funding from one node
to another affects the asset flows. Similarly, if asset prices suddenly drop,
collateral flows will be affected, which in turn will then change available
funding.
The key issue that makes the analysis of systemic risk difficult in a net-
work setting is not so much the size or complexity of the network, but these
varied characteristics and functions of the nodes, and the many facets of the
flows from one node to another. What is more, the flows do not simply move
from one institution to another; the institutions take the funding, credit,
and asset flows and transform them in various ways. There are maturity
transformations, the standard banking function of taking in short term de-
posits and making longer-maturity loans; credit transformations through the
repackaging of debt instruments as structured products, for example where
assets such as mortgages are broken up into tranches of varying credit risk;
liquidity transformations, where less liquid assets are restructured and sup-
ported by market-making, such as ETFs which provide intra-day liquidity to
less liquid assets; and risk transformations, where the return distribution of
assets is changed, such as by issuing options with an asset as the underly-
ing. Any analysis of systemic risk that does not account for this multifaceted
characteristic of the financial network will fail to track the dynamics of the
process.
Network science has been found to provide valuable insights in different
scopes in the financial system: descriptions of systemic structure, analysis
and evaluation of the penetration or contagion effects (Summer, 2013; Lillo,
2010; Kenett et al., 2012a,c; Cont, 2013; Glasserman and Young, 2015; Garas
et al., 2010); studies that assess the impact of the insolvency of one or a
particular group of actors in the system, depending on its relevance and
connectivity within the structure (Jackson, 2010; Battiston et al., 2012);
studies that allow to evaluate the impact of liquidity problems at specific
times and initiated in different nodes of the system (Haldane and May, 2011;

2
Haldane et al., 2009; Cont et al., 2010; Amini et al., 2012; Kenett et al., 2010,
2012b); the structure of interbank exposure networks (Boss et al., 2004, 2006;
Elsinger, 2009; Hüser, 2015). Considering the problem of contagion, Allen
and Gale (1998) study how shocks can spread in the banking system when
it is structured in the from of a network; Drehmann and Tarashev (2013)
develop a measure that captures the importance of an institution, in term of
its systemic relevance, in the propagation of a shock in the banking system;
Acemoglu et al. (2013c,b,a) develop a model of a financial network through
its liability structure (interbank loans) and conclude that complete networks
guarantee eficiency and stability, but that when negative shocks are larger
than a certain threshold, the effects of contagion prevail, leading to systemic
instability.
The manifest characteristic of this wide range of inquiry is that the fi-
nancial system is depicted as a single network, with nodes operating in a
homogenous fashion and with the links between the nodes treating one type
of flow. Of course, in fact the financial system has many types of agents,
and the flows that are relevant differ from one agent to another, and even
differ within the same agent from one activity to another. Thus, the finan-
cial system operates on a number of layers, and cannot be represented as a
single-layer network.
In this paper we present a multilayer network depiction of the financial
system. Very little research has been focused on the multilayer network
properties of the financial system, and these have been focused at particular
aspects of the financial system, and not the system as a whole (see Bargigli
et al. (2015); Bravo-Benitez et al. (2014)). The network also has heteroge-
neous nodes or agents that in some cases sit on one of the network layers,
and in other cases span layers. Indeed, the types of agents can be defined by
the nodes in which they operated, whether they are core or peripheral to the
nodes, and whether they are providers, users or intermediaries for the flows
in the various layers on which they operate. The multilayer network has
implications for systemic risk that are qualitatively different than those for a
single layer network. Perhaps not unexpectedly, the mode of propagation of
risk, the path a shock takes, and the value of integration versus segregation
of the functions of various agents or nodes all have a different and richer
nature as we move to a multilayer view of the financial system.

3
2. Mapping the structure of the financial system
The multilayer network encompassing the financial system has specific
agents operating on each layer, and some types of agents spanning between
layers. In particular, we present the financial system as a three-layer network,
with an asset, funding, and collateral layer. In this section we will describe
the nature of the agents in each of the layers within the broad structure of a
map of the key agents and their interactions. For each layer of the network
we will highlight the components of this map that are relevant to that layer.
The asset and collateral layer both interact with the funding layer and so we
will describe that layer first.

