The recent financial market turbulence caused considerable divergence in the banking market interest rate determination process of the euro area member countries (e.g. Illes and Lombardi 2013, Paries et al., 2014). The purpose of this study is to investigate the factors determining the banking market interest rates in the euro area countries during the pre-crisis and the post-crisis periods, and to highlight possible regional asymmetries in the interest rate determination processes. To this end, we employ a set of country specific factors, such as variables capturing macroeconomic conditions, financial risk and loans market conditions, together with common monetary policy factors at euro-zone level. Instead of using specific bank market interest rates, we base our analysis on the ECB’s harmonized cost of bank borrowing indicators of euro area members, in order to avoid cross-country and cross-product data heterogeneity. With the use of principal component analysis, we obtain a number of latent factors that describe unobserved movements in the cost of borrowing, originating either in certain Euro-zone regions or outside the euro area, or constitute common factors for all euro area members. Such factors are identified as macroeconomic conditions, financial risk, loans market conditions and euro area monetary policy variables. These obtained factors, are then used in order to estimate country specific structural equations of the cost of bank borrowing determination. Employing cluster analysis on the parameter coefficients of these models, we then identify euro area regions with similar characteristics regarding the determination of the cost of borrowing. Next, the member states are pooled within the regions identified and structural models are estimated for these regions. By comparing the estimated distinct regional models and the different dynamic effects of the latent factor shocks across the regions, we highlight the differences in the determination of the cost of bank borrowing between the euro-zone core and periphery, and how it has been evolved through the period of the 2007-9 global financial crisis and the subsequent euro area debt crisis.

1. The determination of the euro area banking market
interest rates in a period of financial turbulence: the
center-periphery asymmetry
George Michalopoulos* and Konstantinos Tsermenidis**
* Dpt. of Accounting & Finance, University of Macedonia, Greece, gmich@uom.gr
** Dpt. of Accounting & Finance, University of Macedonia, Greece, tsermeni@uom.edu.gr
At the 14th Biennial Athenian Policy Forum Conference on ‘“CONTEMPORARY ECONOMIC, FINANCIAL,
BUSINESS AND POLICY ISSUES”, University of Piraeus, Greece, July 6-8, 2018
Presented by Konstantinos Tsermenidis
1

2. The aims of this study
1. Explore the determinants of the banking market interest rates
of euro zone members
2. Identify structural factors that underlie the process of banking
interest rate convergence/divergence across countries,
focusing especially at the post-financial crisis banking
environment
3. Study the effects of the global financial and the subsequent
euro area dept crisis on the determination mechanism of the
cost of borrowing for firms and households
4. Identify certain Eurozone regions that exhibit similar behavior
in the determination of the cost of borrowing
2

3. Introduction
• Banking market integration within Euro area, has been much slower
compared to money market integration, especially at the retail banking
level, (Cabral et al. (2002), Affinito et al. (2006))
• Banking market integration in the Eurozone is a necessary condition for
the efficient conduction of the single monetary policy
• Possible Sources of the observed divergence: differences related to
business cycle, financial environment, regulations, structural
characteristics, barrier producing bank strategies, and asymmetrical
effects of the ECB monetary policy.
3

4. Introduction (cont.)
• The 2007-9 global financial crisis and the subsequent euro area debt crisis
disrupted the process of banking market integration within the euro area
(ECB, 2009-2012)
• The banking market interest rates have shown considerable divergence
across countries during the crisis years (ECB, 2013)
• Divergence can be observed both across countries, as well across Eurozone
regions (Louri and Migiakis, 2015)
• The purpose of this study is to investigate the factors explaining the
divergence of banking market interest rates among euro area regions, namely
the center and periphery for the two periods, before and after the beginning
of the crisis
4

