Credit-Market-of JapanComputational-Economics

Computational Economics (2022) 60:529–569
https://doi.org/10.1007/s10614-021-10157-y (0123456789().,-volV)
(0123456789().,-volV)
Portfolio Correlations in the Bank-Firm Credit Market

of Japan
Duc Thi Luu1
Accepted: 15 July 2021 / Published online: 31 July 2021

Ó The Author(s) 2021
Abstract
The recent global financial crisis has shown portfolio correlations between agents as
one of the major channels of risk contagion and amplification. In this work, we
analyse the structure and dynamics of the cross-correlation matrix of banks’ loan
portfolios in the yearly bank-firm credit network of Japan during the period from
1980 to 2012. Using the methods of Random Matrix Theory (RMT), Principal
Component Analysis and complex networks, we aim to detect non-random patterns
in the empirical cross-correlations as well as to identify different states of such
correlations over time. Our findings suggest that although a majority of portfolio
correlations between banks in lending relations to firms are contributed by noise, the
top largest eigenvalues always deviate from the random bulk explained by RMT,
indicating the presence of non-random patterns governing the correlation dynamics.
In particular, we show that this dynamics is mainly driven by a global common
factor and a couple of ‘‘groups’’ factors. Furthermore, different states in the credit
market can be identified based on the evolution of eigenvalues and associated
eigenvectors. For example, during the asset price bubble period in Japan from 1986
to 1991, we find that banks’ loan portfolios tend to be more correlated, showing a
significant increase in the level of systemic risk in the credit market. In addition,
building Planar Maximally Filtered Graphs from the correlations of different
eigenmodes, notably, we observe that the local interaction structure between banks
changes in different periods. Typically, when the dominance of a group of banks in
one period gradually vanishes, the credit market starts to build-up a different
structure in the next period in which another group of banks will become the main
actors in the backbone of the cross-correlations.
Keywords Bank lending Portfolio correlations Systemic risk Random

matrix theory Principal component analysis Correlation-based filtered
methods
& Duc Thi Luu

d.t.luu@economics.uni-kiel.de
1
Institute of Economics, University of Kiel, Kiel, Germany
123
530 D. T. Luu
1 Introduction
Over the last few years, catastrophic cascade of failures in interdependent systems
has received a remarkable attention in network science. Findings in this line of
research show that the robustness of a system crucially depends on both its internal
structure as well as its pattern of relations to other interdependent ones (e.g. Smart
et al. 2008; Buldyrev et al. 2010; Huang et al. 2011; Brummitt et al. 2012; Reis
et al. 2014; Liu et al. 2016; Bianconi 2018).
In the banking system, the recent global financial turmoil has demonstrated that
besides direct credit linkages1, indirect relations between banks are also the relevant
channels of risk contagion and amplification. For instance, overlapping portfolios
due to common asset holdings have been widely considered as one of the major
sources of risk contagion (e.g. May and Arinaminpathy 2010; Beale et al. 2011;
Huang et al. 2013; Caccioli et al. 2014, 2015; Greenwood et al. 2015; Lillo and
Pirino 2015; Levy-Carciente et al. 2015; Corsi et al. 2016). Analogously, credit
relationships between banks and firms have been suggested as another important
channel of the propagation of distress between the financial system and the real
sector of the economy (e.g. Aoyama et al. 2013; Aoyama 2014; de Castro Miranda
and Tabak 2013; Lux 2016). Accordingly, since banks can be indirectly linked
through a set of joint exposures to firms (i.e. due to loan portfolio overlaps), the
distress originating from a group of banks or firms can be propagated through the
whole system. For these reasons, a number of empirical studies have been devoted
to uncovering the complexity of the topological structure of bank-firm credit
markets world-wide (e.g. Fujiwara et al. 2009; De Masi et al. 2011; De Masi and
Gallegati 2012; Fricke 2016; Marotta et al. 2015, 2016; Fricke and Roukny 2020;
Luu and Lux 2018, 2019; Lux 2020).
It is worthwhile to emphasize that besides portfolio overlaps, portfolio
correlations also play a crucial role in risk management and have been seen as
one of the major sources of financial instability. From a perspective of bankers, the
understanding of the structure of correlations (or co-movements) between banks in
providing loans to non-financial firms as well as the identification of factors
governing such correlations are essential for managing credit risk (e.g. Bülbül and
Lambert 2012; Fenech et al. 2015). The systematic factors driving the correlations
represent market risks that cannot be diversified a way. In contrast, idiosyncratic
risks associated with non-systematic components can be diversifiable as the
dimension of the system grows (e.g. Chamberlain and Rothschild 1983). Further-
more, from a macroeconomic perspective, in a banking system where bank
portfolios are highly unified or coupled, a common shock might trigger the entire
system to follow the similar adjustment strategies. Even if the original shock is
relatively small, it would result in a significant change in the credit supply to the
other sectors of the economy, which in turn might lead to a large impact on
macroeconomic fluctuations (e.g. Kiyotaki and Moore 1997; Khwaja and Mian
2008; Gertler and Kiyotaki 2010; Jorda et al. 2013; Bassett et al. 2014;
1
See, for example, Allen and Gale (2000), Nier et al. (2007), Haldane and May (2011), Acemoglu et al.
(2015), among others.
123
Portfolio Correlations in the Bank-Firm Credit Market of... 531
Brunnermeier and Schnabel 2016; Cingano et al. 2016; Gambetti and Musso 2017;
Bentolila et al. 2018; Azariadis 2018; Alfaro et al. 2018; Amiti and Weinstein
2018). However, so far, the structure and the key drivers of correlations between
banks in lending to firms still remain a puzzling mystery.
This study, therefore, aims to contribute to a deeper understanding of the
structure and the dynamics of the cross-correlation matrix of banks’ loan portfolios.
In particular, our main objectives are to answer the following research questions that
have not been intensively studied in related literature: (i) how to detect non-random
patterns and latent factors (if they exist) in the empirical cross-correlation matrix of
banks’ loan portfolios?; and (ii) how to identify different states in the correlation
dynamics over time?; and (iii) how to quantify systemic risk from loan portfolio
correlations in the credit markets?.
To guide our analysis, we employ the methods of Random Matrix Theory (RMT)
and Principal Component Analysis (PCA). The methods of RMT have been widely
used to filter noise as well as to extract latent information embedded in empirical
correlations for time series data on stock prices/returns (e.g. Laloux et al.
1999, 2000; Plerou et al. 2000, 2002; Kim and Jeong 2005; Bouchaud and Potters
2009; Wang et al. 2011; Meng et al. 2014; Jiang et al. 2014; Uechi et al. 2015;
MacMahon and Garlaschelli 2015), on financial transactions (e.g. Kyriakopoulos
et al. 2009; Fricke 2012), etc. Empirical evidence also shows that RMT is a useful
tool to enhance the Markowitz mean-variance optimization procedure (e.g. Plerou
et al. 2000; Laloux et al. 2000; Sharifi et al. 2004; Daly et al. 2008; Bai et al. 2009;
Quintana et al. 2015). Recently, RMT has been used to study synchronizations
between different macroeconomic indicators (e.g. Ormerod and Mounfield 2000;
Ormerod 2008; Iyetomi et al. 2011; Yoshikawa et al. 2015; Lux et al. 2020).
Indeed, the applications of RMT to the spectral analysis of various types of complex
networks have also received considerable attention in related literature (Bandy-
opadhyay and Jalan 2007; de Carvalho et al. 2009; Potestio et al. 2009; Jalan 2009;
Tran et al. 2013; Dumitriu and Johnson 2016), but to the best of our knowledge
none of them has studied the spectral properties of cross-correlations between nodes
in one set in the formation of links or the strength of interactions with nodes in
another set (e.g. like the correlations between banks in providing loans to firms in
the other sectors of the economy as in the present study).
Practically, in this study, we will, first, compare the eigenvalue spectrum of the
empirical cross-correlation matrix with the theoretical one implied by RMT.
Second, focusing on a group of the largest eigenvalues significantly deviating from
random bulk predicted by RMT, we extract latent information and infer the
significant factors governing the correlation dynamics. This approach allows us to
have a proper selection of the strongest factors containing genuine information
about the properties of the observed data that are different from what would be
expected under a null model. In the next step, in order to identify different states in
the credit market, we examine the evolution of the largest eigenvalues and their
corresponding eigenvectors over time. In addition, building Planar Maximally
Filtered Graphs (PMFG) (Tumminello et al. 2005; Pozzi et al. 2008; Di Matteo
et al. 2010) from the correlation structure for different eigenmodes, we also analyse
the network structure of correlations and investigate how this network changes in
123
532 D. T. Luu
different periods. Furthermore, to infer systemic risk arising from credit extensions
from banks to firms, we measure the absorption ratios that capture the fraction of the
total variance explained by a finite number of the first principal components
(Kritzman et al. 2011; Billio et al. 2012; Zheng et al. 2012; Lux et al. 2020). It
should be emphasized that the procedure used in the present paper can be employed
to filter noise signals and extract significant patterns out of different large bipartite
networks including those in economics and finance such as bank-borrower credit
networks, investor-asset networks.
Perhaps, among different studies on the bank-firm credit relations, the work of
Fujiwara et al. (2009) is closest to our study. Based on eigenvalue analysis,
Fujiwara et al. (2009) measure the fragility scores in terms of dynamical
propagation of influences from lenders to borrowers and vice versa. However,
there exist important distinctions between ours and theirs. To name a few, first, it
should be emphasized that the main difference to the present paper is, instead of
analysing the correlations between banks in lending to firms, Fujiwara et al. (2009)
define a new weighted matrix P capturing the accumulated interactions between the
two sectors. Hence, their approach cannot answer whether the loan portfolios of
banks tend to be more/less correlated in the credit market in different periods.
Furthermore, in such a setting, the largest eigenvalue is always equal to 1, and few
top largest eigenvalues (i.e. the second and the next largest eigenvalues of P) are
important in the propagation of influences from banks to firms and vice versa over a
finite time-scale.2 Different from their approach, we focus on the largest eigenvalues
that significantly deviate from the spectral distribution explained by an appropriate
null model (e.g. the distribution given by the so-called Marchenko-Pastur law
(Marčenko and Pastur 1967; Tao and Vu 2012)).3 Such eigenvalues and the
corresponding eigenvectors are analysed as they might contain the genuine
information of particular economic mechanisms responsible for an unexpectedly
high level of correlations between banks in lending to the non-financial sector of the
economy.
The remainder of this paper is structured as follows. Section 2 briefly describes
the data and methods used to analyse portfolio correlations. Section 3 reports the
main findings for the weighted version of the bank-firm credit network of Japan.
Discussions and concluding remarks are in Sect. 4. In the ‘‘Appendix’’, we provide
additional results for portfolio overlaps and similarities, the fractions of eigenvalues
in different ranges, the distributions as well as the evolution of the eigenvector
2
In their study, briefly, from the matrix of weights W bf representing loans from banks to firms defined in
the next part of the present study, they obtain a matrix A that stands for relative amounts of lending by
banks to firms and a matrix B that stands for the relative amount of borrowing by firms from banks. Then
the matrix P ¼ AB will capture the accumulated reflections and influences between banks and firms.
ðPÞ ðPÞ ðPÞ ðPÞ
Assume that k1 k2 ::: kN are the eigenvalues of P, it can be shown that k1 ¼ 1. Therefore,
ðPÞ
over a finite time-scale, only few next largest eigenvalues k2 ... are important in the dynamics of the
propagation of influences from banks to firms and vice versa. For more details, we refer readers to the
study of Fujiwara et al. (2009).
3
Note that to deal with potential effects of the presence of extreme loan values on the largest
eigenvalues, we also consider the other upper bounds obtained from heavy-tailed random matrices (see
Sect. 2.2.2).
123
elements, and the more detailed correlation network structure between banks based
on the method of PMFG. We also summarize the main results for the binary version
of the credit network.
2 Data and Methods
2.1 Data
We use a yearly data set containing credit relations between banks and large firms in
Japan over the period from 1980 to 2012. The data were obtained from the firms’
financial statements and the survey by Nikkei Media Marketing, Inc. in Tokyo
(commercially available).4 This covers the most critical time and events over the
past few decades in the Japanese economy, e.g. the bubble period from 1986 to
1991, the burst of the bubble economy and the stagnation time in the subsequent
years (the so called ‘‘lost decade’’) (e.g. see Hoshi and Kashyap 2000; Brunnermeier
and Schnabel 2016; Himino 2021), the recent global financial crisis.5
Based on the maturity of loans, we separately consider three layers of lending
relationships from banks to firms, i.e. total lending relationships (namely layer 1), short-
term lending relationships (namely layer 2), and long-term lending relationships (namely
layer 3). Note that the layer 2 captures all loan contracts lasting less than or equal to one
year. In contrast, the layer 3 only consists of loan contracts that exceed one year.
For the sake of conciseness, we exclude other financial institutions including
insurance companies and a small number of aggregated banks.6 In addition, it
should be emphasized that for each maturity-based layer, we focus on the aggregate
amount of loans from each bank to each firm at the end of the fiscal year, meaning
that we ignore contract-specific frequencies (i.e. separate times that a bank lends to
a firm during the fiscal year).
2.2 Methods
2.2.1 Representation of a Bank-Firm Network, Portfolio Overlaps and Projection

Matrices
We shall start with some basic notations and definitions of a bank-firm credit
network. Assume that in each year we have a weighted bipartite network consisting
of N banks lending to NF firms. Let W bf ¼ fwbf
ij gNxNF be the matrix of weights (i
runs from 1 to N standing for N banks, and j runs from 1 to NF representing NF
4
For a more detailed explanation about the data set, we refer readers to, for example, Fujiwara et al.
(2009), Marotta et al. (2015), Marotta et al. (2016).
5
Although additional analysis for more recent data could provide more insights on the evolution of the
portfolio correlations among banks in lending to non-financial firms, we leave it for future study once data
is available.
6
Each aggregated bank consists of a group of small/local banks. However, the main conclusions do not
change very much when we include insurance companies and aggregated banks in the data, since the
market shares of these companies and banks are relatively small.
123
534 D. T. Luu
firms). Each element wbf 7

ij is non-negative and implies the monetary amount of loan that
bank i lends to firm j. Notice that from the matrix W bf , we obtain the adjacency matrix
Abf ¼ fabf bf bf
ij gNxNF , where aij ¼ 1 if wij [ 0, and zero otherwise. This adjacency matrix
is associated with the binary version of the bank-firm credit network.
From a perspective of network theory, one can measure the structural similarity
between banks’ loan portfolios based on various measures of the similarity between
two vectors (e.g. Cai et al. 2018; Pool et al. 2015; Alvarez-Socorro et al. 2015;
Fricke 2016; Luu and Lux 2018, 2019). Briefly, for each pair of nodes in the same
set in the bipartite network, the similarity between them is based on the similarity of
their neighborhoods. For instance, in the weighted version, we can, respectively,
define the similarity matrix of banks and that of firms based on the generalized
Jaccard index as
P NF
k¼1 min wbf
ik ; wbf
jk
SBB
ij ¼P ; ð81 i 6¼ j NÞ; ð1Þ
NF bf bf
k¼1 max wik ; wjk
and
PN
bf bf
k¼1 min w ki ; wkj
SFF
ij ¼P ; ð81 i 6¼ j NF Þ: ð2Þ
N bf bf
k¼1 max wki ; w kj
With non-negative weights in the bank-firm credit network, we have that

P F P NF P
0 Nk¼1 minðwbf bf
ik ; wjk Þ
bf bf
k¼1 maxðwik ; wjk Þ and 0 Nk¼1 minðwbf bf
ki ; wkj Þ
PN bf bf
k¼1 maxðwki ; wkj Þ. Furthermore, since in the present study we focus on banks that
lend to at least one firm and firms that borrow from at least one bank, the two
P NF bf bf PN bf bf
denominators k¼1 maxðwik ; wjk Þ and k¼1 maxðwki ; wkj Þ are strictly positive.
Therefore, SBB
ij and SFFij range in [0, 1].
Notice that SBB
ij ¼ 1 if and only if wbf bf
ik ¼ wjk for all firms k. Furthermore,
SBB
ij ¼ 0 if and only if the portfolios of the two banks i and j have no overlaps at all.
Therefore, each element of the matrix SBB demonstrates the degree of similarity
between the lending portfolios of a pair of banks. In a similar vein, each entry of the
matrix SFF indicates the degree of similarity between the borrowing portfolios of
two firms.
Furthermore, the one-mode projection matrices obtained from the original bank-
firm credit network also indicate the lending portfolio overlaps between banks or the
borrowing portfolio overlaps between firms (e.g. Luu and Lux 2018, 2019). More
specifically, defining
ABB ¼ Abf Afb ; ð3Þ
we then obtain a matrix in which each off-diagonal element is the number of firms
7
Depending on the considered layer, it can be the long-term loan, the short-term loan, or the total loan.
123
that a pair of banks co-finance. In a similar way, each off-diagonal element of the
AFF matrix, with
AFF ¼ Afb Abf ; ð4Þ
indicates the number of common lenders that a pair of firms have.