2.1. Funding layer

Recently, Aguiar et al. (2014) introduced the first mapping of the flow of
funding in the financial system. The dynamics of the financial system and
the undercurrents of its vulnerabilities rest on the flow of funding, which
are typically represented as a network with banks and financial entities as
the nodes and the funding links as the edges. Aguilar et al. extend this
focus to treat the funding operations within the nodes: the intermediation
and transformation functions provided by the Bank/Dealers (BD) and the
decision heuristics of the funding providers and users. This adds a critical
level of detail about potential funding risks. We use this map as the schematic
for the funding layer of the multilayer network.
The key aspects of this funding map are highlighted within the broader
map in Figure 1. It shows how funding and securities move among key ele-
ments of the bank: the prime broker, which interacts with customers such as
hedge funds; the trading desk, which provides market-making for customers
and hedging for internal risks; the derivatives desk; and the corporate trea-
sury of the Bank/Dealer, which provides the equity and debt issuance. At
the center, where all roads seem to lead, is the Bank/Dealers financing op-
eration. This is where securities purchased or received from counterparties
as collateral are rehypothecated as collateral to obtain funding through the
repurchase (repo) market, and where securities are obtained through reverse
repo and securities lending transactions to fulfill short requirements, provide
financing to clients, or for other internal Bank/Dealer needs (e.g., liquidity
investment).
Furthermore, the funding map provides more detail about the nature of
the Bank/Dealers relationships with its customers with respect to sources

4
5
Figure 1: The funding map of a typical Bank/Dealer and related entities. This provides detail into the internal workings of a
Bank/Dealer as the intermediary for passing funding from the cash providers to the cash users. It highlights the key components
of the entities and the Bank/Dealer functions that are involved in funding. At the center, where all roads seem to lead, is
the Bank/Dealers Financing Operation. This is where funding is obtained, largely through the repurchase (repo)market, and
where securities are obtained through reverse repo and securities lending transactions to fulfill short requirements. See also 1
and uses of funding and securities. The map shows the Bank/Dealer operat-
ing within the interbank market, the standard relationship in most network
approaches to the financial system, and in a broader financial landscape
that includes money market funds, pension funds, hedge funds, and others.
The Bank/Dealer obtains securities to lend to clients and to cover expo-
sures in its own trading operations through many of these same parties. The
Bank/Dealer is also connected to other entities in its role of providing fund-
ing and securities, often to the same types of entities that provide its funding
and securities.
The funding map provides a detailed view of the business activities per-
fromed by financial market participants with a directional display of the
exchange of cash or securities, a representation of the durability of funding
sources, and the illumination of the stress triggers and amplifiers of funding-
related risks between participants, with the objective of understanding the
funding risks within the financial system as a whole and the potential for
contagion given the interrelationship of participants.

2.2. Collateral layer

All flows of secured funding are met by collateral flows in the opposite
direction. Thus, as noted in Aguiar et al. (2014), the funding map is im-
plicitly a collateral map as well. The two-way street of secured funding and
collateral is depicted on Figure 1 with connecting lines that have arrows on
both ends. A network depicting funding flows thus implicitly reveals a net-
work of collateral flows. However, collateral can also be presented as its own
network, depicted by the collateral arrangements, ranging from bilateral, to
tri-party, to centrally cleared; by the transformation of collateral though up-
grades and re-use; and by the risk management of collateral based on haircuts
and quality.
Aguiar et al. (preprint) have performed a detailed mapping of the flow of
collateral in the financial system. In a similar fashion to the funding map,
they charted out the collateral map, focusing on the key entities involved in
collateral transactions, and the dynamics of its flow, and provide insight into
how collateral can contribute to a financial crisis by creating vulnerability
that accelerates and propagates from one institution to another, ultimately
affecting the financial stability. The collateral map shows the path of col-
lateral to its end points, differentiating the paths for bilateral, tri-party, and
CCPs. Each of these paths facilitates different transformations of collateral,
such as reuse, upgrades, and credit transformations.

6
7
Figure 2: Collateral map for a typical Bank/Dealer and related entities. In a similar way to Figure 1,, here we highlight in the
general schematic the typical entities involved in the fl w of collateral, the different flow and use mechanisms involved in the
transfer of collateral. Furthermore, there are different roles played by the entities involved in the transfer of collateral.
 Figure 2 highlights the key components that are part of this collateral
map, which depicts the pathway of collateral across the various agents. This
figure contains more detail than other network analyses in at least two re-
spects. First, it gives detail into the internal workings of a Bank/Dealer and
its central roles in intermediating the flow of collateral. It shows that collat-
eral comes into the Bank/Dealer through a number of channels, and it is then
dispatched through a number of routes: bilateral, tri-party, and CCP. The
Prime Broker is the conduit of collateral from the hedge funds; the Financing
Desk for securities lending and repo; and the Derivatives Desk for futures,
forwards, swaps, options, and related activities. The Bank/Dealers Financ-
ing Operation is the engine for key collateral transformations. It is through
the Financing Desk that collateral is re-used and where collateral upgrades
are managed. Underpinning all of this activity is the collateral management
function at the Bank/Dealer, which dictates the level of collateral and the
quality of collateral that can be used for Securities Financing Transactions
and Derivatives obligations. Second, a key aspect of the collateral map is
the pipelines for collateral flows. Collateral can be passed directly to the
funding agent as a bilateral flow, can be held by a tri-party agent, where all
counterparties have their collateral pooled but where that pooling remains
distinct for each borrower; or can be passed to a CCP, where the collateral
could pass through to other CCP members.