5. Literature review
• Empirical evidence on banking interest rate convergence (an indicator of banking
integration), showed that the introduction of the single currency facilitated the
integration process in the euro zone banking market (Vajanne 2007, Cabral et al.
2002, Sander and Kleimeier 2003, Toolsema et al. 2002).
• The global financial crisis (2008-9) and the subsequent eurozone debt crisis,
disrupted this convergence process. Convergence indicators deteriorated, as the
prevailing segmentation of the Eurozone was transferred to the banking sector. Thus,
the member state banking market interest rates have shown considerable
divergence during the crisis years:
• ECB 2012, 2013 reported interest rate divergence in the post crisis period
• Karagiannis et al. 2010, Illes and Lombardi 2013, Paries et al. 2014 reported
heterogeneity of the monetary policy pass-through process on banks’ interest rates
• Neri, 2013, Acharya και Steffen, 2013, Gennaioli et al, 2014, Becker και Ivashina,
2014 reported negative effects of debt crisis and divergent sovereign bond yields on
banking integration 5

6. Methodology: Principal components
• Cost of borrowing variables
• (4 X 11 countries)
• Financial risk variables
• (2 X 11 of countries)
• Macroeconomic variables
• (3 X 11 of countries)
• Loans market variables
• (2 X 11 of countries)
• Monetary policy variables
• (5 X 11 countries)
At country level
• 1 factor for the cost of borrowing
• 1 factor of financial risk
• 2 macroeconomic factors
• 1 loans market factor
• 1 factor for the monetary policy
Common latent factors
• Different sets of variables were used in order to obtain separate factors from each set.
• Factors extracted from the same set of variables, are uncorrelated by definition.
• However, factors across different sets could be correlated
• In order to make them all uncorrelated, factors are orthogonalized by regressing each subsequent factor
with all previous ones and obtain the residuals as the orthogonalized version of each factor.
At Eurozone level
6

7. Methodology: Estimation of individual observation
equations for interest rate indicators
Xt : Individual cost of borrowing indicators
Fpt : Monetary policy factor
Fmt: Macroeconomic factors
Fft : Financial risk factor
Fbt : Bank loans market factor
Frt : Interest rate factor
ΛFp : Monetary policy factor loading
ΛFm : Macroeconomic factors loading
ΛFf : Financial risk factor loading
ΛFb : Bank loans market factor loading
ΛFr : Interest rate factor loading
Χt= ΛFr
Frt+ ΛFm
Fmt+ ΛF f
F ft+ ΛF p
Fpt+ ΛFb
Fbt+ ξt
• The equation, is estimated a) for individual countries, and b) pooling countries within Eurozone regions
• The regions are identified using clustering:
• K-means clustering, using 2 groups, for the center and periphery
• Hierarchical clustering to examine whether similar results may be obtained without imposing any a-priori
restriction on the number of groups in total and the number of groups each country may belong to
7

8. Data*
• Cost of borrowing indicators (ECB, 2013): Main variable of interest, a
better indicator of banking market interest rates across countries, avoids cross-country
and cross-product heterogeneity
• Non-financial institutions (NFIs)
• Households
• Financial risk:
• 10 year sovereign bond yield spreads
• Composite indicator of systemic stress (CISS) (Holló et al., 2012)
• Macroeconomic conditions:
• Unemployment rate
• CPI inflation rate
• Industrial production index
• Loans market conditions
• loans to deposits ratio
• Annual growth rate of loans
• Monetary policy indicators:
• EONIA rate
• EURIBOR of one month, three months, six months and one year maturities
• Austria
• Belgium
• Finland
• France
• Germany
• Greece
• Ireland
• Italy
• Portugal
• Netherlands
• Spain
• Eurozone level
* Monthly observations from January 2003 to June 2015 8

9. Note: Hierarchical clustering based on coefficients of models of regressing the latent factors on the cost of borrowing of individual countries.
Hierarchical clustering of countries based on the coefficients of individual cost of borrowing models
9

10. Results: Clustering of countries, based on estimated observation
equation coefficients
Note: Clustering of countries is based on the estimated coefficients of the individual regression models of the cost of
borrowing for NFIs and households for the periods before and after the beginning of the crises
NFI before NFI after Households before Households after
Austria 1 1 1 1
Belgium 1 1 1 2
Finland 1 1 2 1
France 1 1 1 2
Germany 1 1 1 1
Greece 2 2 1 1
Ireland 1 1 2 2
Italy 1 2 2 2
Netherlands 1 1 1 2
Portugal 2 2 2 2
Spain 2 2 2 2
10