A major disadvantage of the network projection approach is that the information
about the monetary value of loans from banks to firms is partially or completely
discarded (e.g. see Lehmann et al. 2008; Luu and Lux 2018). In addition, often
sparser information about the original bipartite structure of credit links is projected
into denser one-mode networks. More importantly, from an economic perspective,
the network-based analysis of the two projection matrices (and the similarity
matrices) alone cannot help to identify which pairs of banks tend to be more/less
correlated in lending to firms. This provides the motivation for introducing an
appropriate procedure used to measure and analyse the loan portfolio correlations
between banks, which shall be briefly explained in the following.8
2.2.2 Analysis of Portfolio Correlations
2.2.2.1 Correlation Matrix We now introduce the method used to measure the
empirical cross-correlation matrix of banks’ loan portfolios. Here we focus on the
weighted version of bank-firm credit relationships. Additional explanations and
results for the binary version (i.e. based on the elements of the matrix
Abf ¼ fabfij gNxNF ) are provided in the ‘‘Appendix’’.
Similar to the procedure often implemented in the analysis of stock returns/
prices, for wbf
i;j [ 0, the normalized amount of the loan (in log-scale) from bank i to
firm j is defined as

ln wbfij hln wbfij i
Xij ¼ ; ð5Þ
rij
where for each bank i, hlnðwbf

ij Þi and rij are the average and the standard deviation of
lnðwbf bf
ij Þ over all NF firms, respectively. As a special case, we let Xij ¼ 0 if wij ¼ 0.
9
Denote X ¼ fXij gNxNF , the cross-correlation matrix of N banks is then defined as
8
Note that although in the present work we focus on loan portfolio correlations between banks, we also
summarize the main results for loan portfolio overlaps and similarities in the ‘‘Appendix’’.
9
It should be emphasized that in general, almost all main conclusions from this study still hold if we
define

ln w~bf
ij hln w~bf
ij i
Xij ¼ ;
r~ij
with w~bf bf bf
ij ¼ ðwij þ 1Þ. One can also use the original weight wij instead of the loan amount in log-scale to
define the normalized quantity Xij . Although for the sake of brevity the results for these alternative
definitions of Xij are not reported here, they available from the author upon request.
123
536 D. T. Luu
1
C ¼ fCij gNxN ¼ XXT ; ð6Þ
NF
where X T is the transposition of X . Note that 1 Cij 1 (for 1 i\j N) and
Cii ¼ 1 (for all i from 1 to N). From an economic point of view, the value of Cij
indicates the correlation between bank i and bank j in lending to all firms. In
particular, Cij [ 0 ð\0Þ implies that two banks i and j are positively (negatively)
correlated, while Cij ¼ 0 demonstrates that there is no correlation between the two
banks at all.
Note that with the particular application to the bank-firm credit network of Japan
in the present work, the ratio NNF is larger than 1, since the number of banks is always
less than that of firms over the sample period. However, if the number of borrowers
was less than the number of lenders, the estimated covariance matrix between banks
in lending to firms would become singular.10 Similarly, in analyses of economic and
financial time series, when the number of observations over a relatively short period
is smaller than the number of the indices (e.g. those represent different stock prices
or macroeconomic indicators), such a singularity problem also emerges. In this case,
the estimated covariance matrix will have the singular Wishart distribution (e.g. see
Srivastava 2003; Bouchaud et al. 2007; Bodnar et al. 2013, 2014). As a an
important consequence in portfolio analysis, if the estimated covariance matrix is
singular, the matrix inversion operation used to estimate the portfolio weights is not
possible. To tackle this problem, several transformation methods with practical
applications to portfolio theory haven been proposed (c.f. Pappas and Kaimakamis
2010; Bodnar et al. 2016, 2017, 2018, 2019, among others for a more detailed
discussion on different transformation methods and their practical relevance).
2.2.2.2 Random Matrix Theory and Selection of Factors We shall now introduce
the relevant results of RMT that will be used to analyse the cross-correlations matrix
defined in Eq. (6). Suppose that fki gi¼N
i¼1 are the eigenvalues of a correlation matrix
C and fC ðkÞ is the probability density function of the eigenvalues of C. According to
RMT (e.g. Marčenko and Pastur 1967; Tao and Vu 2012), if the elements of
iid
X N ð0; r2 Þ, then in the limit N; NF ! 1 with Q ¼ NNF ! a (a is constant and
a [ 1), the eigenvalue distribution fC ðkÞ is given by the Marchenko-Pastur law
8 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
< Q ðkmax kÞðk kmin Þ
fC ðkÞ ¼ 2pr2 if kmin k kmax ; ð7Þ
: k
0 elsewhere.
In (7), r2 is the variance of the elements of X (r2 is equal to 1 in our case), and kmax
and kmin are the upper and lower bounds of eigenvalues predicted by RMT,
respectively. With unit variance, these two bounds are given by
10
This could also happen in other bipartite financial networks such as bank-asset or investor-asset
networks. In particular, when the number of assets is smaller than the number of banks or investors, it also
leads to the singularity problem in the bank-bank covariance matrix or the investor-investor covariance
matrix.
123
pffiffiffiffiffiffiffiffiffi 2 pffiffiffiffiffiffiffiffiffi
kmax ¼ ð1 þ 1=QÞ and kmin ¼ ð1 1=QÞ2 : ð8Þ
Generally speaking, if some of the eigenvalues of the empirical cross-correlation

matrix are beyond the interval ½kmin ; kmax , they carry genuine information about the
correlations between banks (Plerou et al. 2000, 2002). Nevertheless, it should be
emphasized that the presence of extreme values in the elements of X (e.g. when X is
a heavy-tailed random matrix) might cause the largest eigenvalues to exceed the
upper bound implied by the Marchenko-Pastur law (e.g. Biroli et al. 2007; Bou-
chaud and Potters 2009). In this case, some eigenvalues that are larger than kmax
might actually contain no genuine information.
As shown in Biroli et al. (2007), whenever the largest entry of X (in absolute
1
terms), S, satisfies that S ðN:NF Þ4 , the upper limit for the largest eigenvalue still
1
remains stuck at kmax . However, if S [ ðN:NF Þ4 , a new upper bound kð1Þ
max will
emerge:

ð1Þ 1 S2 NF
kmax ¼ þ 1þ 2 : ð9Þ
Q NF S
Furthermore, when the elements of X have power-law tails with exponent l, then
the large values among fXij g will dominate the top largest eigenvalues. The new
upper bound kð2Þ
max now depends on l as
kð2Þ
max ¼ N
4=l1 2=l1
Q : ð10Þ
Hence, in the present study, to capture the potential effects of extreme values of
loans from banks to firms, we also compare the top largest eigenvalues with these
different bounds. As shown later, in our analysis the top five largest eigenvalues of
the empirical correlation matrix C are always larger than all three upper bounds
kmax , kð1Þ ð2Þ
max and kmax .
2.2.2.3 Decomposition of Correlation Matrix We now introduce one of the most

important applications of RMT, i.e. how to extract latent information from the
largest eigenvalues deviating from the random bulk. Without loss of generality, we
assume that k1 k2 ::: kN are the eigenvalues of the empirical correlation
matrix C, and u1 ; u2 ; . . .; uN are their corresponding eigenvectors. The correlation
matrix C can be diagonalized as
C ¼ UKUT ; ð11Þ
with K ¼ diagfk1 ; . . .; kN g, and fUgNxN is an orthonormal matrix, whose ith column

is the normalized eigenvector ui corresponding to the eigenvalue ki . As a result, for
each ki we have
ki ¼ uTi Cui ¼ uTi CovðX Þui ¼ VarðuTi X Þ: ð12Þ
Let yi ¼ uTi X , we obtain the following property for the total variance of fX i gNi¼1 :
123
538 D. T. Luu
X
N X
N X
N
VarðX i Þ ¼ N ¼ ki ¼ Varðyi Þ: ð13Þ
i¼1 i¼1 i¼1
From Eqs. (12) and (13), we can easily show that the ratio kNi indicates the percentage
P
of the total variance Ni¼1 VarðX i Þ explained by the principal component yi ¼ uTi X
(Jolliffe 1986; Wang et al. 2011; Gewers et al. 2018).
In addition, since Eq. (11) can be re-written as
X
N
C¼ ki ui uTi ; ð14Þ
i¼1
we shall now decompose the empirical correlation C into different parts capturing
the effects of different factors. First, if the largest eigenvalue k1 deviates signifi-
cantly from the random bulk, the so-called ‘‘market’’ mode, i.e. the global common
factor (or the systematic factor) that influences the lending of all banks in the credit
market, can be extracted from this eigenvalue and its corresponding eigenvector
(Plerou et al. 2000, 2002). Its effect on C is represented by
C m ¼ k1 u1 uT1 ; ð15Þ
and the rest of C is
FCm ¼ C C m ; ð16Þ
which is the filtered matrix after the market-wide effect is subtracted.

Next, genuine information can be extracted from the next largest eigenvalues if
they are still larger than the selected upper bound (e.g. the bound kmax implied by
ð1Þ ð2Þ
RMT or the more ‘‘conservative’’ one given by kallmax ¼ maxðkmax ; kmax ; kmax Þ). Such
information may be associated with the level of sub-groups (namely the ‘‘groups’’
modes) (e.g. Kim and Jeong 2005; Shen and Zheng 2009; Jiang and Zheng 2012). In
that case, following Kim and Jeong (2005), we can further decompose C into three
parts Cm , C g , and C r as
C ¼ Cm þ Cg þ Cr ; ð17Þ
where C g and C r indicate the effect of the ‘‘groups’’ modes and the rest of C,
respectively. From an economic perspective, the matrix Cg captures the influence of
non-systematic components (e.g. sectoral factors, regional factors, etc.) that only
affect the lending of a subset of banks in the credit market. The residual represented
by the matrix C r is refereed to as the random noise part.11
11
Indeed, the interpretations for different factors used in the decomposition of the matrix C are
somewhat similar to those used for the dynamic factors models, which have also become one of the
important dimension reduction techniques in macro-econometrics as well as financial econometrics (e.g.
Forni et al. 2000; Forni and Lippi 2001; Barigozzi et al. 2014; Forni et al. 2015; Matteo and Marc
2016b, a; Forni et al. 2017).
123
Mathematically,
Ng
X
Cg ¼ ki ui uTi ; ð18Þ
i¼2
and the rest

X
N
Cr ¼ ki ui uTi ; ð19Þ
i¼Ng þ1
where Ng is determined based on the number of the next largest eigenvalues larger
than the selected upper bound (see also Kim and Jeong (2005), MacMahon and
Garlaschelli (2015) for further discussions about the selection of Ng ). From
Eqs. (17) to (19), we can define FCm;g ¼ ðC Cm Cg Þ ¼ C r as the filtered matrix,
in which the effects of the ‘‘market’’ and ‘‘groups’’ modes are subtracted.
2.2.2.4 Absorption Ratios and Systemic Risk From Eq. (13), for each
i ¼ 1; 2; . . .N, we define the absorption ratio Ei as
Pi Pi
j¼1 kj j¼1 VarðX j Þ ð20Þ
Ei ¼ ¼ ;
N N
which indicates the fraction of the total variance of fX j gNj¼1 explained by (or
absorbed by) the first i principal components (e.g. Pukthuanthong and Roll 2009;
Wang et al. 2011; Billio et al. 2012).
PN PN
Since i¼1 VarðX i Þ ¼ N ¼ i¼1 ki , the ratio EN is equal to 1, representing
100% of the total variance explained by all components together. Hence, comparing
different absorption ratios E1 ; . . .; Ek to EN ¼ 1 allows us to assess the contribution
of the significant principal components to the total variance.12 On top of this, we can
use the absorption ratios to analyse the systemic risk in the market. For example, if
E1 accounts for a larger part of EN when the largest eigenvalue k1 increases, it
indicates a higher level of systemic risk. This is because it implies that banks in the
credit market now become more unified and strongly coupled, and a shock, thus, can
be quickly and widely propagated throughout the whole system (e.g. Pukthuanthong
and Roll 2009; Billio et al. 2012; Zheng et al. 2012; Meng et al. 2014).
2.2.2.5 Inverse Participation Ratios We use the Inverse Participation Ratio (IPR)
as an indicator of the contribution to the eigenvectors of the correlation matrix
(Plerou et al. 1999, 2002). It is given by
X
N
IPRi ¼ ui ðjÞ4 : ð21Þ
j¼1
The inverse of IPRi indicates the number of elements that contribute significantly to
12
Here k is the largest integer such that kk is larger than the selected bound, e.g. kmax implied by RMT or
ð1Þ ð2Þ
the more ‘‘conservative’’ one kallmax ¼ maxðkmax ; kmax ; kmax Þ.
123
540 D. T. Luu
ui . More specifically, a lower value for IPR implies that banks contribute more
equally (i.e. a less localized behavior). In contrast, a higher value for IPR shows that
a smaller number of banks dominate in the eigenvector (i.e. a more localized
behavior), signaling the presence of a more hierarchical structure.
P
In the normalized version of eigenvectors, if Nj¼1 ui ðjÞ2 ¼ 1 (8i ¼ 1; 2; . . .N),
the ratio IPRi will have two limiting cases
1
1 IPRi ; 8i ¼ 1; 2; . . .N: ð22Þ
N
qffiffiffi
Notice that IPRi ¼ N1 if and only if ui ðjÞ ¼ N1 for all j ¼ 1; 2; . . .N (i.e. under the
completely homogeneous structure). In contrast, IPRi ¼ 1 if and only if only one
element contributes to the eigenvector ui while all other elements are equal to zero
(i.e. under the most hierarchical structure).
2.2.2.6 Complex Network Approach to Correlation Structure Analysis In order to