2.3. Asset layer

The third layer is that of the asset network. Although, in light of the
funding- and leverage-induced failures of the 2008 crisis, much of the recent
network literature in finance focuses on the funding network in one from
or another, the asset network is what is more commonly thought of as the
embodiment of the financial system and has historically been the topic of
network representations. Figure 3 presents a canonical asset network. The
asset network includes the flows of assets and cash between asset managers
and the security markets. The network thus includes mutual funds, hedge
funds, the trading and investment arms of insurance companies and pension
funds as peripheral nodes, and the exchanges and market makers as nodes
in the core. In most cases the asset managers do not span to the funding
layer, but as is clear from the funding map of Figure 1, leveraged investors
that use funding and securities lenders that provide funding will be nodes in
the funding layer as well.

8
 Figure 3 presents the asset map in the context of the funding map and
collateral map. The central components of the asset map are the market
makers. There are exchanges for a range of asset types, most notably equities.
For other asset types, most notably the rates and credit instruments, the
principle market making activity rests with the trading desks of the broker
dealers within the bank dealers. Some of these markets are moving toward
electronic trading, for example the swap execution facility. However, the
mode of trade communication and clearing is not the focus of the map; rather
it is the channels through which the trading occurs. The mode of execution
may differ in periods of day-to-day levels of trading versus trading during
periods or market dislocation, with voice trading becoming more dominant
in the latter case.

3. The financial system as a multilayer network

Multilayer networks (Boccaletti et al., 2014; De Domenico et al., 2014)
explicitly incorporate multiple channels of connectivity and constitute the
natural environment to describe systems interconnected through different
categories of connections. Each channel is represented by a layer and the
same node or entity may have different kinds of interactions with different
set of neighbors in each layer. For instance, in social networks, one can
consider several types of different actors relationships: friendship, vicinity,
kinship, membership of the same cultural society, partnership or coworker-
ship, etc. Such a change of paradigm, which has been termed in disparate
ways (multiplex networks, networks of networks, interdependent networks,
hypergraphs, and many others), has already led to a series of relevant and
unexpected results (though up to this point not explicitly to chart the struc-
ture of the financial system). One of the first examples of multilayer networks
that has been thoroughly investigated and modeled, and sparked this strain
of research, is the 2003 Italian blackout. On September 28, 2003, Italy expe-
rienced a country-wide blackout that propagated through the vulnerabilities
of a multilayer network. Electric power moved as edges emanating from
nodes of power stations. Italy had a second network for communications
that among other things was used to send the controls for the operation of
the power network. On that day, one node in the power network failed. That
failure of that one node started off a cascade which enveloped the countrys
entire power system. The one node in the power network spanned to the
communication network by disabling a nearby server. One server failure af-

9
10
Figure 3: Asset map for a typical Bank/Dealer and related entities. Following on from Figure 1 and Figure 2, here we highlight
the typical entities involved in the asset market, most notably the entities that have asset positions and the exchanges and
market makers that facilitate the asset flows. Different avenues are used depending on the asset type. Equities mostly use
exchanges, credit and rate instruments trade through broker-dealer desks, and swaps and other derivatives through banks
either on a bilateral or CCP basis.
fected adjacent servers because they depended on the one server as a part
of their communication cluster. The power stations that were connected to
these servers then shut down due to the loss of their controllers, which in
turn affected the servers adjacent to that node. The failure propagated in
hop scotch fashion from one power node on the power layer to a node in the
communication layer, from that one communication network node to other
nodes on the communication layer, and then back to the power node through
these servers. The result then propagated to other networks ranging from
railways, to healthcare, to the financial system, and to other communication
networks distinct from the communication network with the close proximity
to the power grid (Rosato et al., 2008; Buldyrev et al., 2010). The vulnerabil-
ities came both because the communication and power networks were highly
interdependent the servers depended on power, and the power depended
on the control of the servers – and also because the networks were similar
in topology and geography, with communication nodes closely matching up
with power nodes.
In Figure 4 we present a schematic representation of this multilayer struc-
ture. As some financial entities participate in more than one layer, this results
in the dependency and connectivity between the different layers. One main
source for this is the Bank/Dealer, which participates in all three layers, with
the mode of participation varying from one of its sub-units to another.
The multilayer network formalism presented Figure 4 is made up of three
layers: the Asset Layer, Funding Layer, and Collateral Layer. The three
layers are of a core-periphery topology. The core nodes are the exchanges
and market makers for the asset layer; the Bank/Dealer intermediaries for the
funding layer; and the CCPs and tri- party repo agents for the collateral layer.
For the assets, the peripheral nodes are those that engage in buying and
selling of assets, such as hedge funds and asset managers. For the Funding
Layer, the peripheral nodes are the suppliers and users of funding that engage
through the Bank/Dealer intermediary. We can also differentiate the nodes
based on the directions of flows or linkages in the different layers, as suppliers,
user, and intermediaries. For funding, the supplier is the cash provider, the
user is the hedge fund. For assets, the peripheral nodes act as both suppliers
and users, where these function might be defined either in terms of being
sellers and buyers, or in terms of being those that provide liquidity and those
that demand it. For collateral, the supplier is the one that is receiving the
funding, the user is the one who receives the collateral. In some cases the
peripheral nodes might interact directly, such as in the case of bilateral swap