11. Cost of bank borrowing for NFI
EA-CEN pre-crisis = EA-PER pre-crisis EA-CEN pre-crisis =EA-CEN post-crisis
Χ2
P-value Χ2
P-value
Macro 1 4,56 0,03 0,25 0,62
Macro 2 9,46 0,00 4,48 0,03
Financial risk 76,01 0,00 0,38 0,54
Loans market 0,10 0,75 1,27 0,26
Market rates 12,04 0,00 1,13 0,29
Cost of borrowing 0,02 0,89 1,75 0,19
EA-CEN post-crisis=EA-PER post-crisis EA-PER pre-crisis =EA-PER post-crisis
Macro 1 161,61 0,00 9,96 0,00
Macro 2 2,33 0,13 10,18 0,00
Financial risk 212,20 0,00 0,17 0,68
Loans market 87,80 0,00 27,82 0,00
Market rates 131,74 0,00 70,03 0,00
Cost of borrowing 4,83 0,03 4,35 0,04
Cost of bank borrowing for households
EA-CEN pre-crisis = EA-PER pre-crisis EA-CEN pre-crisis =EA-CEN post-crisis
Χ2
P-value Χ2
P-value
Macro 1 8,75 0,00 27,75 0,00
Macro 2 4,33 0,04 32,70 0,00
Financial risk 7,55 0,01 4,63 0,03
Loans market 12,76 0,00 52,89 0,00
Market rates 12,90 0,00 79,97 0,00
Cost of borrowing 4,90 0,03 11,74 0,00
EA-CEN post-crisis=EA-PER post-crisis EA-PER pre-crisis =EA-PER post-crisis
Macro 1 2,92 0,09 0,33 0,57
Macro 2 24,77 0,00 0,23 0,63
Financial risk 9,30 0,00 0,02 0,88
Loans market 8,36 0,00 0,04 0,83
Market rates 65,89 0,00 4,18 0,04
Cost of borrowing 0,74 0,39 0,00 0,96
Note: Wald tests for the difference of model coefficients for the regressions of the latent factors on the cost of borrowing of NFIs and households for the Euro area center (EA-CEN) and Euro
Area periphery (EA-PER) and for the periods before and after the beginning of the financial crises.
Χ2 Wald tests for the difference of model coefficients for Euro area center and periphery
• Models show significant
differences, across
regions even before the
beginning of the crises
• The differences continued
for the period that
followed
• Statistically similar
models were observed for
the cost of borrowing of
NFI for the center, as well
as for the cost of
borrowing of households
for periphery comparing
the two periods
11

12. Methodology: Estimation of vector auto-regression for the latent factors (A)
Zt= μ+ ∑i= 0
∞
Φi εt− i
Σ=PP'
P-1ΣP'-1=IK
• A vector autoregressive model is estimated from all latent factors (Zt= Fpt, Fmt ,Fft, Fbt, Frt),
where A's define the coefficients of the individual factors in the dynamic system:
• Ε(εt)=0, E(εt ε't)=Σ και Ε(εt εs)=0 για t≠s. Off-diagonal elements in Σ may be non-zero, meaning that shocks on
any of the equations of the system, may be contemporaneously correlated with shocks in other equations.
• The VAR system, given that it is stable, can have a moving average representation:
• We consider a Cholesky factorization of the covariance matrix of the reduced-form residuals, in order to identify the
structural shocks. The covariance matrix of the errors Σ is broken down respectively, while the matrix that defines the
contemporaneous relations P, is a lower triangular.
12