identify the backbone of the interdependencies between banks, we use the so-called
Planar Maximally Filtered Graph (PMFG). Broadly speaking, this method allows us
to filter and map the correlation matrix into a sparser graph and retain only the
important correlation parts (e.g. Tumminello et al. 2005; Di Matteo et al. 2010;
Aste et al. 2010; Jiang et al. 2014; Massara et al. 2015). Based on the decompo-
sition of the empirical correlations into different eigenmodes (see Eq. (17)), we can
also construct the PMFG for the different correlation matrices, e.g. the filtered
correlation matrix preserving only the ‘‘market’’ mode (i.e. Cm ), the filtered
correlation matrix retaining only the ‘‘groups’’ modes (i.e. C g ), or the noise-filtered
correlation matrix maintaining all significant correlations (i.e.
C m;g ¼ C m þ C g ¼ C C r ).
Indeed, the construction algorithm of the PMFG is similar to the one used for the
so-called Minimal Spanning Tree (MST)–one of the simplest connected graphs
widely used to capture a minimal set of relevant interactions associated with the
strongest correlations (e.g. Mantegna 1999). To construct the MST graph, one can
follow the following steps: First, sorting the pairwise correlations fCij g in
descending order, we obtain a list L of the N 2 reordered correlation elements. In the
second step, choosing the two banks i and j associated with the first element in L,
we put an edge between banks i and j to the MST graph. In the third step, we select
the next edge in the list L, and then add the corresponding link and nodes to the
graph if it is not connected to the existing nodes and does not form any cycles. We
then repeat the third step until all the correlation elements in the list L have been
examined.
Notice that since in the third step in the construction of the MST, all new edges
are excluded if they form a cyclic structure, the resulting filtered MST is a tree
graph consisting of N banks with ðN 1Þ edges. Under such a strong constraint on
an acyclic structure, the presence of significant correlations that have loops will be
definitely discarded (e.g. Tumminello et al. 2005; Di Matteo et al. 2010; Wang
et al. 2018). To provide a more significant and richer structure, Tumminello et al.
123
(2005) propose the PMFG in which the resulting connected planar graph retains
ð3N 6Þ edges and possibly contains 3- and 4- closed cliques (in the third step of
the aforementioned construction algorithm of the MST). In fact, it can be shown that
the resulting MST graph is always a subgraph of the PMFG (e.g. Tumminello et al.
2005; Pozzi et al. 2008).
3 Portfolio Correlations in the Bank-Firm Credit Market of Japan
To begin with, we briefly summarize the main statistical properties of the elements
of the correlation matrices in the three lending layers of the bank-firm credit market
of Japan over the period from 1980 to 2012. The time-varying distributions of Cij in
the different layers are shown in Fig. 1, while the fundamental statistics including
the mean, median, skewness and kurtosis of Cij are reported in Fig. 2. Overall, we
observe that the averages of correlations in all layers are relatively small but always
positive. In addition, these distributions exhibit positive skewness and high kurtosis.
In order to investigate whether non-random patterns can be identified in the
correlations between loan portfolios, in each layer, first, we compute the
eigenvalues as well as eigenvectors of the empirical correlation matrix C. After
that, we will compare the distribution of eigenvalues of C with the one predicted by
RMT. Note that in our analysis we also consider the potential effects of extreme
values on the upper limit of eigenvalues by taking into account the other two upper
bounds kð1Þ ð2Þ
max and kmax , besides the bound kmax implied by RMT.
As shown in Fig. 3 and in Fig. 11 (see the ‘‘Appendix’’), a majority of the
eigenvalues of the empirical correlation matrix lie within the random bulk [kmin ,
kmax ]. Compared to this, we find that about additional 10% of the eigenvalues are in
the broader range [kmin , kallmax ] when the more conservative upper bound kmax ¼
all
ð1Þ ð2Þ
maxðkmax ; kmax ; kmax Þ is used. Furthermore, we also observe a group of the smallest
eigenvalues deviating from the RMT’s left edge (i.e. less than the lower bound
kmin ). However, these eigenvalues are not of interest in the present work as their
contribution to the total variance is relatively small.
Moving on to the largest eigenvalues exceeding the RMT’s right edge, typically
from 10 % to 20 % of eigenvalues are always larger than kmax over the years, while a
smaller fraction is observed when we only count the eigenvalues that exceed kall max .
All taken together, the inspection of the empirical distribution the eigenvalues
suggests that genuine information from the bank-bank correlation matrices and
significant factors driving the correlation dynamics over the years in the three layers
can be extracted from some of the largest eigenvalues.
In the following, for the sake of conciseness, we select the top five largest
eigenvalues of the empirical correlation matrix in each lending layer and then
investigate their temporal evolution over time. As shown in the panels (a), (c), and
(e) of Fig. 4, over the years, these eigenvalues persistently deviate from the random
bulk, and they are always larger than the all three upper bounds kmax , kð1Þ ð2Þ
max and kmax .
To show the contribution of the first five principal components to the total variance
and to infer the systemic risk, we also report the associated absorption ratios defined
123
542 D. T. Luu
7000 7000
2010 2010
2000 2000
6000 6000
1990 1990
1980 1980
5000 5000
frequency
frequency
4000 4000
3000 3000
2000 2000
1000 1000
0 0
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.5 0 0.5 1
Cij Cij
(a) Dist. of Cij , layer 1 (b) Dist. of Cij , layer 2

10000
2010
2000
1990
8000
1980
6000
frequency
4000
2000
0
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Cij
(c) Dist. of Cij , layer 3

Fig. 1 Temporal evolution of the distributions of Cij in the three layers. Panels a–c: The distributions of
Cij in the three layers in the four selected years 1980, 1990, 2000, 2010
in Eq. (20). Overall, to a certain degree, the layer 1 (total loans) and layer 3 (long-
term loans) display a similar trend. As demonstrated in Fig. 4, a couple of regimes in
the bank-firm credit market of Japan can be identified from these two layers. For
example, the temporal evolution shows that almost all top five eigenvalues and the
absorption ratios increase sharply during the period of the lending boom and
Japanese asset price bubble (1986–1991).13 This implies that in this period the top
five significant factors become more important driving forces of the correlation
dynamics, and Japanese banks tend to be more tightly coupled in lending to the non-
financial sector of the economy. Hence, it indicates a higher level of systemic risk in
the credit market during the bubble time in the sense that banks’ lending becomes
more vulnerable to negative shocks (Kritzman et al. 2011; Billio et al. 2012; Meng
et al. 2014). As we can see in the consecutive years after the bubble bursting, all of
these eigenvalues as well as the corresponding absorption ratios substantially
13
See, for example, Hoshi and Kashyap (2000), Posen (2003) and Brunnermeier and Schnabel (2016) for
further discussions on the lending boom and the emergence of asset price bubble in Japan in 1980s.
123
0.4 0.4 0.4

Cij Cij Cij
median median median
0.3 0.3 0.3
0.2 0.2 0.2

Cij
Cij
Cij
0.1 0.1 0.1
0 0 0
-0.1 -0.1 -0.1
-0.2 -0.2 -0.2

1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
years years years
(a) mean(Cij ), layer 1 (b) mean(Cij ), layer 2 (c) mean( Cij ), layer 3
3 skewness skewness skewness
4 2.5
2.5 3.5
2
3
2
2.5
1.5
1.5
2
1.5 1
1
1
0.5 0.5
0.5
0 0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
years years years
(d) skewness(Cij ), layer 1 (e) skewness(Cij ), layer 2 (f) skewness(Cij ), layer 3
kurtosis
35 kurtosis kurtosis
20 16
30
14
15 25 12
20 10
10 8
15
6
10
5 4
5
2
0 0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
years years years
(g) kurtosis(Cij ), layer 1 (h) kurtosis(Cij ), layer 2 (i) kurtosis(Cij ), layer 3
Fig. 2 Fundamental statistics of Cij in the three layers. Panels a–c The averages (together with their 95 %
confidence intervals) and the medians of Cij in the three layers. Panels d–f The values of the skewness of
Cij in the three layers. Panels g–i The values of the kurtosis of Cij in the three layers
decline, and even decrease further after the Asian financial crisis (1997–1998). Such
a trend somewhat stops in 2002, when the Bank of Japan (BOJ) implements a
monetary stimulus policy (e.g. Bowman et al. 2015). In addition, as a consequence
of the economic contraction in Japan hit by the recent global crisis (e.g. Kawai and
Takagi 2011), we observe a significant decrease in all of these eigenvalues and
absorption ratios in 2009 in all three layers.
Furthermore, interestingly, compared to the layers 1 and 3, the layer 2 (short-term
loans) possesses a different dynamics: the largest eigenvalues and their contribu-
tions to the total variance show a decline from the Asian financial crisis, suggesting
that from then on Japanese banks tend to be less unified in this layer. Notice that as
shown in panel (b) of Fig. 2, the correlations between banks in the layer 2 also
exhibit a similar dynamics: on average, banks’ short-term loans to firms become less
123
544 D. T. Luu
1.4 1.5 1.4

empirical empirical empirical
M.P law M.P law M.P law
1.2 1.2
1 1
1
0.8 0.8
( )
( )
( )
0.6 0.6
0.5
0.4 0.4
0.2 0.2
0 0 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
(a) Dist. of λ, layer 1, 1980 (b) Dist. of λ, layer 2, 1980 (c) Dist. of λ, layer 3, 1980
1.8 1.4 1.6
1.6 M.P law M.P law M.P law
1.2 1.4
1.4
1.2
1
1.2
1
1 0.8
( )
( )
( )
0.8
0.8 0.6
0.6
0.6
0.4
0.4
0.4
0.2 0.2
0.2
0 0 0
0 2 4 6 8 10 12 0 1 2 3 4 5 0 5 10 15
(d) Dist. of λ, layer 1, 2012 (e) Dist. of λ, layer 2, 2012 (f) Dist. of λ, layer 3, 2012
Fig. 3 Distributions of the eigenvalues of the correlation matrices, compared with RMT, in the three
layers, in 1980 and 2012. Panels a–c show the distributions of of k in 1980, and d–f show the distributions
of k in 2012. In all panels, the red curve represents the spectral distribution explained by the Marchenko-
Pastur law. Note that over the years, we observe that the top five largest eigenvalues of the empirical
correlation matrix in each lending layer are always larger than the three upper bounds kmax , kð1Þ ð2Þ
max and kmax
(see also Fig. 11 in the ‘‘Appendix’’)
correlated since the aftermath of that crisis. In fact, by a closer inspection from a
network perspective, we find that credit linkages in this lending layer also become
sparser and the average of weights (in absolute terms and in log-scale) substantially
drops since then.
Having discussed the important statistics of the eigenvalues of the correlation
matrix C in each lending layer, we now move on to the analysis of the eigenvectors
of C. Note that the ith element of an eigenvector uk indicates the contribution of the
bank i to uk , thus, showing the role of the bank i in the kth principle component.
To begin with, we compare the distributions of the empirical eigenvector
elements with the theoretical distribution predicted by RMT. According to the
Gaussian prediction of RMT, the elements of each normalized eigenvector will
follow (e.g. Guhr et al. 1998; Laloux et al. 1999; Plerou et al. 2002):
1 u2
PRMT ðuÞ ¼ pffiffiffiffiffiffi e 2 : ð23Þ
2p
The detailed comparison results are shown in Fig. 12 in the ‘‘Appendix’’. Overall,
again we typically find that the distributions of the elements of eigenvectors asso-
ciated with the largest eigenvalues significantly deviate from the Gaussian predic-
tion of RMT. In contrast, the distributions of the elements of eigenvectors associated
with eigenvalues in the random bulk are somewhat more similar to PRMT . This
123
0.5 E1
1
20
E2
2 0.45
18 E
3 3
0.4 E
16 4 4
Largest Eigenvalues
E5
0.35
Absorption Ratios
5
14 (1) E
max 10
(2)
0.3 E
12 all sig. factors
max
* 0.25
10 max
8 0.2
6 0.15
4 0.1
2 0.05
0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years
(a) Largest eigenvalues, layer 1 (b) Absorption ratios, layer 1

E
16 1 1
0.45 E2
2
14 E
3 0.4 3
E
4 4
12 0.35
Largest Eigenvalues
5
Absorption Ratios E5
(1) E
max 0.3 10
10 E
(2) all sig. factors
max
0.25
*
8
max
0.2
6
0.15
4
0.1
2 0.05
0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years
(c) Largest eigenvalues, layer 2 (d) Absorption ratios, layer 2

E
1 0.6 1
30 2
E2
E
3 3
0.5 E4
25 4
Largest Eigenvalues
E
Absorption Ratios
5 5
(1) 0.4 E
10
20 max
(2) E all sig. factors
max
* 0.3
15 max
10 0.2
5 0.1
0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years
(e) Largest eigenvalues, layer 3 (f) Absorption ratios, layer 3
Fig. 4 Evolution of the top five largest eigenvalues and absorption ratios for the correlation matrices in
the three layers. All panels on the left show the evolution of the top five largest eigenvalues. These
eigenvalues are persistently larger than the bound kmax implied by RMT as well as the two more
conservative bounds kð1Þ ð2Þ
max and kmax . The right panels show the five associated absorption ratios (E1 to E5 )
and the absorption ratio of the top ten largest eigenvalues (E10 ) and that of all eigenvalues that exceed
kmax (namely Eallsig:factors )
observation is in agreement with those reported in analyses of stock returns (e.g.

Laloux et al. 1999; Plerou et al. 1999).
In the next step, based on the eigenvector elements, we investigate the degree of
homogeneity/heterogeneity of banks’ contribution to C. To this end, we compute the
123
546 D. T. Luu
Inverse Participation Ratios fIPRk gk¼N k¼1 and then examine the relationship between
these ratios and the corresponding eigenvalues. Large values of IPRk reveal that
only few banks contribute to uk , suggesting the presence of a localized behavior of
banks. In contrast, a small value of IPRk indicates that many banks contribute to uk
together, showing that the factor (if it exists) extracted from that eigenvector has a
pervasive effect in the market.
All panels of Fig. 5 demonstrate a common feature among the three layers: some
smallest eigenvalues have the highest level of IPR and their IPR ratios deviate from
the average of fIPRk gk¼N
k¼1 . This indicates that their associated eigenvectors are more
localized, i.e. only a small group of banks contribute to them. Nevertheless, these
smallest eigenvalues are less than 1 and the contribution of their associated
factors/components to the total variance (recalling also Eqs. (12) and (13)) is
actually negligible. Moving on to the IPRs associated with the eigenvalues located
in the center of the eigenvalue distribution, we find that these eigenvalues often have
a relatively low level of IPR. In fact, these features of the smallest eigenvalues and
those located at the center of the distribution are analogous to what universally
found in similar analyses of stock return correlations (e.g. Plerou et al. 1999).
However, interestingly, we do observe a distinct feature for the eigenvector
elements of the largest eigenvalue: over the years, we find that IPR1 is typically less
than the average of fIPRk gk¼N k¼1 , showing that there are many banks contributing to
the eigenvector associated with the first largest eigenvalue. From an economic
perspective, this indicates a wide effect that the ‘‘market’’ mode has on the banking
system. Such a behavior still emerges in some of the next largest eigenvalues
deviating from the random bulk. This result implies that besides the global common
factor, some ‘‘groups’’ factors also play a certain role in explaining the correlation
dynamics. Furthermore, it also signals the presence of large communities (clusters)
of banks in which the members within each community (cluster) tend to be more
correlated in their lending activities, since they are influenced by some common
pervasive factors.
It should be emphasized that under a more detailed inspection, in each year, we
also observe a more localized behavior in at least one of the eigenvectors associated
with the largest eigenvalues other than k1 . In this case, the associated factor only
drives the co-movements among a smaller subset of banks. For instance, as shown
in panels (b), (f) of Fig. 5, in 2012, in the first and the third layers, IPR4 is relatively
larger than the average of fIPRk gk¼N k¼1 , while in the second layer, such a behavior can
be found in the case of IPR8 .14 Nevertheless, since we typically observe distinct
localized eigenvectors in different years, it is hard to track the economic imprints
responsible for the emergence of this behavior. We therefore leave this issue for
future research.
Furthermore, if the largest eigenvalue and the next largest ones stand for the
‘‘market’’ and ‘‘groups’’ modes (i.e. systematic and ‘‘non-systematic’’ factors),
respectively, it would be interesting to examine their effects by comparing the raw
14
In Fig. 13 in the ‘‘Appendix’’, we also show the visualization for a more localized eigenvector and a
more homogenized eigenvector as examples.
123
0.15 0.6
IPR IPR
avg. of IPR avg. of IPR
0.5
0.1 0.4
IPR( )
IPR( )
0.3
0.05 0.2
0.1
0 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12
(a) IPR, layer 1, 1980 (b) IPR, layer 1, 2012

0.3 0.5
IPR IPR
avg. of IPR 0.45 avg. of IPR
0.25
0.4
0.35
0.2
0.3
IPR( )
IPR( )
0.15 0.25
0.2
0.1
0.15
0.1
0.05
0.05
0 0
0 5 10 15 0 0.5 1 1.5 2 2.5 3 3.5 4
(c) IPR, layer 2, 1980 (d) IPR, layer 2, 2012

0.35 0.6
IPR IPR
0.3
0.5
0.25
0.4
0.2
IPR( )
IPR( )
0.3
0.15
0.2
0.1
0.1
0.05
0 0
0 5 10 15 0 2 4 6 8 10 12 14
(e) IPR, layer 3, 1980 (f) IPR, layer 3, 2012

Fig. 5 Inverse Participation Ratios of the eigenvectors of the correlation matrices in the three layers.
Panels a, b show IPR versus k in the layer 1, in 1980 and 2012. Panels c, d show IPR versus k in the layer
2, in 1980 and 2012. Panels e, f show IPR versus k in the layer 3, in 1980 and 2012. The red lines stand
for the averages of fIPRk gk¼N
k¼1
correlation matrix with the filtered ones. As can be seen in the panels (a) to (f) of
Fig. 6, after excluding the influence of the ‘‘market’’ mode on cross-correlation
matrix C, some significant correlations still remain in the filtered matrix FC m . Such
significant correlations are actually mainly driven by the ‘‘groups’’ modes
associated with the next largest eigenvalues. A further decomposition can be
implemented to split the ‘‘market’’ mode-filtered matrix into the correlations driven
123
548 D. T. Luu
1 1 1
0.8 0.8 0.8

20 20 20
0.6 0.6 0.6
40 0.4 0.4 40 0.4

40
0.2 0.2 0.2

Banks
Banks
Banks
60 60
0 60 0 0
-0.2 -0.2 -0.2

80 80
80
-0.4 -0.4 -0.4
100 100
-0.6 -0.6 -0.6
100
-0.8 -0.8 -0.8

120 120
120
-1 -1 -1
20 40 60 80 100 120 20 40 60 80 100 120 20 40 60 80 100 120
Banks Banks Banks
(a) C in 2012, layer 1 (b) C in 2012, layer 2 (c) C in 2012, layer 3