11
12
Figure 4: A multi-layer network view of the financial focused on the Bank/Dealers and related entities. This multilayer view
restates the components shown in the funding, collateral, and asset maps of Figures 1, 2, and 3, with the various entitles
represented as nodes, and with examples between them. Some financial entities participate in more than one layer, resulting
in the dependency and connectivity between the different layers. Particularly notable are the Bank/Dealers, which participate
in all three layers, due to activities of its various sub-units.
transactions.
The main agents of each of the various layers include, with notation for
whether they are suppliers and/or users, and whether they are core or pe-
riphery within the respective network are detailed in Table 1.
As is evident from Figure 4 and Table 1, some financial entities span
across layers with the Bank/Dealers being unique in spanning all three layers
– while some only operate in an individual layer. The role of an institution
that does span layers may differ from one layer to another. For example,
a hedge fund is defined as being a buyer and seller of assets and a user of
funding. The nature of the spanning, the position in the core-peripheral
topology, and the function of the node can provide a blueprint to define new
categories into which financial institutions will fall, based on the level of their
activity in the different layers. That is, we can define financial institutions by
type based on which layer they operate, whether they are core or periphery,
and whether they are providers or users. The degree to which an institution
that acts as a core node also spans layers gives a measure of its importance
from a systemic risk standpoint. As we will see below, these characteristics
can lead to more fragility than is exhibited by core nodes within a one-layer
network.
Although it is only suggestively illustrated in Figure 4, the number of
central and periphery nodes vary in a notable way from one layer to the
next. While all portraying core-periphery topologies, the number of nodes
vary in order of magnitude, roughly speaking, from one layer to the next.
This is shown in Table 2, where we outline the core and periphery entities
in each layer, and their representative number within the financial system.
In terms of core nodes, there are hundreds of exchanges and market mak-
ing institutions, perhaps a few dozen intermediaries for funding, and only a
handful of central counterparties of note. In terms of the peripheral nodes,
there are tens of thousands of institutional investment firms (ignoring retail
investors), with only a fraction of these involved in funding. And the periph-
eral institutions for collateral are those that are involved in either bilaterally
and through a central counterparty, and this is broadly restricted to a sub-
set of the Bank/Dealers. An important question which will remain beyond
the scope of this paper is whether this order of magnitude difference in the
number of nodes contributes to the nature or severity of the propagation of
shocks in and between the different layers, suggests specific points of vulner-
ability, and thus is an additional key feature of the financial system that can
be manifest through the multilayer network. For example, is there a higher

13
 Table 1: The agents that occupy the various layers of the network, along with their node type and function within that layer,
and their primary federal regulator.

Node Function Primary regulator

Layer Agent Core Peripheral Supplier Intermediary User
Hedge Funds x x x SEC, CFTC
Asset Manager x x x SEC
Asset Bank/Dealer x x OCC, FED, SEC
Market Makers
Bank/Dealer x x OCC, SEC
Derivatives
Exchanges x x SEC, CFTS, SRO
and Non-bank
market Makers
Hedge Funds x x SEC, CFTC
Cash Providers x x NA
Funding
Bank/Dealer Fi- x x OCC, FED

14
nance Desk
Bank/Dealer x x SEC
Prime Broker-
age
Cash Providers (pen- x x Department of
sion funds, insurance Labor, State
Collateral companies insurance de-
partments, FIO
Derivatives x x x CFTC
Bank/Dealer Fi- x x x OCC, FED
nancial Desk
Central Coun- x x SEC, CFTC, FED
terparties
(CCPs)
Tri-party Repo x x FED
Agents
Table 2: Summary of key entities in the different layers, their role (core or periphery) and
their representative number.