13. Methodology: Estimation of vector auto-regression for the latent factors (B)
• From the moving average representation and using the Cholesky decomposition, the system of equations takes the form:
Ζt= μ+ ∑
i= 0
∞
Φi εt− i
Ζt= μ+ ∑
i= 0
∞
Φi PP
− 1
εt− i
Ζt= μ+ ∑
i= 0
∞
Θi P
− 1
εt− i
Ζt= μ+ ∑
i= 0
∞
Θi wt− i
• From the above transformations, the resulting covariance matrix of the errors has all off-diagonal elements equal to zero,
allowing for inference regarding individual equation shocks and calculation of impulse responses and forecast error
variance decompositions.
• The structure imposed regarding the contemporaneous effects does not allow macroeconomic, loans market and financial
conditions to be affected by shocks in monetary policy or retail interest rates.
• However, under this structure, shocks originating from macroeconomic, banking or financial factors are allowed to
contemporaneously affect monetary policy and interest rates.
13

14. Results: Results on forecast error variance decompositions (FEVDs) for core and
periphery
Note: Clustering of countries is based on the estimated coefficients of the individual regression models of the cost of borrowing for NFIs and
households for the periods before and after the beginning of the crises
Factors explaining the forecast error variance decomposition of the cost of borrowing, in order of contribution
a) Cost of borrowing of Non-Financial Institutions (bank loans)
Eurozone Centre Eurozone Periphery
Pre-crisis period Post-crisis period Pre-crisis period Post-crisis period
Short -run Long-run Short -run Long-run Short -run Long-run Short -run Long-run
Macro II Cost of
borrowing
Macro II Macro II Macro I Macro I Financial
Risk
Financial
Risk
MacroI Loans
Market
Financial
Risk
Financial
Risk
Money
Market
Macro II Macro II Macro II
Macro I Macro I Macro I Financial
Risk
Loans
Market
Loans
Market
Macro II
b) Cost of borrowing of Households (bank loans)
Eurozone Centre Eurozone Periphery
Pre-crisis period Post-crisis period Pre-crisis period Post-crisis period
Short -run Long-run Short -run Long-run Short -run Long-run Short -run Long-run
Macro II Macro II Macro II Macro I Macro II Macro II Macro II Financial
Risk
Financial
Risk
Financial
Risk
Money
Market
Macro II Financial
Risk
Financial
Risk
Financial
Risk
Macro I
Macro I Macro I Loans
Market
Macro I Macro I Macro I Macro II
14

15. Factors’ shocks impacting cost of borrowing of
NFIs
Center pre-crisis:
In the short run, macroeconomic factors were dominant
In the long run, cost of borrowing was the most significant
Center post-crisis:
Macroeconomic factor II came first, while financial risk increased its role, both in
the short and long run
Periphery pre-crisis:
• Macroeconomic factor I played the most significant role in the short and long run
Periphery post-crisis:
• Financial risk became the dominant factor for both short and long run
15

16. Factors’ shocks impacting the cost of
borrowing of Households
Center pre-crisis:
• Macroeconomic factor II was the most significant for both short run and long run
Center post-crisis:
• Macroeconomic factor II remained significant, while monetary policy increased
its significance in the short run
• Macroeconomic factors prevailed in the long run
Periphery pre-crisis:
• Macroeconomic factor II and financial risk played significant role for both short
and long run
Periphery post-crisis:
• Financial risk became the most significant in the long run
16

17. Conclusions
1. Factors identified are macroeconomic conditions, financial risk, loans market
conditions and euro area monetary policy variables
2. The mechanisms which shape the cost of borrowing indicators across countries, show
significant differences.
3. It was also found that certain groups of countries, showed similar behavior regarding
the dynamics that shape the cost of borrowing. The similarities-differences found here,
fit well to the classification of the Eurozone countries in two separate regions, namely
eurozone core and eurozone periphery
4. It is shown here that the structural models determining the cost of borrowing in
eurozone core and peripheral countries already exhibited significant differences, even
before the beginning of the 2007 financial crisis.
17

18. Conclusions
5. A change in these models was also observed for the regions when comparing the pre-
crisis and post-crisis periods
6. Forecast error variance decompositions, show that the latent factor representing
“financial risk” has become the main contributor of the cost of borrowing variation in
the financially troubled eurozone periphery during the post-crisis period
7. On the contrary, for the eurozone center, factors related to macroeconomic conditions
are more important in explaining the cost of borrowing variation during the same
period.
8. Post-crisis cross region interest rate divergence is attributed not only to divergent
economic conditions but also to divergent interest rate structural models. Therefore,
convergence of economic policies and conditions is necessary but not sufficient for
achieving higher levels of integration in the eurozone banking markets in the future.
Structural convergence of the banking markets is also important for such integration.
18