1 1 1
0.8 0.8 0.8

20 20 20
0.6 0.6 0.6
40 0.4 0.4 0.4

40 40
Banks
Banks
0.2 0.2 0.2

Banks
60 60
0 60 0 0
-0.2 -0.2 -0.2

80 80
80
-0.4 -0.4 -0.4
100 100
-0.6 -0.6 -0.6
100
-0.8 -0.8 -0.8

120 120
120
-1 -1 -1
20 40 60 80 100 120 20 40 60 80 100 120 20 40 60 80 100 120
Banks Banks Banks
(d) F C m in 2012, layer 1 (e) F C m in 2012, layer 2 (f) F C m in 2012, layer 3

1 1 1
0.8 0.8 0.8

20 20 20
0.6 0.6 0.6
40 0.4 0.4 0.4

40 40
Banks
Banks
0.2 0.2 0.2
Banks
60 60
0 60 0 0
-0.2 -0.2 -0.2

80 80
80
-0.4 -0.4 -0.4
100 100
-0.6 -0.6 -0.6
100
-0.8 -0.8 -0.8

120 120
120
-1 -1 -1
20 40 60 80 100 120 20 40 60 80 100 120 20 40 60 80 100 120
Banks Banks Banks
(g) F C m,g in 2012, layer 1 (h) F C m,g in 2012, layer 2 (i) F C m,g in 2012, layer 3
Fig. 6 Correlations and filtered correlations between banks in lending to firms in the three layers, in 2012
as an example. The three upper panels from a to c show the raw correlation matrices. The three panels
from d to f are the filtered correlation matrices, subtracting the influence of the ‘‘market’’ mode. The three
panels from g to i show the filtered correlation matrices, subtracting the effects of the ‘‘market’’ mode as
well as ‘‘groups’’ modes. In each layer, banks are sorted in descending order of eigenvector centrality of
the raw correlation matrix
by the ‘‘groups’’ modes and by noise. The ‘‘market’’ as well as ‘‘groups’’ modes-
filtered matrices FC m;g in three layers are shown in the panels (g) to (h) of Fig. 6.
We observe that almost all correlations are significantly subtracted in FC m;g .
All taken together, the results we have obtained so far demonstrate that the
analysis based on the methods of RMT and PCA can be used to identify important,
significant patterns in the correlations between banks in lending to firms in the other
sectors of the economy. In particular, we have shown that the correlations are
mainly driven by a global common factor influencing the lending of all banks and
some non-systematic factors that only affect the lending of a subset of banks in the
credit market of Japan. To gain a deeper understanding of the structure and the
dynamics of these correlations, we make one more step by combining these methods
with those of complex networks. More specifically, for each year, we use the PMFG
graph to extract the skeleton structure from the different parts of the cross-
correlation matrices, i.e. the part represents the effects of ‘‘market’’ mode (C m ), the
part captures the effects of ‘‘groups’’ modes (C g ), the part represents the effects of
123
(a) (b) (c)
(d) (e) (f)

Fig. 7 The interaction structure between banks for C m , C g , and C m;g obtained with the PMFG graph, in
layer 1
75 6 6
85 114
68 81
79
3 5 2
84 71 2 1 85 71
3
72 104
47 5 81 72
86 96 73
35 101 54 106
113 6 31 97 73 1
57 4
24 99 53 70
69
100 19 56 52 70 84 51
72 96105 112 74 96
92 55
91
25 23 82 58 46 86 79 78
69 102 63 28 55
37 88 50 70 67
4 65 64
66 8 57
105
4734 30
95 106 9 15
107 56 54
14 15 28 44 30 111 11
39 42 108 52 35
33
11 2 114
61 27
7 16 50 46
4143 109
45 101 36
41
66 63
111
104
98 68 60 59
58 53 89 40 113
48 109 67 73 9 100 38 39 11014 43
20
18
16 8 51 37 64 112 17
30 107 5817 61 78 49 90 40 62 48
107 69 31 44 19
97 37 26
43 13 19 35 48 50 110 93 109 42
52 59 92 66
47 12
104 103 95 91 94 83 61 62 111 2321
60 62 42 108 59
26 99
10 63 114 38
11010 27
45 21 105 39
80
33 4177 57 24 75
29 101 18 25 12
20 36 106 102 92 22
3640 18 5 15 78 77 13 67 65 29
32 7 20 5671 46 16
8 88 82
64 26 38 23 9
32 77
60 93
80 85 82 108 91
83 51 4 1 34 113 7
74 83 76 49 94
55 81 79 98
89 5374 22 65 76 11 10 99 68
4498 94 88 80 84 45
24 95
87 13 12 25 87 90
21 31 97 103
3 32 87 27 28
9354 89 33
4976 103
34 14 17 22 29
102 90 75 86 100
(a) “market” mode, in 1980 (b) “groups” modes, in 1980 (c) “market” and “groups” modes, in 1980
88 104 72 5
36 46 1
1 112
1 102 4
112 56 36 37 73
13
71 102 73 5 55 48
79
111 10 63
17 28 52 70113
51 86 77 7 8 53 101
8 2 396170 15 2
101 74 51 35
112 83 70 113 32
4 75 64 20 92 27 4 50 100 34
29 12 56 82 16
106 103
88 31 33 16 20
27 62 97 78 79 10 99 89 103 83
89 57 21 108 98 17 25 54 77
44 23 46 74 82 106
109 5811 84 47 76 49 64 87
41 115 45 25
5 14 3 78
24 100 78 6 15 28 23 13
30 59
53 26 72 90
3520 41 26 80 17
94 9 96 86 71 7
43 8 21
40 95 29 114 85
19 54
16 96 23 37 11 30 19
18 9 24 30
91
98
60 66 52
72 36
21
42
81 27 12 7911575
69 50 93 110 87 61
107 95 80 80 22
94 93 22 10 11 39
64
43 18 65 7 92 297658 47 38 93 109
63
35 31 73 8118 75 3389 14 66 67 68
47
113
45 19 13 59 62 65
14 34 38 55 40 6 82 37 39
31 88 41 60 107
74 90 85 15 92 108 110 111
49 105 97 91 3812 34 53 69
9440
99 81 22 26 102
77 105 96
7642 28 3 101 95
48 114 24 9 71 32 42 46
25 2 100
52 43
84 57 51 6
49
83 3 60
68 111 50 44
33 99 45
91 86 106 65 98 84
55 48 97
44 66 56
103 114 87 62 69 57
59 63 58 67
32 107 90
108 109 104 54
115 104 85 61 110 105
(d) “market” mode, in 2012 (e) “groups” modes, in 2012 (f) “market” and “groups” modes, in 2012
layer 2
123
550 D. T. Luu
48
67 53 57 52 104 67
88 33 56
55
87
114 85 51 9778
96
67 96 44
65
6 4068 68 9
110 28
16
5 33 63111 89
7 73 84 40 39
3 71 64 27
37 59 1 85 7 88
41 3164 38 59
35 32 2 1566 47
87 30 264 76 70 9 42 41 14 101 34
48 36 29 112 113 74 76
27 20 83 11572 43
16
46
52
491353 5612 81
21 86 95
14
28 18 19 23 8 25 102 105 51 37
87
78 22 17
78 106 4243
24 79 54 12 75 89 777510 29
107 107 36
114 628
25 80 17
10 47 95 55 86
9 44 77
16 11 100 82 77 13 3445 57 82
24
112 30
5
30 947534 91 81 31 44 94 54 21 2
806855 63 84 8 73 101 50 3526 4 72
3 84
33
46 109
61 45 32 23 19 70
70 90 108
83 20
95 92 11 39 88 49 93 60 10448 73
4 109 99 1
69 102
28 19
49 26 22 104 39 14 62 58
3
108 72 27 2 38 103 105 85 98 71
112 57 114 92 69 9493 90 18 115
52 24
64 37 98 115
106
38 79 22 80 113
89 110 1596 51 58
60 12
29 21 108
6
65 74 60109
17 8213103 23 66 54 40 102 716 106
7 62 63 74 50 100
11
5 79 42 58110 15
46 61 10153 90 45 61
98 69
111 91
41 66
36 50 97 111
18 31 20 56 47 92
65
93 1 83 43 99
10 91 25 81 35
99 113
105 97 103
(a) “market” mode, in 1980 (b) “groups” modes, in 1980 (c) “market” and “groups” modes, in 1980
78 49 72
152 71 115
87 4
6 69 18 6768
64 56 43
22
32 963046 66 80
4 34 79 11370
73 98 35 1819 11 75 6
775347
65
26
1 71
73 3
5
6
9760 8426 81 78 5
33 112 3573
93 8311 51 21 82 10
80
22 76 24
14 2023 4
30
2 58 114
48 62 28 31 71 70 17 29 99 101 48
86 9 113 105 2 44 102
1 8 15 115 69 111 86
41 16 12 60
29 7581 100 65 83
112 251
53
14
88 87
70 72 13 64
66
45 50 98
105
16 67 13
14 106 15 51 52
2468
63 101 25 62 8 28 96
104 55
36 5 9 8 95 3893 107 16 100
25 40
17 37
3 114
84
94 61
59 9 27 97 10347
22
57 54
85 97 99 91 63 76
11010 112
82 28
27
83
103 104 92 39 110 78 89 84
7 44 109 108 19 31 56
91 114 27 96 54 47 43 46 4241 20 67 80 3432 88
45
58107 72 99 57 40 68
106 86 17 79 75
39 90 37
33
3 109
38
77 55 98 56 48 95
87 81
40
9255 52 106104
79 50 51 43
57 50 100 30 18 12
7 53
59 4921 89 54 105
102 36 90 24 21
11
29
91 46
103 108 19 85 74 10152 10 49 77 42 95

42
76 61 82 90 37
20 26 41
12 74 111 61 31 88 64 74 7 35
113
115 23 45 94 69
62 65 23 94
13 32 63 85
33 60
66 111
89 34 58 38
108 36 39
102 109
59
107 92
44 110 93
(d) “market” mode, in 2012 (e) “groups” modes, in 2012 (f) “market” and “groups” modes, in 2012
layer 3
both ‘‘market’’ as well as ‘‘groups’’ modes (C m;g ¼ C m þ C g ). We then study the

persistence and changes in the skeleton structure over the years.
For the illustration purpose, in Figs. 7, 8 and 9, we show PMFG graphs extracted
from different parts of the correlation matrices in the two years 1980 and 2012. In
each panel of these Figures, each number corresponds to a bank, and each link (in
green color) indicates the presence of a significant relation between a pair of banks.
The numbers in red color represents the hub banks that have the highest levels of the
degrees in the PMFG graph. Overall, the correlations between banks reveal different
backbone structures in different periods. Indeed, by having a closer inspection over
the years (for example, see Figs. 17 and 18 in the ‘‘Appendix’’) we often observe
that when the dominance of a group of banks in one period gradually disappears, the
credit market starts to build-up a different structure in the next period in which
another group of banks emerge as the key players in the backbone of the cross-
correlations.
123
4 Concluding Remarks
In this paper, we have examined the structure and the dynamics of the correlation
matrices for the banks’ loan portfolios in a large data set of the bank-firm credit
network of Japan during the period from 1980 to 2012. Our results show that only a
subset of the eigenvalues contain genuine information while the rest actually
corresponds to noise or just have a small effect on the correlations between banks in
lending to firms in the other sectors of the economy. This implies that the methods
of RMT and PCA can be used for filtering noise from the empirical correlation
matrices of banks’ loan portfolios.
The dynamics of the largest eigenvalues and the associated absorption ratios
reveal different milestones in the credit market of Japan over the years. More
specifically, there is a dramatic increase in the level of systemic risk during the
period of the Japanese asset price bubble (1986–1991). In contrast, we observe a
significant decrease in all of these eigenvalues and the associated absorption ratios
in 2009, which might be considered as a consequence of the economic contraction
in Japan hit by the global crisis (e.g. Kawai and Takagi 2011).
Based on the eigenvector elements, we have further investigated the localization
and clustering behaviours of banks. One the one hand, we find that in the case of the
eigenvectors corresponding to a group of smallest eigenvalues, there are only few
banks contributing to them. Such a localization behavior is similar to what is often
found in analyses of stock price changes (e.g. Plerou et al. 1999). However, since
these eigenvalues are smaller than 1 and the top largest eigenvalues, the contribution
of their corresponding principal components to the total variance is actually
negligible. On the other hand, we find that the eigenvectors corresponding to the top
largest eigenvalues are typically not very localized. This indicates that latent factors
extracted from them have a more pervasive effect on banks in providing credit to the
firms.
Interestingly, building Planar Maximally Filtered Graphs (PMFG) from the
correlations driven different eigenmodes, we find that the local interaction structure
between banks changes in different periods. More specifically, when the dominance
of a group of banks in one period gradually disappears, the credit market starts to
build-up a different structure in the next period in which another group of banks
become the new hubs in the backbone of the cross-correlations.
Several directions for future research can be suggested from our present work.
First, as we can see in the analysis of eigenvector elements, it is often observed that
in a couple of the eigenvectors corresponding to the top largest eigenvalues, a large
group of banks tend to have a higher degree of correlations among themselves.
Therefore, the clustering behaviours and the community structure of banks in the
correlation matrices in the different lending layers should be studied further (e.g.
Almog et al. 2015; MacMahon and Garlaschelli 2015). Second, it would also be
interesting to examine the effects of the bank characteristics (e.g. bank locations,
bank types, balance sheets’ information) on the formation of such clusters and
communities. Last but not least, another important direction for future research is to
analyse the multilayer architecture of the interdependencies between banks, e.g. to
123
552 D. T. Luu
consider together various layers including the network of loan portfolio overlaps,
the network of loan portfolio correlations, the correlation network of bank stock
returns, and the physical interbank trading network (e.g. Brunetti et al. 2015;
Montagna and Lux 2017). We believe that this direction will help to gain a deeper
understanding of the complex structure of interactions within the banking system
and between the banking system and the other sectors of the economy.
Appendix
Here we provide additional results for portfolio overlaps and similarities, the
fractions of eigenvalues in different ranges, the distributions as well as the evolution
of the eigenvector elements, and the more detailed correlation network structure
between banks based on the method of PMFG. We also report the main results for
the binary version of the bank-firm credit network of Japan.
Portfolio Overlaps and Similarities
In this part, we briefly summarize the main results for portfolio overlaps and
portfolio similarities between banks. Recall that in each lending layer, each element
of the binary version of the bank-bank projection matrix ABB explained in Eq. (3)
implies whether a pair of banks share at least one common borrower. Meanwhile,
each element of the similarity matrix SBB , which is based on the generalized
Jaccard index defined in Eq. (1), indicates the degree of structural similarity
between two loan portfolios. Figure 10 demonstrates the binary version of the bank-
0 0 0
20 20 20
40 40 40
Banks
Banks
Banks
60 60 60
80 80
80
100 100
100
120 120
120
0 20 40 60 80 100 120 0 20 40 60 80 100 120 0 20 40 60 80 100 120
Banks Banks Banks
(a) Overlaps, layer 1 (b) Overlaps, layer 2 (c) Overlaps, layer 3

1 1 1
0.9 0.9 0.9

20 20 20
0.8 0.8 0.8
40 0.7 0.7 40 0.7

40
0.6 0.6 0.6
Banks
Banks
Banks
60 60 60
0.5 0.5 0.5
0.4 0.4 0.4

80 80
80
0.3 0.3 0.3
100 0.2 0.2 100 0.2

100
0.1 0.1 0.1
120 120
0 120 0 0
20 40 60 80 100 120 20 40 60 80 100 120 20 40 60 80 100 120
Banks Banks Banks
(d) Similarity, layer 1 (e) Similarity, layer 2 (f) Similarity, layer 3

Fig. 10 Binary overlaps and portfolio similarities between banks in the three layers, in 2012. Banks are
sorted in descending order of the eigenvector centrality of the adjacency matrix obtained from ABB .
Panels a to c The binary version of the bank-bank projection matrices in the three layers. Each blue point
in these panels indicates a pair of banks have at least one common borrower. Panels d to f The similarity
matrices in the three layers. The colorbar indicates the range [0, 1] for the elements of the similarity
matrices .(Colour figure online)
123
100 100 100

larger than * in [ * , * ] less than *
90 max 90 min max 90 min
larger than all in [ * , all ]
80 max 80 min max 80
fraction of eigenvalues (%)
fraction of eigenvalues
70 70 70
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years years
(a) fractions of λ, layer 1 (b) fractions of λ, layer 1 (c) fractions of λ, layer 1