Assets
Type Number
Market maker
Central ∼ 100
Exchange
Hedge Funds
Peripheral ∼ 10, 000
Asset Managers
Funding
Type Number
Central Funding Agent ∼ 10
Hedge Funds
Peripheral ∼ 1000
MMF
Collateral
Type Number
Central CCP ∼1
Bank/Dealer
Peripheral ∼ 10
FCM

risk of propagation when a node is disrupted or a shock occurs in a more

sparsely populated layer.
Characterizing the financial system as a multilayer network provides the
opportunity to rethink and redefine its structural properties, which are con-
stantly evolving and adapting, and relationships that might be come to the
fore during a financial crisis. In the current financial system, financial insti-
tutions can be involved in activities that do not directly fall under the specific
title or definition of the given institution. As such, an important issue that
arises is how to apply regulations, which often addresses specific financial
institutions. Viewing the institutions within the context of the multi-layer
network can assist policy makers in monitoring cross-layer institutions, and
provide the means of calibrating regulation policies for the diff t categories.
Furthermore, it will provide the means for tracking how fi institutions change
their nature, in terms of their activity in the different layers, and monitoring
how a financial entity changes its activity in a given layer. Understanding
how it spans across the different layers will provide a new monitoring frame-
work to ensure financial stability. The multilayer formalism provides new
insights into the structure of the financial system, and its function through

15
the dynamics of processes in and between the layers.
When first presented, the distinction between what occurs in a multilayer
versus single-layer network was a surprising result and spurred an active sub-
field in network science. Intuitively, the links between layers give an avenue
for propagation that makes it less likely that a shock will remain contained,
creating a branching channel akin to how an elevator shaft can allow a fire to
spread from floor to floor. Utilizing the knowledge on the spread of damage in
a multilayer interdependent network, we can identify three critical stages, as
highlighted schematically in Figure 5. This phenomenon has been observed
for several different multilayer network systems, and can thus be predicted
also for the financial system. Such knowledge provides new insights into
cascade process in the financial system, and highlight the different stages.
Making use of such knowledge provides new opportunities for policy and
decision makers to decrease the extent of the effect of the damage once it
begins, and ultimately contain it and introduce strategies that will push the
system into a recovery mode (see for example Majdandzic et al. (2014)). The
multilayer interdependent network framework describes the propagation of
damage in three distinct stages: 1) a fast, strong collapse of the system; 2)
a slow, long period in which the damage propagates slowly throughout the
system, in an analogous fashion to cracking ice (see Zhou et al. (2014)); and
3) a final fast strong collapse of the system. This dynamical process results
from a balance that exists between different states of the system. These
two states can be defined as damage amplification (two cascading stages
highlighted in Figure 5) and damage saturation (plateau stage in Figure 5).
In the damage amplification stage, any damage in the system leads to yet
more extensive or severe damage. For example, removal of one node will lead
to a failure of ten nodes, which will lead to a failure of one hundred nodes,
and so forth. In the damage saturation stage, the damage is constant, and a
damage of one node will lead to a damage of one node, and so forth.
Understanding the effect of financial crises and how they propagate through
out the entire system is critical in order to maintain and ensure financial sta-
bility. While it is important to understand the vulnerability of each layer by
itself, when considering the system as a multilayer network, new threats and
vulnerabilities are discovered. Thus, applying the lessons learned from this
framework to the financial system will provide new insights and tools in order
to maintain its stability and functionality. Multilayer networks encode sig-
nificantly more information than their single layers taken in isolation, since
they include correlations between the role of the nodes in different layers

16
Figure 5: Stages of the propagation of damage across the multilayer networks. The mul-
tilayer interdependent network framework describes the propagation of damage in three
distinct stages: 1) a fast, strong collapse of the system; 2) a slow, long period in which
the damage propagates slowly throughout the system, in an analogous fashion to cracking
ice; and 3) a final fast strong collapse of the system.

17
and between statistical properties of the single layers. As discussed above,
multilayer networks have been found to be significantly more fragile in com-
parison to the case of single isolated networks. This can be attributed to the
connectivity and dependence between the different layers, or networks, in the
system. Recently Reis et al. (2014) and Boccaletti et al. (2014) have distin-
guished between different types of similarities of the network layers, which
result in inter-layer connectivity and dependence. These include interlayer
degree correlations (indicates if high (low) degree nodes in layer A will be
high (low) degree nodes in layer B); overlap of connectivity patterns in two
or more layers of the system (the overlap of the links can be quantified by
the global or local overlap between two layers, or by the multidegrees of the
nodes that determine the specific overlapping pattern); node pairwise multi-
plexity (correlation of two, or more, nodes connectivity patterns in the case
where not all nodes are active in all the layers of the system); layer pair-
wise multiplexity (correlation of two, or more, layers in terms of connectivity
patterns, when not all nodes are active in all layers); full dependence; and
partial dependence. Whereas random links between interdependent networks
represent dangerous liaisons, enhancing the fragility of the entire system, the
trustworthy links between the networks are not random, but correlated in a
specific way. First, the links between the layers must be such that the highly
connected nodes, or hubs, of the single layers are also the nodes with more
interlinks. And second, there must be multilayer assortativity. This means
that for two layers, A and B, the hubs in layer A (layer B) are more likely to
be linked with the nodes in layer B (layer A) that are connected with other
hubs in layer B (layer A) (Bianconi, 2014). However, current studies have
shown that interbank networks have low assortativity (Bargigli et al., 2015),
which raises the question about the stability of the interbank system. Thus,
we aim to study the level of overlap and correlation between the structure and
function of the different layers. This knowledge, together with the insights
learned from the previous two propositions, will provide the full picture of
the structure, function, and resilience of the financial system. These results
provide critical quantitative information to address this important question
of integration versus segregation in the financial system (addressing such de-
bates as the Glass-Steagall regulation and too concentrated to fail). We can
replicate these results using real network topologies, as outlined in the previ-
ous section, and simulate different scenarios of dependency level between the
different layers. One such approach to achieve this goal is to use an Agent
Based Model (ABM) approach, as recently introduced by Bookstaber et al.

18
(2014). Combining the ABM with the multilayer framework would provide
the means to stress test different scenarios for varying levels of dependency
between the layers of the financial system.
Finally, the multilayer network has important implications for the issue
of systemically important financial institutions. To the size of an institution,
and even its centrality in terms of any one financial function we can add the
importance of the institution for spanning the various layers. That is, if an
institution can express the disruption in one layer through another layer, it
is moving a local (layer-specific) disruption in to a systemic (multilayer) one.

4. Applications, challenges and data gaps

The specific application and implications of these properties to the finan-
cial system are beyond the scope of this paper; our intent is to extend on the
mapping of the financial system and place it more explicitly in the multilayer
network framework. The previous section highlights critical aspects of the
financial system, when investigated as a multilayer network:

1.) Understanding the financial system as a multilayer network results in

the need of modifying contagion models for the financial system. These
new models must take into consideration the spread of the shocks be-
tween the layers, and how the connectedness and dependence between
the layers lead to the amplification of the shock.
2.) Rethinking how financial institutions are defined, according to their
activity in the different layers. This will require rethinking regulation
and monitoring policies, and provide new definitions into systemically
important financial institutions.
3.) Provide quantitative evidence into the effect of integration versus segre-
gation in the financial system. This is becoming increasingly important
considering how financial institutions, such as banks, are branching out
into new financial activities.
4.) Develop new models to characterize the resilience and vulnerability of
the financial system. This should lead to the development of a new
class of stress tests, and ultimately to a class of intervention strategies.
5.) The multilayer network formalism can lead to new understandings on
the evolution of the financial system, and on link formation. When

19
 links are formed in one layer, this can lead to link formation in related
networks, as well as link deletion.

These challenges require both theoretical, numerical, and empirical ef-

forts. The existing theoretical framework for multilayer networks needs to
be adapted for the case of the financial system, such as the definition of
the dependence between the layers. The models of shock propagation and
contagion for financial networks needs to be revised to address the multilayer-
ness of the system (see for example Kenett and Havlin (2015)). Finally, to
calibrate the models and provide quantitative insights into the challenges
mentioned above, there is a need to tie in several different counterparty data
sources. This requires a significant data collection effort, as the different
financial institutions answer to different regulators (see Table 1). Data col-
lection efforts have significantly increased and expanded following the 2008
financial crisis (see for example Baklanova (2015); Balklanova et al. (2015);
Johnson (2015)). However, while more aggregated data is being collected,
these data collection efforts rarely include counterparty exposure information.
Currently, available counterparty datasources, in the context of this frame-
work, include tri-party repo data, swap data, CDS data and bond market
dealer counterparty data. However, these provide only partial information,
and are far from being sufficient for describing the full structure of the three
layers discussed above, nor the connections or dependencies between the lay-
ers. This is a crucial data gap and challenge that needs to be resolved in
order to uncover the underlying structure of the financial system and provide
full information on sources of systemic risk.
Some initiatives are attempting to address this challenge. One such ini-
tiative that is currently being discussed is the Qualifying Financial Contracts
(QFC) rule proposed by the Financial Stability Oversight Council (FSOC),
which is aimed at closing this gap (FSOC, 2015). A QFC is a securities con-
tract, commodities contract, forward contract, repurchase agreement, swap
agreement, or any similar agreement that the FDIC determines by regulation,
resolution, or order to be a qualified financial contract; and a covered finan-
cial company, which is governed by the rule, is a financial company, other
than an insured depository institution. The rule, if passed, would deem
certain financial companies as record entities, which would require them to
keep electronic records of all QFCs, and provide them to its primary regu-
lator within 24 hours of a request to do so. These filings would include full
position data and collateral counterparty data.