100 100 100
* * * *
larger than max in [ min, max ] less than min
90 90 90
all * all
larger than max in [ min, max ]
80 80 80
70 70 70
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years years
(d) fractions of λ, layer 2 (e) fractions of λ, layer 2 (f) fractions of λ, layer 2

100 100 100
* * * *
larger than max in [ min, max ] less than min
90 90 90
all * all
larger than max in [ min , max ]
80 80 80
70 70 70
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years years
(g) fractions of λ, layer 3 (h) fractions of λ, layer 3 (i) fractions of λ, layer 3
Fig. 11 Fractions of eigenvalues in different ranges for the three layers. Panels a, d, g Fractions of
eigenvalues larger than the considered upper bound kmax (blue lines) or kall max (red lines) for layers 1, 2, 3,
respectively. Panels b, e, h Fractions of eigenvalues lie within ½kmin ; kmax (blue lines) or ½kmin ; kall
max (red
lines) for layers 1, 2, 3, respectively. Panels c, f, i Fractions of eigenvalues less than the lower bound kmin
for layers 1, 2, 3, respectively .(Colour figure online)
bank projection matrices (panels (a) to (c)) and the similarity matrices (panels (d) to
(f)) in the three lending layers, for the data in 2012 as an example. We observe that
many pairs of banks have overlaps in their loan portfolios, especially in the layer 1
and layer 3. Furthermore, the loan portfolios also tend to have a relatively high
degree of similarities.15
Fractions of Eigenvalues in Different Ranges
Figure 11 shows the fractions of eigenvalues in different ranges, i.e. larger than the
considered upper bound (kmax or kall 1 2
max ¼ maxðkmax ; kmax ; kmax Þ), within the interval
from the lower bound to the considered upper bound (½kmin ; kmax or ½kmin ; kall

max ), and
less than the lower bound kmin . Although the fractions change over the years, in
15
To illustrate the similarity matrices, for the sake of convenience, we still use the same order of banks
as in the analysis of binary overlaps in Fig. 10a–c, i.e. banks are sorted in descending order of the
eigenvector centrality (e.g. see Bonacich (2007)) of the adjacency matrix obtained from ABB . We do not
conclude any nature of the relationship between that order and the elements of the portfolio matrix SBB .
123
554 D. T. Luu
general we find that a majority of the eigenvalues lie within the random bulk or are
smaller than the lower bound kmin . In contrast, only a small subset of the top
eigenvalues persistently exceed the upper bound kmax or kall
max over the whole period.
1 1 1
0.9 RMT prediction 0.9 RMT prediction 0.9 RMT prediction
0.8 0.8 0.8
0.7 0.7 0.7

relative frequency
relative frequency
relative frequency
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
-10 -8 -6 -4 -2 0 2 4 6 8 -10 -8 -6 -4 -2 0 2 4 6 8 -8 -6 -4 -2 0 2 4 6 8
eigenvector components eigenvector components eigenvector components
(a) uN , layer 1 (b) uN , layer 2 (c) uN , layer 3

0.4 0.4 0.4
RMT prediction RMT prediction RMT prediction
0.35 0.35 0.35
0.3 0.3 0.3

relative frequency
relative frequency
relative frequency
0.25 0.25 0.25
0.2 0.2 0.2
0.15 0.15 0.15
0.1 0.1 0.1
0.05 0.05 0.05
0 0 0
-4 -3 -2 -1 0 1 2 3 -4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4
(d) uRM T , layer 1 (e) uRM T , layer 2 (f) uRM T , layer 3

0.4 0.4 0.4
RMT prediction RMT prediction RMT prediction
0.35 0.35 0.35
0.3 0.3 0.3

relative frequency
relative frequency
relative frequency
0.25 0.25 0.25
0.2 0.2 0.2
0.15 0.15 0.15
0.1 0.1 0.1
0.05 0.05 0.05
0 0 0
-5 -4 -3 -2 -1 0 1 2 3 4 -3 -2 -1 0 1 2 3 -4 -3 -2 -1 0 1 2 3 4
(g) u4 , layer 1 (h) u4 , layer 2 (i) u4 , layer 3

0.45 0.4 0.4
0.4 RMT prediction RMT prediction RMT prediction
0.35 0.35
0.35
0.3 0.3
relative frequency
relative frequency
relative frequency
0.3
0.25 0.25
0.25
0.2 0.2
0.2
0.15 0.15
0.15
0.1 0.1
0.1
0.05 0.05 0.05
0 0 0
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
(j) u1 , layer 1 (k) u1 , layer 2 (l) u1 , layer 3

Fig. 12 Distributions of the eigenvector elements of four selected eigenvalues of C, in 2012, in the three
layers. Panels a to c show the distribution of the eigenvector elements of uN . Panels d to f show the
distribution of the elements of an eigenvector in the random bulk. Panels g to i show the distribution of
the eigenvector elements of u4 . Panels j to l show the distribution of the eigenvector elements of u1
123
0.25 0.2
0.2 0.1
0.15 0
1
4
Eigenvetor of
Eigenvetor of
0.1 -0.1
0.05 -0.2
0 -0.3
-0.05 -0.4
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Bank Bank
(a) Evec. 1 in 2012-layer 1 (b) Evec. 4 in 2012-layer 1

0.25 0.25
0.2
0.2
0.15
0.15
0.1
1
4
Eigenvetor of
Eigenvetor of
0.1 0.05
0.05 0
-0.05
0
-0.1
-0.05
-0.15
-0.1 -0.2
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Bank Bank
(c) Evec. 1 in 2012-layer 2 (d) Evec. 4 in 2012-layer 2

0.25 0.2
0.2 0.1
0.15 0
1
4
Eigenvetor of
Eigenvetor of
0.1 -0.1
0.05 -0.2
0 -0.3
-0.05 -0.4
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Bank Bank
(e) Evec. 1 in 2012-layer 3 (f) Evec. 4 in 2012-layer 3

Fig. 13 The eigenvectors (sorted in descending order) of k1 and k4 , in the three layers, in 2012. Without
loss of generality, we fix the signs of eigenvector elements such that their sum is non-negative. Panels a, b
show the eigenvectors of k1 and k4 in the layer 1. Panels c, d show the eigenvectors of k1 and k4 in the
layer 2. Panels e, f show the eigenvectors of k1 and k4 in the layer 3
Additional Results for Eigenvector Elements
Distribution of Eigenvector Elements

In Fig. 12, we compare the distributions of the elements of the four selected
eigenvectors of C, i.e. the eigenvector uN associated with the smallest eigenvalue
123
556 D. T. Luu
1 0.5 1 0.5
11 0.45 11 0.45
21 21
0.4 0.4
31 31
0.35 0.35
41 41
0.3 0.3
51 51
0.25 0.25
61 61
71 0.2 71 0.2
81 0.15 81 0.15
91 0.1 91 0.1
101 101
0.05 0.05
111 111
0 0
19801982198419861988199019921994199619982000200220042006200820102012 19801982198419861988199019921994199619982000200220042006200820102012
years years
(a) Elements of u1 (b) Elements of u2

1 0.5 1 0.5
11 0.45 11 0.45
21 21
0.4 0.4
31 31
0.35 0.35
41 41
0.3 0.3
51 51
0.25 0.25
61 61
71 0.2 71 0.2
81 0.15 81 0.15
91 0.1 91 0.1
101 101
0.05 0.05
111 111
0 0
19801982198419861988199019921994199619982000200220042006200820102012 19801982198419861988199019921994199619982000200220042006200820102012
years years
(c) Elements of u3 (d) Elements of u4
Fig. 14 The evolution of the elements of the eigenvectors corresponding to k1 (panel (a)), k2 (panel (b)),
k3 (panel (c)), and k4 (panel (d)), for 115 banks activating in all years, in the layer 1
kN , the eigenvector uRMT corresponding to a representative eigenvalue located in the

random bulk, the eigenvector u4 associated with k4 , and the eigenvector u1
corresponding to the largest eigenvalue k1 . Note that for the illustration purpose, in
PN 2
this Figure we re-normalize the eigenvector elements such that j¼1 ui ðjÞ ¼ N
(8i). We can see that uN has only few significant elements, revealing that this
eigenvector is very localized. In addition, the distributions of the elements of u1 and
u4 are different from the distribution of the elements of uRMT .
Localization Behavior
As two examples, with one for a more localized behavior of eigenvector elements
and another for a less localized behavior of eigenvector elements, Fig. 13 shows the
entries of u1 and u4 in descending order. We can see that in the first and third layers,
the homogeneity degree is much stronger in u1 than in u4 . However, it should be
emphasized that this does not implies that all banks contribute equally to the factor
extracted from k1 and u1 . Indeed, as shown in Fig. 13a, c and e, we still observe a
certain level of heterogeneity among banks.
123
1 0.5 1 0.5
11 0.45 11 0.45
21 21
0.4 0.4
31 31
0.35 0.35
41 41
0.3 0.3
51 51
0.25 0.25
61 61
71 0.2 71 0.2
81 0.15 81 0.15
91 0.1 91 0.1
101 101
0.05 0.05
111 111
0 0
19801982198419861988199019921994199619982000200220042006200820102012 19801982198419861988199019921994199619982000200220042006200820102012
years years

1 0.5 1 0.5
11 0.45 11 0.45
21 21
0.4 0.4
31 31
0.35 0.35
41 41
0.3 0.3
51 51
0.25 0.25
61 61
71 0.2 71 0.2
81 0.15 81 0.15
91 0.1 91 0.1
101 101
0.05 0.05
111 111
0 0
19801982198419861988199019921994199619982000200220042006200820102012 19801982198419861988199019921994199619982000200220042006200820102012
years years

Evolution of Eigenvector Elements

From Figs. 14, 15, 16, we show the evolution of the eigenvectors (in absolute
terms) associated with the top four largest eigenvalues in the three lending layers.
Here we focus on 115 banks that activate in the bank-firm credit market of Japan in
all years from 1980 to 2012. As we can see, some banks correspond to a higher level
eigenvector elements in certain sub-periods, but overall, in almost all years, the
distribution ranges of these elements are not very wide. This result is consistent with
those obtained from the analysis of IPR.
123
558 D. T. Luu
1 0.5 1 0.5
11 0.45 11 0.45
21 21
0.4 0.4
31 31
0.35 0.35
41 41
0.3 0.3
51 51
0.25 0.25
61 61
71 0.2 71 0.2
81 0.15 81 0.15
91 0.1 91 0.1
101 101
0.05 0.05
111 111
0 0
19801982198419861988199019921994199619982000200220042006200820102012 19801982198419861988199019921994199619982000200220042006200820102012
years years

1 0.5 1 0.5
11 0.45 11 0.45
21 21
0.4 0.4
31 31
0.35 0.35
41 41
0.3 0.3
51 51
0.25 0.25
61 61
71 0.2 71 0.2
81 0.15 81 0.15
91 0.1 91 0.1
101 101
0.05 0.05
111 111
0 0
19801982198419861988199019921994199619982000200220042006200820102012 19801982198419861988199019921994199619982000200220042006200820102012
years years