20
 The QFC record keeping framework, or other alternatives, would provide
a first full description of the interconnectedness in the U.S. financial sys-
tem. Such information would provide the means to calibrate the model and
challenges discussed above, and provide new insights into the stability of the
financial system, and how to manage it.

5. Summary and discussion

Financial risks can be described as originating and then propagating
through a three-layer system of assets, funding, and collateral. We can con-
sider risks that are layer specific - asset prices and liquidity; funding liquidity
and leverage; and the collateral of secured lending as well as the counter-
party and credit risk that the collateral seeks to mitigate - but sources of
vulnerability are exposed when considering the dependency and connectivity
between the different layers that are not manifest when the financial system
is modeled as a single layer network, all the more so when it is modeled as
a single layer network with homogeneous nodes. As we move from one layer
to the next, the risk is transformed, so that a price shock becomes a funding
risk as the impact moves down the network to that layer, and similarly as it
moves to the collateral layer it is expressed as a counterparty or credit risk.
Or, looking at the dynamic as a shock moves up from the collateral layer,
if a counterparty risk leads a CCP to have to sell off positions, they then
are pushing activity and risk to the asset layer, and they face liquidity risk
due to the market impact of their selling. The structural implications of the
multi-layer approach include:

1. The definition of an agent or entity is determined by the layers on which

it operates.

2. The type of risk is determined by the layer on which the risk occurs.

3. The transmission of risk through the system occurs through those

agents or entities that span the respective layers. This is why banks
are so critical, because they span all three.

We present a schematic for each of these layers, and for the agents in the
financial system that connect them. These schematics can provide a structure
for investigating how the effect of the multilayer financial system is capable
of withstanding damage or shocks. By understanding the mechanisms of

21
propagation of damage in a multilayer network it will be made possible to
present a deeper understanding of the structure, function, and resilience of
the financial system.

Acknowledgements
We would like to thank Amir Bashan, Sergey Buldyrev, Jill Cettina, Shlomo
Havlin Michelle Farrel, Greg Feldberg, Mark Flood, Jianxi Gao, H. Eugene
Stanley, and Stathis Tompaidis for insightful discussions and comments on
this work.

22
References
Acemoglu, D., Malekian, A., Ozdaglar, A., 2013a. Network security and
contagion. Tech. rep., National Bureau of Economic Research.

Acemoglu, D., Ozdaglar, A., Tahbaz-Salehi, A., 2013b. The network origins
of large economic downturns. Tech. rep., National Bureau of Economic
Research.

Acemoglu, D., Ozdaglar, A., Tahbaz-Salehi, A., 2013c. Systemic risk and
stability in financial networks. Tech. rep., National Bureau of Economic
Research.

Aguiar, A., Bookstaber, R., Kenett, D. Y., Wipf, T., preprint. A map of col-
lateral uses and flows. Office of Financial Research Working Paper Series.

Aguiar, A., Bookstaber, R., Wipf, T., 2014. A map of funding durability and
risk. Office of Financial Research Working Paper Series 14 (03).

Allen, F., Gale, D., 1998. Optimal financial crises. The Journal of Finance
53 (4), 1245–1284.

Amini, H., Cont, R., Minca, A., 2012. Stress testing the resilience of financial
networks. International Journal of Theoretical and applied finance 15 (01).

Baklanova, V., April 2015. Repo and securities lending: improving trans-
parency with better data. Brief 15-03, Office of Financial Research.

Balklanova, V., Copeland, A., McCaughrin, R., September 2015. Reference

guide to u.s. repo and securities lending markets. Working paper 15-17,
Office of Financial Research.

Bargigli, L., Di Iasio, G., Infante, L., Lillo, F., Pierobon, F., 2015. The
multiplex structure of interbank networks. Quantitative Finance 15 (4),
673–691.

Battiston, S., Puliga, M., Kaushik, R., Tasca, P., Caldarelli, G., 2012. Deb-
trank: Too central to fail? Financial networks, the Fed and systemic risk.
Scientific Reports 2.

Bianconi, G., 2014. Multilayer networks: Dangerous liaisons? Nature

Physics.

23
Boccaletti, S., Bianconi, G., Criado, R., Del Genio, C., Gómez-Gardeñes, J.,
Romance, M., Sendiña-Nadal, I., Wang, Z., Zanin, M., 2014. The structure
and dynamics of multilayer networks. Physics Reports 544 (1), 1–122.

Bookstaber, R., Paddrik, M., Tivnan, B., 2014. An agent based model for
financial vunerability. Office of Financial Research Working Paper Series
14 (05).