Additional Results for PMFG Graphs Over the Years
In the following we provide additional results for the evolution of PMFG graphs
based on the matrix C m;g ¼ ðC m þ C g Þ, which captures the effects of significant
factors (i.e. the global common factor as well as the ‘‘groups’’ factors) on the
correlations between banks in lending to firms. Again, we focus on 115 banks
activating in the bank-firm credit market of Japan in all the years from 1980 to 2012.
For the sake of conciseness, here we only show the PMFG graphs in the layer 1 (the
total lending layer).16 In each panel of Figs. 17 and 18, the green links indicate
significant pairwise correlations retained in the PMFG graph. The numbers
represent banks, and those in red color are the banks that have highest degrees in the
PMFG network. We can see that the correlation structure between banks changes in
different periods. In particular, when the dominant role of a group of banks in one
period gradually vanishes, the credit market starts to build-up a different structure in
the next period in which another group of banks will become the new hubs in the
backbone of the cross-correlations.
16
However, the results for the other layers are available from the author upon request.
123
6 98 59 102
96 45 89 101
5 108 44
3 42 62 69
29
1 115 111 102104 40 53 52
43 43 98 56
2 54 6 34 33
96 40 109 47 48 108
42 3081 85 112 4 73 51 46 58
79 72 110 62 5
26 59 69 101 37
7835 36 71
54 46
38 92 70 107 61 56 36 104
11155
61 64
41 52 55 2 1 106 60 66 65
47 37 86 53 115 7341 19 35 109
107 71
3
93
1915 113 32 58 51 110
39 57 30 26 22 4
50 49 72 71 23 57
20 70
113 18
24
2 73 70 113
1874 94 9 28 88 31 64
48 20
80 112 50 63 26 21
6 115 1
21 25 68 3 21 29
65 4 84
32 17 22 91
84
24 27 63 66 60
82
33 5 72
20
88 808 23 83
38 39
92 88 32 49 23 25
99 87
111
795 82 76 13 75 7 16
31 34
16 36 19 24 82
31 37 80
28 22
110 45 14 12 77
8427 67 94 67
74 7 41 30 35
90
18 29
10 8 95 93 9
109 62 16 98
86 74 9 91 43
107 61 64 114 106 14 28 14 40
87 112
102 44 25 83 11 15 8 46 4738
58 10 15 79 27 42
69 59 63 11 86
54 39
53 87 48 76 44
105 66 76
50 90 13 11 10 96 92
6534 12 93
101 60 57 106 77 12
100 17 94
100 49 52 56 33 99 13 81
97 67 97 75
90 55 108 83 95
68 114 105 17 45 85
103 78 75 77 99 91
89 103 81 78 97
79 114
105 103
51 89
104 85 100 68
(a) “market” and “groups” modes, in 1980 (b) “market” and “groups” modes, in 1981 (c) “market” and “groups” modes, in 1982
59 81 88 95
10262 86
85
69 77103 94
101 114
108 107 106
32
33
93
97 37
99 927638
53 91 96
55 58 48 104
104 44 100 31
110 39
56
52
61 110
100 109 95
98
90 98 89 15 44
88 36
54 103 4548 49 35 54
92 28
51 94 44 34 105 9 76 41 45 32
96 45 106 27 13 47 80
64 78 46 30 29
93 91 79 40
50 96 8 68 31 43 22 81 112
63
38 42
43 57 14 10
12 33 9868 18
66 47 51 97 75 42
11165 60 57 39
55 101 16 11 19 83 24
40 74 34 48
50 46 114 85 74
84
7
49 74 56
102 52
82
114 20 25 26 3
89 413067 54 20 55 51 4
17 97 89 7
33
37
34 90 9 15 53 24 82
16 57 17 23 21
31 16 7 87 2321 113 50 106 113 70 72
87 92 67 9 85 115
88 32 36 35 8 112 83
49 79 87
14 10 11 94 93 91
47
2919 22 18 103 8 14 101 1
2 71
17 27 83
39 30 35
58
37 84 80
8125 11 52 73
75 95 46 10
28 15 105
19 12 38 42 110 56
64 36 13 27 5 6
18 13
76
78 43 61 26 77
53
20 68 79 41 86
28
12
86 40 99
69 104
80 78
84 62 64
77 99 111 63 59 58
2623 25 24 22 65 60 70107 4 75 102
72 66 5 108
115 70 113 82 29 105 73 69 61 107
21 109 111
71 67 2 60 63 66
1 4 109 3
2 73 112 65
3 71
6 1 62 59 90
108 11572
5 6 100
(d) “market” and “groups” modes, in 1983 (e) “market” and “groups” modes, in 1984 (f) “market” and “groups” modes, in 1985
98 88 87 3 93
45 91
32
68 4 71 111 67
66 38
94
60 39 78
31 34 79 78 72 112 11 92
48 76
44 96 33 70 65 15 13
47 12 6
17 61 64 110 10 111
115 45
57 51 1375
73 69 95 86 77
62 58 109 28 43 99
54 5 96 108 59 27
57 63 107 12 41 48
7
9 8 47 65 110
104 14 89 100 2 98 44
50 37 10 114 1 52
53
79
32
42
66 59
55 55
16 56 99 102 14 40 36
87
42 46
43 101 24 82 51 74 90 37 17 88
46 35 108 112
67
56 52 49
83
36 87
83 84 23 50 49101
61 109
105 40
8 28 33 54 16 30 107
102 11 105104 100 60 34 62 68
27 35 29 26 48
112 80 53 68 31 1922 25
21 89 106 7 25 31 33 19
96 85
81 30 24 9
40 34
30 81 20 100
22 18 35
41 36 15 113 20 21 63
82
29
58 19 25 44 47 46 80 18 114
83 58 69
18 22
80
75
64 74 81
77 45 41 95 84
95 69 38 20
84
64 23 113 24
60 82 23 106 76 38 42 43 75 52
108 61
39 99 7
16 50 101 26 29
107 63 94 86 14 37 12
39 9 53 102
21 90 94 92 5457 56 5 6
109 66 62 9265 67 113 26 90 74 51
73 27 55
110 70 1
70 93 28 8 15 17 98
106
73
5 115 91 11 10
59 13 103 49
103 93 6 89
91 2 71 105
85 77 115
111 1 79 2
3 4 86 76 104
114 88 103
32 78 3 71 72
97 85 4
72 97 97
(g) “market” and “groups” modes, in 1986 (h) “market” and “groups” modes, in 1987 (i) “market” and “groups” modes, in 1988
93 83 84 84
111
85 76 83
94 84 81
92 112 79 99 77 85 78
95 83 114 68
91 81 107 109 65 107 103
10 111 110 99
87 75
11 110
13 15 109 79 77 12
12 36 44 98 108 75 78 112 62 87 108
17 9 77103
82
76 35
81 2159
606166
68
8 25
42 67
15
32 74 38 59 45
67 47 45 100 9
28
2788 16 100 31 18 19
30
62
14 66 49 88 32 34 96 91
46 89
48
69 64
74
42 40 39 60
40 41
7 68 43 38 33 97 51
86 65 80 43 61 8
39 98 114
50 92 95 102 5856 54
94 50
101 80 66 65 26
41 51 63 23
98
45 96
52 35 22 90 93 52
53 20
47 54 55 89 46 55 114 14
44 33 106 34
29 49
53 56
58 64 49
10155
56 19 37
36 82 69 63 96 16 27 6 4
31 57 51 105
72
10253
3024 18 52 5457 88 13 115
5 113
64 58 26 101 105
57 1 10
104 105 69 74
16 29 102 106
113
32 33 50104 25 17 3
11 2 104 70
20 63 86 12 7
26 23 72 31 82 28 11
23 60 28 27 8 149 17 20 115
73 1 6
34 44 22 73
4 61
10 24 7071 46106 24 86 71
62 78 25 18
113 75 15
22
5 47 19 80 72
3 109 21 13
6 70 59 87 4
5 1 76
3
95 41
79 91 30
73
108 39 38
48 67 94 35 29
85 48 36
2 110 100 92
97
115 111 89 42 37
72 103 93 43
107 99 21 37 90
112
90
71 97 40
(j) “market” and “groups” modes, in 1989 (k) “market” and “groups” modes, in 1990 (l) “market” and “groups” modes, in 1991
45 91 91 97
92 93
97 93
103 92 95
89 94
94 40 95 39 38
39 38 44 42
67 43
81 83 47 22 46
42 47
111 109 100 41 19 45 41
78 44 43 18
107 30
99 76 77
46 3029 80
85 113
110 15 108 12 62 65
35 36 23 106 89 354 36
90
37
66 57 105 106 115
87 75 11
59 104 20 104 37
68
79
10 60 69 102 101 90 105 57 6 1 70 88 3
61
54 55 20 80
21 70 5 73 72 32 19
14 58 6452
55 105 113 3 4056 98
99 71 26 22
63 26 49
13 56 63 89 115 96 51
52 2 33 31 112 84 18
8 109 49 1 4
53 49 54
57 60
104 106 71 101 50 34 7 23
17 9 8851 98 54
73 72 64 53
7 48 56 96 58 2 16 148
28 27
16 74 32 98 68
48 55 101 61 5350 5 62
63
69 58 74 25 13
10
113 1 2 112 100 51 102 69 6 29
17 12
6
86 50 233133 45 67
114 6452 60 61 87 28 24
5 73 62 112 9 27
86
115 70 82 22 65 66
7 65
66 11
34 79
71 72 46 114
47 20 80 24 34 16 111
75
96 84 33 84
74 86 21
78
76 77 15
4 26 82 28
3 95
18 31
14
25 27
82
30 111 8 110
25 32 85
44 39 41 19 107 88
17 83 81
91 99 13 108
87
29 110 85
92 94 38 35 59 11 10 103 107
40 79 83 12 59
93 42 36 37 103 75 10877 114
43 102 100 109
97 81 76 9
78 15 48
24 67
90 21 68
(m) “market” and “groups” modes, in 1992 (n) “market” and “groups” modes, in 1993 (o) “market” and “groups” modes, in 1994
87 25 108 82
22 37 114
81 90 24
96 90 25
87 78 85 111
107 109
19 80 91 96 8586 77
12 36 86 79 83 103 84
3 30 77 82 97 29112 98 83
26 4 35 24 27 84 75
17 28 104100 45 89 97
7615
1874 45 95
27 21 9 29 115 10 48 89 98 12
86 10
12 11 101 105
106 100103
14 75
70 29 16
99 56 57 44 79
104
44
23 1 72 9 55 21
28 13 158 20 71 21 51
82 73 54 78 106
11 105 7 49 33
17 7 104 74 8 50
52 87 32
5 31
24
16 84 98 13 14
7876 2 101 53 61 34 88
6 102
31 112 55 2 6963 58 6062 108
113
56 57 48 112 65
59 107 67
65 50 92 94 8
33 88 64 50 106 54 73 111 68 9 76
40 109110
66 110 99 93
47 43
67 5846 69 52 51 5 72
70 64 67 66 59
13
79 40 115 46 38 15
99 1 2
66 60 62 4132 53 6
4 71 68
61 101 49
39
42 74 10
68 61 96 62 60 63 53 40
100 107
65 59 114
3
113 26 23 14
63 75 69 102 41 7 27
25
109 42
43 47 49
102 20
80 71
11573
1 30 35
17
28
34 108 58 11
72 16
19 22 52 4 70 113
23
89 45 39 93 31 18 5 6 20 36
91 32 33 30 36
9294 81 83 88 34 46 37
64
3
18 37
44 38 77 47
35 105 26
19
95 41 56
43 51
103 95 42 80 81
111 85 38 54 5755
110 114
39 22
90 92 94
97 93 91 48
(p) “market” and “groups” modes, in 1995 (q) “market” and “groups” modes, in 1996 (r) “market” and “groups” modes, in 1997
123
560 D. T. Luu
b Fig. 17 The interaction structure between banks for the correlation matrix C m;g obtained with the PMFG
graph, in layer 1, from 1980 to 1997
57 76 81
54 26 43 80 22
32
15 9
51 88 22
81 6 87
55 33 8 70
4
42 19 82
31 13 80 18
56 11
71 3 5
72
99 22 17 82 46 24 2930
59 65 105 74
7 10 115 73
75 12 61 47
61 52 64 66
60 82 114 14 99 21 78
79 67
63 1
2 23
62
49 50 63 23
24
80 81
83
85 68 62 113112 20
77 1025049 99 60
102101
58 86 16 110
89 84 49
69
58 37 28
12 86 59 102
83 6 111 50 52 17
69 27 72 63 64 96
106 91
51 38 53 66 36 35 16 27
53 29 85 28 23 5 65 51 64 110 41 76
104 25 95 2671 66 53 55 52 77 101 65 109
14
90 21 93 54 56
40 34 96 19 4 40 56
54 97 57 104
103 45 115 70 69 95 55 111 3938
5 111 98 91 36 61 101 39 57 105 107 10
106 100 97 18113 2 98 25
48 7887 41 20 1 73 58 62 105 48 92 94 93 1311
4 73
70 2 84 44 35
24 112 108
43 112 109 107 3 59 60
103 106 90
113
72 10889 37 46 84 114
42 26 1 71 30 91 94 8 77
46 75 45 86 44 92
90 44
3
6 110 67
29 34 107
100 97103 45
100
95
15
115 87 43 47 7 108 78
104 96
47 42 21 9
31
33 85 34
18 20 98 40
30 68 25 83 89 114
75 48
35 16 79 7 74
74 109 12
38 19 76
39 41 28 88
94 36 14 17 11
37 27 32 33 31
92 93 9
10 15 68
8
13 88
79 67 32
(a) “market” and “groups” modes, in 1998 (b) “market” and “groups” modes, in 1999 (c) “market” and “groups” modes, in 2000
96 88 34 32 67
77 45 86 68 81
67 22
1 29
98 68 21
85 44
109 91 97 71 33 19 26
89
36
2 109 59 7030 115
110 90 31
1 61 62 18 71
63 60 103 54 72
61
37 6
72 60
2073 4
65 66 115 57 40 23
104 114 5 17
58 105 100
108
80
64 66 74 28 24
16113 5
62 69 47
69
68 73 5856 7
74
27 10 2 82
59 64 46 6 59
102 107 83 3 63 79 88
32 56 52 25
4
65 55 78 34
31 7 14 109
53 111 26
113 53 51 9 32
33
9 11 76
62
67 106 75 8 41 42 61 60
101 57 47 70 21 63
88 33 31
75 79 8 11 15 66
101 39 15 4313 65
15 51 12 21 78 46 23 82 52 29 12 69
58
74 55
50 20 16 10 14 77 53 64
99 76 80 35
7
9 76
8
49 54
42 50 111
106
38
17 27 112 56
71
24 18 28 92
54 51 37
36
2 84 19 13 49
11 43 30 102 93 91
72
7010 1 112 35
107 110 49 52 85 79
115
14 29 81 100 94
50
55
26 95 104 48 101 90
3 35 2287 95
73 113
27
28
40 108 89 78
4 112
94 103 12
86
84 105
16 42 39 105 103
5 20 91 80
17
41
39
92 100 75108 97
6 8246 38 93 83 44 48
38 41 2585 97 102 107
43 89 84
24 86 98 96
19 23 37 90
47
18 13 30
48 92 93 83 104 111 87
94 44 96 99 110 45
81 77 99 106 57
87
40 114 98
34 95 36 3
22 45 11425
(d) “market” and “groups” modes, in 2001 (e) “market” and “groups” modes, in 2002 (f) “market” and “groups” modes, in 2003
92 24 74 43
23 111 88 32 42
39 30
20 29 63
38 40 31 7 41
99
93 16 65 33 34 3893 40
95
66 92
60
74 13 17 64
41 27 56 61
28 53 94
67 68
94
21 44
53 7
9 8 69 9139 9
101
95 109
63
58 62 8 29
45 87 21 14 10
1 47 50 52 100 96 1117584 10 68 13 20
59 14 30
51 49 102
107 97 69 58 65 45 107 110 11
17 16
29 112 61 11 77 15 76 27 46 28
55 9990 59 64 98 3 82 47105
46
735 56 104 62 66 109 90 10878 48
67
22
19
30
115
5491 12 108110 83
60
15 96
114 103
106 35 87
70 48 89 103 76 83
47 44 97 89104
4 26 85 25
81 18
113 105106 57 25
77
86 84 61 63 96 98 7921 10237 36 57
42 43
114
49 97 109 53 44 85 86
72 80 78 98 3 85 90 58 45 25 12
2 37 37 62 64 83 80 100
107
79 35 3610479 78 89 108 59 65 19 75 52
71 23
15 36 48
22
11099 82 69 111 114
26 84 18
76 77 19
12
75 60
66 3 101
6
24 82
57 80 103 105 68 22
102 55
106 81 87 54
100 81 49
91 18 67 51 50
11 14
8 39 52
112
16 101
20
10 13 41 35
27 17 93 92 38 56 5150 72 1
73 70
115
9 4
28 34 43 42 55
113 112 113
72
34 115 1
94 54 46
86 33
7
74 95 40
5
73 5 6
31 33 31 71 2
4 70 2 6 23
32 32 71
26
88 88 24
(g) “market” and “groups” modes, in 2004 (h) “market” and “groups” modes, in 2005 (i) “market” and “groups” modes, in 2006
42 85 22
7 74 80
78 87
54 47 28 90
43 10
46 20
86 16 97
4 45
5 27 99
55
48 11
17
8 48 83
111
115 18 19
70
50 51 75 14 13 44 84
1 113 56
47 9630 25
73 6 26 49 45
21 81 29 114
66
72 76 77 46 59 107
65
15 62
2 52101 7 64 60 89 21 3
71
109 61 67 6 41 91 110 60
23 100
74 8768 35 64
57 112 2043 95 405 58
75 9 58 62 106 100 55
11 10 69 4 112 16 61 109
27 48 101 90 82 23 42 69
24 13 21 28 86 56 53
79
107 113 39 63
107108 102 30
14
8 50 102 59 110 104 108 38
7393
82 12 17 51 52 97 13 24 70 37
68 15 76 4465 66 12 71 36
67 109 55 10581 63 19 80 22 75 102
16 20 92
87 77 29 34 57 9
98 111 18 26 72
2 115
53
99 49 37 88 54 76 52
59 79 18 34
110 12103
71 77 85 94 101
62 89 32 31 15 33
33 19 81 95 94 8
36 103 91 92 40 105
29 2491 9
35 93 4 1 56
25 85 37 5 104
58 106 41 39
22 82 70 57
104 78 38 113 11 14 98 51
60 61 69 83 105 98 35 30 3447 112 73
72 115
89 50 100 106
80 39 36 43 32 88
53 84 95 23 103
25 83
94
64
96 97
38 33 96
1 27 10 78 68 54
3 42 46 32 28
7 79
67
31
65 90 92 93 41 45 2 108
84
66 63 114 44 26 88 6 99 49
40 114
3
86
111 31 74 17
(j) “market” and “groups” modes, in 2007 (k) “market” and “groups” modes, in 2008 (l) “market” and “groups” modes, in 2009
45 6 4 23 36
73 115 13 86
48
98 5
54 72 70 113 31
53 34
55 24 35 10
112 2 101 82 29
50 51 56
1
52 27
105 99 7520 28 88
33
98 12
10252 101
81 50 48 50 37 32
18 91
51 56 19 30
101 1852 58 98 48
102 63 53 51 100 11
20 111 69 55 55 80 87
1 19 49 47
2
53 109 54 104 81 45
72
66 65
60 56 57 21 68 99
22 97
115
22
44
29 64 54 46
89 67
79 10396 84
5 73 58 78 44
6 70 69 80112 30 61 105 42 95
4 21 62 77 106
96 17 40 90
11336 33 59 43 17 78
71 97 84 13 16 31 107 108 107
41 7 58
63 77 74 2423
12 86 110 44 94 74
60 57 10 34
35 37 7 25 88 32 100 89 114 92 14 25 64
64 106 6790 7528 82
14 76 8 96 83
3
102 8
61 69
62 90
49
68 97 84 25 38
9 76 26 83 59 62 60
68 27 83 87 21 93
88 34 47
108
11 15 26 9 18 109
89 59
33 32 104 85 39 110
31 43 79 103 114
85 36
37
35 74 43
13 108
67
80 22
3
107 49 103 11 19
65 46 45 6515 114
66 61 100 110 87 3 7
79 82 81 26 16
42 104 57 10646 111
109 91 8 47 12 23 20 113 112 66 85
77 75 9 24
71
95 99 105
10 91 29 71 6
73
111 38 14 76 2728 95 41 5
63
70 115
15 16 78 4
39 38 72
17 86
40 39 92 1
41 92 40
30 42 94
93 94 93 2
(m) “market” and “groups” modes, in 2010 (n) “market” and “groups” modes, in 2011 (o) “market” and “groups” modes, in 2012
Fig. 18 The interaction structure between banks for the correlation matrix C m;g obtained with the PMFG
graph, in layer 1, from 1998 to 2012
123
1.4 1.4 1.4

1.2 1.2 1.2
1 1 1
0.8 0.8 0.8

ρ(λ)
ρ(λ)
ρ(λ)
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
0 5 10 15 0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
λ λ λ
(a) Dist. of λ, layer 1, 1980 (b) Dist. of λ, layer 2, 1980 (c) Dist. of λ, layer 3, 1980
1.5 1.4 1.4
1.2 1.2
1 1
1
0.8 0.8
ρ(λ)
ρ(λ)
ρ(λ)
0.6 0.6
0.5
0.4 0.4
0.2 0.2
0 0 0
0 2 4 6 8 10 0 1 2 3 4 0 2 4 6 8 10 12
λ λ λ
(d) Dist. of λ, layer 1, 2012 (e) Dist. of λ, layer 2, 2012 (f) Dist. of λ, layer 3, 2012
Fig. 19 Binary data: Distribution of eigenvalues of C bin , compared with RMT, in the three layers, in 1980
and 2012. Panels a–c show the distributions of of k in 1980, and panels d–f show the distributions of of k
in 2012. In all panels, the red curve represents the spectral distribution explained by the Marchenko-
Pastur law .(Colour figure online)
Binary Analysis
Using the similar methods, we can also analyse the binary version of the bank-firm
credit relationships. Interestingly, in general, our results suggest that this version
also contains genuine information about the structure and dynamics of the cross-
correlation matrix of banks’ loan portfolios.
Recalling that under the binary case, for each pair (i, j), we have
( bf
ai;j ¼ 1 if bank i lends to firm j;
ð24Þ
abf
i;j ¼ 0 otherwise:
We then define the matrix X bin ¼ fXi;j;bin gNxNF with each element is given by
abf bf
i;j hai;j i
Xi;j;bin ¼ ; ð25Þ
rðabf
i;j Þ
where for each bank i, habf bf

i;j i and rðai;j Þ are the average and the standard deviation of
abf
i;j over all NF firms, respectively. The correlation matrix C
bin
of the binary data is
then defined as
1
C bin ¼ fCijbin gNxN ¼ X bin X Tbin ; ð26Þ
NF
with X Tbin is the transposition matrix of X bin .
For the sake of conciseness, in the following we will only summarize the main
results for the analysis of the binary data. First, we find that a group of largest
123
562 D. T. Luu
0.3
E
1 1
16 E2
2
0.25 E
3 3
14
E
4 4
Largest Eigenvalues
E5
12
Absorption Ratios
5 0.2
* E
max 10
10
0.15
8
6 0.1
4
0.05
2
0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years
(a) Top eigenvalues, layer 1 (b) Absorption ratios, layer 1

1
E1
0.25 E2
12 2
3
E3
4
E4
10 0.2
Largest Eigenvalues
5
Absorption Ratios E5
* E
max 10
8
0.15
6
0.1
4
0.05
2
0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years
(c) Top eigenvalues, layer 2 (d) Absorption ratios, layer 2

1 0.4 E1
25 E
2 2
0.35 E
3 3
E
4 4
20 0.3
Largest Eigenvalues
E5
Absorption Ratios
5
* E
10
max 0.25
15
0.2
10 0.15
0.1
5
0.05
0 0
1980 1984 1988 1992 1996 2000 2004 2008 2012 1980 1984 1988 1992 1996 2000 2004 2008 2012
years years
(e) Eigenvalues, layer 3 (f) Absorption ratios, layer 3

Fig. 20 Binary data: Evolution of the top 5 largest eigenvalues and the associated absorption ratios for
C bin in the three layers. The panels on the left show the evolution of the top 5 largest eigenvalues. These
eigenvalues are always larger than the upper bound kmax implied by RMT. The three panels on the right
show their associated absorption ratios as well as the absorption ratio of the top 10 largest eigenvalues
eigenvalues (typically from 10 % to 15 %) always deviate from the random bulk

predicted by RMT (see Fig. 19). In addition, we often observe that more than half of
the eigenvalues lie between the two RMT bounds kmin and kmax .
Compared to the weighted data, we also find a similar evolution dynamics for the
top largest eigenvalues and their absorption ratios, although in general the factors
extracted from these eigenvalues contribute slightly less to the total variance in the
123
0.16 0.7
IPR IPR
0.14 0.6
0.12
0.5
0.1
0.4
IPR(λ)
IPR(λ)
0.08
0.3
0.06
0.2
0.04
0.02 0.1
00 0
5 10 15 −2 0 2 4 6 8 10
λ λ
(a) IPR, layer 1, 1980 (b) IPR, layer 1, 2012

0.35 0.7
IPR IPR
0.3 0.6
0.25 0.5
0.2 0.4
IPR(λ)
IPR(λ)
0.15 0.3
0.1 0.2
0.05 0.1
0 0
0 2 4 6 8 10 12 14 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 4
λ λ
(c) IPR, layer 2, 1980 (d) IPR, layer 2, 2012