Boss, M., Breuer, T., Elsinger, H., Krenn, G., Lehar, A., Puhr, C., Summer,
M., 2006. Systemic risk monitor: Risk assessment and stress testing for
the austrian banking system. Internal technical document. OeNB.

Boss, M., Elsinger, H., Summer, M., Thurner 4, S., 2004. Network topology
of the interbank market. Quantitative Finance 4 (6), 677–684.

Bravo-Benitez, B., Alexandrova-Kabadjova, B., Martinez-Jaramillo, S., 2014.

Centrality measurement of the mexican large value payments system from
the perspective of multiplex networks. Computational Economics, 1–29.

Buldyrev, S., Parshani, R., Paul, G., Stanley, H., Havlin, S., 2010. Catas-
trophic cascade of failures in interdependent networks. Nature 464 (7291),
1025–1028.

Cont, R., 2013. Networks: Introduction. Handbook on Systemic Risk, 285.

Cont, R., Moussa, A., Bastos e Santos, E., 2010. Network structure and
systemic risk in banking systems (December 1, 2010).

De Domenico, M., Porter, M. A., Arenas, A., 2014. Multilayer analysis and
visualization of networks. arXiv preprint arXiv:1405.0843.

Drehmann, M., Tarashev, N., 2013. Measuring the systemic importance of

interconnected banks. Journal of Financial Intermediation 22 (4), 586–607.

Elsinger, H., 2009. Financial networks, cross holdings, and limited liability.
WP 156.

FSOC, 2015. Qualified financial contracts recordkeeping related to orderly

liquidation authority.

24
Garas, A., Argyrakis, P., Rozenblat, C., Tomassini, M., Havlin, S., 2010.
Worldwide spreading of economic crisis. New journal of Physics 12 (11),
113043.

Glasserman, P., Young, H. P., 2015. How likely is contagion in financial

networks? Journal of Banking & Finance 50, 383–399.

Haldane, A. G., May, R. M., 2011. Systemic risk in banking ecosystems.

Nature 469 (7330), 351–355.

Haldane, A. G., et al., 2009. Rethinking the financial network. Speech deliv-
ered at the Financial Student Association, Amsterdam, April.

Hüser, A.-C., 2015. Too interconnected to fail: A survey of the interbank

networks literature.

Jackson, M. O., 2010. Social and economic networks. Princeton University

Press.

Johnson, D. C., July 2015. Private fund data shed light on liquidity funds.
Brief 15-05, Office of Financial Research.

Kenett, D. Y., Havlin, S., 2015. Network science: a useful tool in economics
and finance. Mind & Society, 1–13.

Kenett, D. Y., Preis, T., Gur-Gershgoren, G., Ben-Jacob, E., 2012a. Depen-
dency network and node influence: Application to the study of financial
markets. International Journal of Bifurcation and Chaos 22 (7), 1250181.

Kenett, D. Y., Raddant, M., Lux, T., Ben-Jacob, E., 2012b. Evolvement of
uniformity and volatility in the stressed global financial village. PloS one
7 (2), e31144.

Kenett, D. Y., Raddant, M., Zatlavi, L., Lux, T., Ben-Jacob, E., 2012c.
Correlations in the global financial village. International Journal of Modern
Physics Conference Series 16 (1), 13–28.

Kenett, D. Y., Tumminello, M., Madi, A., Gur-Gershgoren, G., Mantegna,

R., Ben-Jacob, E., 2010. Dominating clasp of the financial sector revealed
by partial correlation analysis of the stock market. PloS one 5 (12), e15032.

25
Lillo, F., 2010. Networks in finance. Encyclopedia of Life Support Systems
(EOLSS), Developed under the Auspices of the UNESCO.

Majdandzic, A., Podobnik, B., Buldyrev, S. V., Kenett, D. Y., Havlin, S.,
Stanley, H. E., 2014. Spontaneous recovery in dynamical networks. Nature
Physics 10 (1), 34–38.

Reis, S. D., Hu, Y., Babino, A., Andrade Jr, J. S., Canals, S., Sigman, M.,
Makse, H. A., 2014. Avoiding catastrophic failure in correlated networks
of networks. Nature Physics.

Rosato, V., Issacharoff, L., Tiriticco, F., Meloni, S., Porcellinis, S., Setola, R.,
2008. Modelling interdependent infrastructures using interacting dynami-
cal models. International Journal of Critical Infrastructures 4 (1), 63–79.

Summer, M., 2013. Financial contagion and network analysis. Annu. Rev.
Financ. Econ. 5 (1), 277–297.

Zhou, D., Bashan, A., Cohen, R., Berezin, Y., Shnerb, N., Havlin, S., 2014.
Simultaneous first-and second-order percolation transitions in interdepen-
dent networks. Physical Review E 90 (1), 012803.

26