0.35 0.5
IPR IPR
avg. of IPR 0.45 avg. of IPR
0.3
0.4
0.25 0.35
0.3
0.2
IPR(λ)
IPR(λ)
0.25
0.15
0.2
0.1 0.15
0.1
0.05
0.05
0 0
−2 0 2 4 6 8 10 12 14 −2 0 2 4 6 8 10 12
λ λ
(e) IPR, layer 3, 1980 (f) IPR, layer 3, 2012
Fig. 21 Binary data: Inverse Participation Ratios of the eigenvectors of C bin in the three layers, in 1980
and 2012. Panels a, b show IPR versus k in the layer 1, in 1980 and 2012. Panels c, d show IPR versus k
in the layer 2, in 1980 and 2012. Panels e, f show IPR versus k in the layer 3, in 1980 and 2012. The red
lines stand for the averages of fIPRk gk¼N
k¼1
binary data than in the weighted data (e.g. see Fig. 20, in comparison to Fig. 4 in the
main text).
Furthermore, as demonstrated in Fig. 21, the localization behavior still emerges
in the eigenvectors corresponding to the smallest eigenvalues, as their IPR ratios are
often much higher than the average of IPR. Besides this, again many banks
contribute to the eigenvectors corresponding to the top largest eigenvalues,
123
564 D. T. Luu
signalling that the ‘‘market’’ mode as well as some of the ‘‘groups’’ modes extracted
from the binary version of the credit network also have a wide effect.
Acknowledgements I am indebted to the editors and anonymous reviewers. I am grateful to Prof. Thomas
Lux, for valuable comments and discussions that helped to improve the paper. I also thank participants to
the different conferences and seminars where earlier versions of this paper were presented. These include:
the Conference on Complex System 2018 (CCS 2018) in Thessaloniki, Greece, September 24–28, 2018;
the Research Seminars at the Chair of Monetary Economics and International Finance, the University of
Kiel, Germany. All remaining errors are my own.
Funding Open Access funding enabled and organized by Projekt DEAL.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,
which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as
you give appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended
use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain
permission directly from the copyright holder. To view a copy of this licence, visit http://
creativecommons.org/licenses/by/4.0/.
References
Acemoglu, D., Ozdaglar, A., & Tahbaz-Salehi, A. (2015). Systemic risk and stability in financial
networks. American Economic Review, 105(2), 564–608.
Alfaro, L., Garcı́a-Santana, M., & Moral-Benito, E. (2018). On the direct and indirect real effects of credit
supply shocks. CEPR Discussion Papers 12794, CEPR Discussion Papers.
Allen, F., & Gale, D. (2000). Financial contagion. Journal of Political Economy, 108(1), 1–33.
Almog, A., Besamusca, F., MacMahon, M., & Garlaschelli, D. (2015). Mesoscopic community structure
of financial markets revealed by price and sign fluctuations. PLOS ONE, 10(7), 1–16.
Alvarez-Socorro, A. J., Herrera-Almarza, G. C., & González-Dı́az, L. A. (2015). Eigencentrality based on
dissimilarity measures reveals central nodes in complex networks. Scientific Reports, 5, 17095.
Amiti, M., & Weinstein, D. E. (2018). How much do idiosyncratic bank shocks affect investment?
Evidence from matched bank-firm loan data. Journal of Political Economy, 126(2), 525–587.
Aoyama, H. (2014). Systemic risk in Japanese credit network. In H. Aoyama, B. K. Chakrabarti, A.
Chakraborti, A. Ghosh, & F. Abergel (Eds.), Econophysics of Agent-Based Models (pp. 219–228).
Berlin: Springer International Publishing.
Aoyama, H., Stefano, B., & Fujiwara, Y. (2013). DebtRank Analysis of the Japanese credit network.
RIETI Discussion Paper Series 13-E-087, The Research Institute of Economy, Trade and Industry,
Japan.
Aste, T., Shaw, W., & Matteo, T. D. (2010). Correlation structure and dynamics in volatile markets. New
Journal of Physics, 12(8), 085009.
Azariadis, C. (2018). Credit cycles and business cycles. Federal Reserve Bank of St Louis Review, 100(1),
45–71.
Bai, Z., Liu, H., & Wong, W. K. (2009). Enhancement of the applicability of markowitz’s portfolio
optimization by utilizing random matrix theory. Mathematical Finance, 19(4), 639–667.
Bandyopadhyay, J. N., & Jalan, S. (2007). Universality in complex networks: Random matrix analysis.
Physical Review E, 76, 026109.
Barigozzi, M., Brownlees, C., Gallo, G. M., & Veredas, D. (2014). Disentangling systematic and
idiosyncratic dynamics in panels of volatility measures. Journal of Econometrics, 182(2), 364–384.
Bassett, W. F., Chosak, M. B., Driscoll, J. C., & Zakrajšek, E. (2014). Changes in bank lending standards
and the macroeconomy. Journal of Monetary Economics, 62, 23–40.
123
Beale, N., Rand, D. G., Battey, H., Croxson, K., May, R. M., & Nowak, M. A. (2011). Individual versus
systemic risk and the Regulator’s Dilemma. Proceedings of the National Academy of Sciences,
108(31), 12647–12652.
Bentolila, S., Jansen, M., & Jiménez, G. (2018). When credit dries up: Job losses in the great recession.
Journal of the European Economic Association, 16(3), 650–695.
Bianconi, G. (2018). Multilayer networks: Structure and function. Oxford: Oxford University Press.
Billio, M., Getmansky, M., Lo, A. W., & Pelizzon, L. (2012). Econometric measures of connectedness
and systemic risk in the finance and insurance sectors. Journal of Financial Economics, 104(3),
535–559.
Biroli, G., & Bouchaud, J. P. (2007). Potters M (2007) Extreme value problems in random matrix theory
and other disordered systems. Journal of Statistical Mechanics Theory and Experiment, 07, P07019.
Bodnar, T., Mazur, S., & Okhrin, Y. (2013). On the exact and approximate distributions of the product of
a wishart matrix with a normal vector. Journal of Multivariate Analysis, 122, 70–81.
Bodnar, T., Mazur, S., & Okhrin, Y. (2014). Distribution of the product of singular wishart matrix and
normal vector. Theory of Probability and Mathematical Statistics pp. 1–14.
Bodnar, T., Mazur, S., & Podgórski, K. (2016). Singular inverse wishart distribution and its application to
portfolio theory. Journal of Multivariate Analysis, 143, 314–326.
Bodnar, T., Mazur, S., & Podgórski, K. (2017). A test for the global minimum variance portfolio for small
sample and singular covariance. AStA Advances in Statistical Analysis, 101(3), 253–265.
Bodnar, T., Mazur, S., Muhinyuza, S., & Parolya, N. (2018). On the product of a singular wishart matrix
and a singular gaussian vector in high dimension. Theory of Probability and Mathematical Statistics,
99, 37–50.
Bodnar, T., Mazur, S., Podgórski, K., & Tyrcha, J. (2019). Tangency portfolio weights for singular
covariance matrix in small and large dimensions: Estimation and test theory. Journal of Statistical
Planning and Inference, 201, 40–57.
Bonacich, P. (2007). Some unique properties of eigenvector centrality. Social Networks, 29(4), 555–564.
Bouchaud, J. P., & Potters, M. (2009). Financial applications of random matrix theory: A short review.
ArXiv e-prints.
Bouchaud, J. P., Laloux, L., Miceli, M. A., & Potters, M. (2007). Large dimension forecasting models and
random singular value spectra. European Physical Journal B, 55(2), 201–207.
Bowman, D., Cai, F., Davies, S., & Kamin, S. (2015). Quantitative easing and bank lending: Evidence
from Japan. Journal of International Money and Finance, 57, 15–30.
Brummitt, C. D., D’Souza, R. M., & Leicht, E. A. (2012). Suppressing cascades of load in interdependent
networks. Proceedings of the National Academy of Sciences, 109(12), E680–E689.
Brunetti, C., Harris, J., Mankad, S., & Michailidis, G. (2015). Interconnectedness in the interbank market.
Finance and Economics Discussion Series 2015-90, Board of Governors of the Federal Reserve
System (U.S.).
Brunnermeier, M. K., & Schnabel, I. (2016). Bubbles and Central Banks: Historical Perspectives (pp.
493–562). Cambridge: Cambridge University Press.
Buldyrev, S. V., Parshani, R., Paul, G., Stanley, H. E., & Havlin, S. (2010). Catastrophic cascade of
failures in interdependent networks. Nature, 464(7291), 1025–1028.
Bülbül, D., & Lambert, C. (2012). Credit portfolio modelling and its effect on capital requirements.
Discussion Papers 11/2012, Deutsche Bundesbank.
Caccioli, F., Shrestha, M., Moore, C., & Farmer, J. D. (2014). Stability analysis of financial contagion due
to overlapping portfolios. Journal of Banking and Finance, 46, 233–245.
Caccioli, F., Farmer, J. D., Foti, N., & Rockmore, D. (2015). Overlapping portfolios, contagion, and
financial stability. Journal of Economic Dynamics and Control, 51, 50–63.
Cai, J., Eidam, F., Saunders, A., & Steffen, S. (2018). Syndication, interconnectedness, and systemic risk.
Journal of Financial Stability, 34, 105–120.
Chamberlain, G., & Rothschild, M. (1983). Arbitrage, factor structure, and mean-variance analysis on
large asset markets. Econometrica, 51(5), 1281–1304.
Cingano, F., Manaresi, F., & Sette, E. (2016). Does credit crunch investment down? New evidence on the
real effects of the bank-lending channel. The Review of Financial Studies, 29(10), 2737–2773.
Corsi, F., Marmi, S., & Lillo, F. (2016). When micro prudence increases macro risk: The destabilizing
effects of financial innovation, leverage, and diversification. Operations Research, 64(5),
1073–1088.
123
566 D. T. Luu
Daly, J., Crane, M., & Ruskin, H. (2008). Random matrix theory filters in portfolio optimisation: A
stability and risk assessment. Physica A Statistical Mechanics and Its Applications, 387(16),
4248–4260.
de Carvalho, J. X., Jalan, S., & Hussein, M. S. (2009). Deformed gaussian-orthogonal-ensemble
description of small-world networks. Phys Rev E, 79, 056222.
de Castro Miranda, R. C., & Tabak, B. M. (2013). Contagion risk within firm-bank bivariate networks.
Working Papers Series 322, Central Bank of Brazil, Research Department.
De Masi, G., & Gallegati, M. (2012). Bank-firms topology in Italy. Empirical Economics, 43(2), 851–866.
De Masi, G., Fujiwara, Y., Gallegati, M., Greenwald, B., & Stiglitz, J. E. (2011). An analysis of the
Japanese credit network. Evolutionary and Institutional Economics Review, 7(2), 209–232.
Di Matteo, T., Pozzi, F., & Aste, T. (2010). The use of dynamical networks to detect the hierarchical
organization of financial market sectors. The European Physical Journal B, 73(1), 3–11.
Dumitriu, I., & Johnson, T. (2016). The Marcenko-Pastur law for sparse random bipartite biregular
graphs. Random Structures and Algorithms, 48(2), 313–340.
Fenech, J. P., Vosgha, H., & Shafik, S. (2015). Loan default correlation using an Archimedean copula
approach: A case for recalibration. Economic Modelling, 47, 340–354.
Forni, M., & Lippi, M. (2001). The generalized dynamic factor model: Representation theory.
Econometric Theory, 17(6), 1113–1141.
Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2000). The generalized dynamic-factor model:
Identification and estimation. The Review of Economics and Statistics, 82(4), 540–554.
Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2015). The generalized dynamic factor model: One-sided
estimation and forecasting. Journal of the American Statistical Association, 100(471), 830–840.
Forni, M., Hallin, M., Lippi, M., & Zaffaroni, P. (2017). Dynamic factor models with infinite-dimensional
factor space: Asymptotic analysis. Journal of Econometrics, 199(1), 74–92.
Fricke, D. (2012). Trading strategies in the overnight money market: Correlations and clustering on the
e-MID trading platform. Physica A Statistical Mechanics and Its Applications, 391(24), 6528–6542.
Fricke, D. (2016). Has the banking system become more homogeneous? Evidence from banks’ loan
portfolios. Economics Letters, 142, 45–48.
Fricke, D., & Roukny, T. (2020). Generalists and specialists in the credit market. Journal of Banking and
Finance, 112, 105335.
Fujiwara, Y., Aoyama, H., Ikeda, Y., Iyetomi, H., & Souma, W. (2009). Structure and temporal change of
the credit network between banks and large firms in Japan. Economics: The Open-Access. Open-
Assessment E-Journal, 3, 1–18.
Gambetti, L., & Musso, A. (2017). Loan supply shocks and the business cycle. Journal of Applied
Econometrics, 32(4), 764–782.
Gertler, M., & Kiyotaki, N. (2010). Chapter 11 - financial intermediation and credit policy in business
cycle analysis. Handbook of Monetary Economics, vol 3, Elsevier, pp 547–599.
Gewers, F. L., Ferreira, G. R., de Arruda, H. F., Silva, F. N., Comin, C. H., Amancio, D. R., &
da F Costa, L. (2018). Principal component analysis: A natural approach to data exploration.
arXiv:CoRR abs/1804.02502.
Greenwood, R., Landier, A., & Thesmar, D. (2015). Vulnerable banks. Journal of Financial Economics,
115(3), 471–485.
Guhr, T., Müller-Groeling, A., & Weidenmüller, H. A. (1998). Random-matrix theories in quantum
physics: Common concepts. Physics Reports, 299(4), 189–425.
Haldane, A. G., & May, R. M. (2011). Systemic risk in banking ecosystems. Nature, 469, 351–355.
Himino, R. (2021). The Japanese Banking Crisis. Singapore: Palgrave Macmillan.
Hoshi, T., & Kashyap, A. (2000). The Japanese banking crisis: Where did it come from and how will it
end? In: NBER Macroeconomics Annual 1999, Volume 14, National Bureau of Economic Research,
Inc, pp 129–212.
Huang, X., Gao, J., Buldyrev, S. V., Havlin, S., & Stanley, H. E. (2011). Robustness of interdependent
networks under targeted attack. Physical Review E, 83, 065101.
Huang, X., Vodenska, I., Havlin, S., & Stanley, H. E. (2013). Cascading failures in bi-partite graphs:
Model for systemic risk propagation. Scientific Reports, 3, 1219.
Iyetomi, H., Nakayama, Y., Yoshikawa, H., Aoyama, H., Fujiwara, Y., Ikeda, Y., & Souma, W. (2011).
What causes business cycles? Analysis of the Japanese industrial production data. Journal of the
Japanese and International Economies, 25(3), 246–272.
Jalan, S. (2009). Spectral analysis of deformed random networks. Physical Review E, 80, 046101.
123
Jiang, X., Chen, T., & Zheng, B. (2014). Structure of local interactions in complex financial dynamics.
Scientific Reports, 4, 5321.
Jiang, X. F., & Zheng, B. (2012). Anti-correlation and subsector structure in financial systems. EPL
(Europhysics Letters), 97(4), 48006.
Jolliffe, I. (1986). Principal Component Analysis. Berlin: Springer Verlag.
Jorda, O., Schularick, M. H. P., & Taylor, A. M. (2013). When credit bites back. Journal of Money, Credit
and Banking, 45(s2), 3–28.
Kawai, M., & Takagi, S. (2011). Why was Japan hit so hard by the global financial crisis? In: The impact
of the economic crisis on East Asia, Edward Elgar Publishing, chap 7.
Khwaja, A. I., & Mian, A. (2008). Tracing the impact of bank liquidity shocks: Evidence from an
emerging market. American Economic Review, 98(4), 1413–42.
Kim, D. H., & Jeong, H. (2005). Systematic analysis of group identification in stock markets. Physical
Review E, 72, 046133.
Kiyotaki, N., & Moore, J. (1997). Credit cycles. Journal of Political Economy, 105(2), 211–248.
Kritzman, M., Li, Y., Page, S., & Rigobon, R. (2011). Principal components as a measure of systemic
risk. The Journal of Portfolio Management, 37(4), 112–126.
Kyriakopoulos, F., Thurner, S., Puhr, C., & Schmitz, S. W. (2009). Network and eigenvalue analysis of
financial transaction networks. The European Physical Journal B, 71(4), 523.
Laloux, L., Cizeau, P., Bouchaud, J. P., & Potters, M. (1999). Noise dressing of financial correlation
matrices. Physical Review Letters, 83, 1467–1470.
Laloux, L., Cizeau, P., Potters, M., & Bouchaud, J. P. (2000). Random matrix theory and financial
correlations. International Journal of Theoretical and Applied Finance, 03(03), 391–397.
Lehmann, S., Schwartz, M., & Hansen, L. K. (2008). Biclique communities. Physical Review E, 78,
016108.
Levy-Carciente, S., Kenett, D. Y., Avakian, A., Stanley, H. E., & Havlin, S. (2015). Dynamical
macroprudential stress testing using network theory. Journal of Banking and Finance, 59, 164–181.
Lillo, F., & Pirino, D. (2015). The impact of systemic and illiquidity risk on financing with risky
collateral. Journal of Economic Dynamics and Control, 50, 180–202.
Liu, R. R., Li, M., & Jia, C. X. (2016). Cascading failures in coupled networks: The critical role of node-
coupling strength across networks. Scientific Reports, 6, 35352.
Luu, D. T., & Lux, T. (2018). Identifying patterns in the bank-sector credit network of Spain. Journal Of
Network Theory in Finance, 4(1), 1–38.
Luu, D. T., & Lux, T. (2019). Multilayer overlaps and correlations in the bank-firm credit network of
Spain. Quantitative Finance, 19(12), 1953–1974.
Lux, T. (2016). A model of the topology of the bank - firm credit network and its role as channel of
contagion. Journal of Economic Dynamics and Control, 66, 36–53.
Lux, T. (2020). On the distribution of links in financial networks: Structural heterogeneity and functional
form. Empirical Economics, 58(3), 1019–1053.
Lux, T., Luu, D. T., & Yanovski, B. (2020). An analysis of systematic risk in worldwide econonomic
sentiment indices. Empirica, 47, 909–928.
MacMahon, M., & Garlaschelli, D. (2015). Community detection for correlation matrices. Physical
Review X, 5, 021006.
Mantegna, R. N. (1999). Hierarchical structure in financial markets. European Physical Journal B, 11(1),
193–197.
Marotta, L., Miccichè, S., Fujiwara, Y., Iyetomi, H., Aoyama, H., Gallegati, M., & Mantegna, R. N.
(2015). Bank-firm credit network in Japan: An analysis of a bipartite network. PLOS ONE, 10(5),
1–18.
Marotta, L., Miccichè, S., Fujiwara, Y., Iyetomi, H., Aoyama, H., Gallegati, M., & Mantegna, R. N.
(2016). Backbone of credit relationships in the Japanese credit market. EPJ Data Science, 5(1), 10.
Marčenko, V. A., & Pastur, L. A. (1967). Distribution of eigenvalues for some sets of random matrices.
Mathematics of the USSR-Sbornik, 1(4), 457.
Massara, G. P., di Matteo, T., & Aste, T. (2015). Network filtering for Big Data: Triangulated Maximally
Filtered Graph. arXiv:CoRR abs/1505.02445.
Matteo, B., & Marc, H. (2016a). Generalized dynamic factor models and volatilities: Recovering the
market volatility shocks. The Econometrics Journal, 19(1), C33–C60.
Matteo, B., & Marc, H. (2016b). A network analysis of the volatility of high dimensional financial series.
Journal of the Royal Statistical Society: Series C (Applied Statistics), 66(3), 581–605.
123
568 D. T. Luu
May, R. M., & Arinaminpathy, N. (2010). Systemic risk: The dynamics of model banking systems.
Journal of The Royal Society Interface, 7(46), 823–838.
Meng, H., Xie, W. J., Jiang, Z. Q., Podobnik, B., Zhou, W. X., & Stanley, H. (2014). Systemic risk and
spatiotemporal dynamics of the US housing market. Scientific Reports, 4, 3655.
Montagna, M., & Lux, T. (2017). Contagion risk in the interbank market: A probabilistic approach to
cope with incomplete structural information. Quantitative Finance, 17(1), 101–120.
Nier, E., Yang, J., Yorulmazer, T., & Alentorn, A. (2007). Network models and financial stability.
Journal of Economic Dynamics and Control, 31(6), 2033–2060 (Tenth Workshop on Economic
Heterogeneous Interacting Agents.).
Ormerod, P. (2008). Random matrix theory and macro-economic time-series: An illustration using the
evolution of business cycle synchronisation, 1886–2006. Economics the Open-Access Open-
Assessment E-Journal, 2, 1–10.
Ormerod, P., & Mounfield, C. (2000). Random matrix theory and the failure of macro-economic
forecasts. Physica A: Statistical Mechanics and its Applications, 280(3), 497–504.
Pappas, D., & Kaimakamis, G. (2010). Optimal portfolio selection with singular covariance matrix.
International Mathematical Forum, 5, 2305–2318.
Plerou, V., Gopikrishnan, P., Rosenow, B., Nunes Amaral, L. A., & Stanley, H. E. (1999). Universal and
nonuniversal properties of cross correlations in financial time series. Physical Review Letters, 83,
1471–1474.
Plerou, V., Gopikrishnan, P., Rosenow, B., Amaral, L., & Stanley, H. (2000). A random matrix theory
approach to financial cross-correlations. Physica A Statistical Mechanics and Its Applications,
287(3), 374–382.
Plerou, V., Gopikrishnan, P., Rosenow, B., Amaral, L. A. N., Guhr, T., & Stanley, H. E. (2002). Random
matrix approach to cross correlations in financial data. Physical Review E, 65, 066126.
Pool, V. K., Stoffman, N., & Yonker, S. E. (2015). The people in your neighborhood: Social interactions
and mutual fund portfolios. Journal of Finance, 70(6), 2679–2732.
Posen, A. S. (2003). It takes more than a bubble to become Japan. Working Paper Series WP03-9,
Peterson Institute for International Economics.
Potestio, R., Caccioli, F., & Vivo, P. (2009). Random matrix approach to collective behavior and bulk
universality in protein dynamics. Physical Review Letters, 103, 268101.
Pozzi, F., Aste, T., Rotundo, G., & Matteo, T. D. (2008). Dynamical correlations in financial systems. In:
Proceedings of SPIE, vol 6802, pp. 6802–6813.
Pukthuanthong, K., & Roll, R. (2009). Global market integration: An alternative measure and its
application. Journal of Financial Economics, 94(2), 214–232.
Quintana, D., Garcı́a-Rodrı́guez, S., Cincotti, S., & Isasi, P. (2015). Combining RMT-based filtering with
time-stamped resampling for robust portfolio optimization. International Journal of Computational
Intelligence Systems, 8(5), 874–885.
Reis, S. D. S., Hu, Y., Babino, A., Andrade, J. S., Jr., Canals, S., Sigman, M., & Makse, H. A. (2014).
Avoiding catastrophic failure in correlated networks of networks. Nature Physics, 10, 762–767.
Sharifi, S., Crane, M., Shamaie, A., & Ruskin, H. (2004). Random matrix theory for portfolio
optimization: A stability approach. Physica A Statistical Mechanics and Its Applications, 335(3),
629–643.
Shen, J., & Zheng, B. (2009). Cross-correlation in financial dynamics. EPL (Europhysics Letters), 86(4),
48005.
Smart, A. G., Amaral, L. A. N., & Ottino, J. M. (2008). Cascading failure and robustness in metabolic
networks. Proceedings of the National Academy of Sciences, 105(36), 13223–13228.
Srivastava, M. (2003). Singular wishart and multivariate beta distributions. The Annals of Statistics 31.
Tao, T., & Vu, V. (2012). Random covariance matrices: Universality of local statistics of eigenvalues.
The Annals of Probability, 40(3), 1285–1315.
Tran, L. V., Vu, V. H., & Wang, K. (2013). Sparse random graphs: Eigenvalues and eigenvectors.
Random Structures and Algorithms, 42(1), 110–134.
Tumminello, M., Aste, T., Di Matteo, T., & Mantegna, R. N. (2005). A tool for filtering information in
complex systems. Proceedings of the National Academy of Sciences, 102(30), 10421–10426.
Uechi, L., Akutsu, T., Stanley, H. E., Marcus, A. J., & Kenett, D. Y. (2015). Sector dominance ratio
analysis of financial markets. Physica A Statistical Mechanics and Its Applications, 421, 488–509.
Wang, D., Podobnik, B., Horvatić, D., & Stanley, H. E. (2011). Quantifying and modeling long-range
cross correlations in multiple time series with applications to world stock indices. Physical Review
E, 83, 046121.
123
Wang, G. J., Xie, C., & Stanley, H. E. (2018). Correlation structure and evolution of world stock markets:
Evidence from pearson and partial correlation-based networks. Computational Economics, 51(3),
607–635.
Yoshikawa, H., Aoyama, H., Fujiwara, Y., & Iyetomi, H. (2015). Deflation/inflation dynamics: Analysis
based on micro prices . CIRJE F-Series CIRJE-F-959, CIRJE, Faculty of Economics, University of
Tokyo.
Zheng, Z., Podobnik, B., Feng, L., & Li, B. (2012). Changes in cross-correlations as an indicator for
systemic risk. Scientific Reports, 2, 888.
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.
123

Credit-Market-of JapanComputational-Economics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Credit-Market-of JapanComputational-Economics

Uploaded by

Copyright:

Available Formats

Computational Economics (2022) 60:529–569

Portfolio Correlations in the Bank-Firm Credit Market

Duc Thi Luu1

Accepted: 15 July 2021 / Published online: 31 July 2021

Keywords Bank lending Portfolio correlations Systemic risk Random

& Duc Thi Luu

2 Data and Methods

2.2.1 Representation of a Bank-Firm Network, Portfolio Overlaps and Projection

firms). Each element wbf 7

With non-negative weights in the bank-firm credit network, we have that

AFF ¼ Afb Abf ; ð4Þ

indicates the number of common lenders that a pair of firms have.

2.2.2 Analysis of Portfolio Correlations

where for each bank i, hlnðwbf

Denote X ¼ fXij gNxNF , the cross-correlation matrix of N banks is then defined as

Generally speaking, if some of the eigenvalues of the empirical cross-correlation

2.2.2.3 Decomposition of Correlation Matrix We now introduce one of the most

with K ¼ diagfk1 ; . . .; kN g, and fUgNxN is an orthonormal matrix, whose ith column

ki ¼ uTi Cui ¼ uTi CovðX Þui ¼ VarðuTi X Þ: ð12Þ

and the rest of C is

which is the filtered matrix after the market-wide effect is subtracted.

and the rest

2.2.2.6 Complex Network Approach to Correlation Structure Analysis In order to

3 Portfolio Correlations in the Bank-Firm Credit Market of Japan

(a) Dist. of Cij , layer 1 (b) Dist. of Cij , layer 2

(c) Dist. of Cij , layer 3

0.4 0.4 0.4

0.2 0.2 0.2

-0.1 -0.1 -0.1

-0.2 -0.2 -0.2

(d) skewness(Cij ), layer 1 (e) skewness(Cij ), layer 2 (f) skewness(Cij ), layer 3

(g) kurtosis(Cij ), layer 1 (h) kurtosis(Cij ), layer 2 (i) kurtosis(Cij ), layer 3

1.4 1.5 1.4

(a) Largest eigenvalues, layer 1 (b) Absorption ratios, layer 1

(c) Largest eigenvalues, layer 2 (d) Absorption ratios, layer 2

(e) Largest eigenvalues, layer 3 (f) Absorption ratios, layer 3

observation is in agreement with those reported in analyses of stock returns (e.g.

(a) IPR, layer 1, 1980 (b) IPR, layer 1, 2012

(c) IPR, layer 2, 1980 (d) IPR, layer 2, 2012

(e) IPR, layer 3, 1980 (f) IPR, layer 3, 2012

0.8 0.8 0.8

40 0.4 0.4 40 0.4

0.2 0.2 0.2

-0.2 -0.2 -0.2

-0.8 -0.8 -0.8

Banks Banks Banks

(a) C in 2012, layer 1 (b) C in 2012, layer 2 (c) C in 2012, layer 3

0.8 0.8 0.8

40 0.4 0.4 0.4

0.2 0.2 0.2

-0.2 -0.2 -0.2

-0.8 -0.8 -0.8

Banks Banks Banks

(d) F C m in 2012, layer 1 (e) F C m in 2012, layer 2 (f) F C m in 2012, layer 3

0.8 0.8 0.8

40 0.4 0.4 0.4

-0.2 -0.2 -0.2

-0.8 -0.8 -0.8

Banks Banks Banks

(a) (b) (c)

(d) (e) (f)

103 108 19 85 74 10152 10 49 77 42 95

both ‘‘market’’ as well as ‘‘groups’’ modes (C m;g ¼ C m þ C g ). We then study the

Portfolio Overlaps and Similarities