19. References
• Acharya V. V., Steffen S., 2013, "The Greatest Carry Trade Ever? Understanding Eurozone Bank Risks", CEPR Discussion Papers 9432, C.E.P.R.
Discussion Papers.
• Affinito M., Farabullini F., 2006, “An empirical analysis of national differences in the retail bank interest rates of the euro area”, Temi di
discussione (Economic working papers) 589, Bank of Italy.
• Banerjee, A., M. Marcellino, and I. Masten, 2014, “Forecasting with factor augmented error-correction models”, International Journal of
Forecasting 30 589–612.
• Becker, B., and V. Ivashina, 2014, "Cyclicality of Credit Supply: Firm Level Evidence," Journal of Monetary Economics 62, 76-93.
• Becker, B. and Ivashina, V., 2014, "Financial Repression in the European Sovereign Debt Crisis", Swedish House of Finance, Research Paper
No. 14-13, Paris December 2014 Finance Meeting EUROFIDAI - AFFI Paper.
• Bernanke, B., J. Boivin, and P. Eliasz (2005), “Measuring the Effects of Monetary Policy: A actor-Augmented Vector Autoregressive (FAVAR)
Approach”, The Quarterly Journal of Economics, Vol. 120(1), pp. 387-422.
• Bernhofer D., van Treeck T., 2013, "New evidence of heterogeneous bank interest rate pass-through in the euro area", Economic Modeling,
Volume 35, September
• Cabral, I., Dierick F. and Vesala J. M., “Banking Integration in the Euro Area” (December 2002). ECB Working Paper No. 6.
• Von Borstel J. & Eickmeier S. & Krippner L., 2015. "The interest rate pass-through in the euro area during the sovereign debt crisis,"
Discussion Papers, Deutsche Bundesbank, Research Centre.
• de Bondt, G., 2005. Interest rate pass-through: Empirical results for the Euro Area. German Economic Review 6, 37-78.
• ECB (2013). Assessing the retail bank interest rate pass-through in the euro area at times of financial fragmentation. ECB Monthly Bulletin
August 2013, 75–91.
• Gennaioli N. and Martin A. and Rossi S., 2014, "Sovereign Default, Domestic Banks, and Financial Institutions", Journal of Finance, Volume
69, Issue 2, pages 819-866.
19

20. References
• Heinemann, F. and Schüler, M. (2002) Integration benefits on EU retail credit markets –Evidence from interest rate pass-through. ZEW
Discussion Papers 02-26.
• Hollo, D., Kremer M. and Lo Duca M. (2012), “CISS – A Composite Indicator of Systemic Stress in the Financial System”, ECB Working Paper
No. 1426.
• Illes A., Lombardi M. J., 2013, "Interest rate pass-through since the financial crisis," BIS Quarterly Review, Bank for International
Settlements, September.
• Louri, H., Migiakis P. (2015), “Determinants of euro-area bank lending margins: Financial fragmentation and ECB policies”, Athens : Bank of
Greece.
• Mojon B., 2000, "Financial structure and the interest rate channel of ECB monetary policy", Working Paper Series 0040, European Central
Bank.
• Neri S., 2013, "The impact of the sovereign debt crisis on bank lending rates in the euro area," Questioni di Economia e Finanza (Occasional
Papers) 170, Bank of Italy, Economic Research and International Relations Area.
• Neri St., Ropele T., 2015, The macroeconomic effects of the sovereign debt crisis in the euro area, Temi di discussione (Economic working
papers) 1007, Bank of Italy, Economic Research and International Relations Area.
• Sander H. and Kleimeier St., 2003, “Convergence in Eurozone Retail Banking? What Interest Rate Pass-Through Tells Us About Monetary
Policy Transmission, Competition and Integration”, Maastricht University LIFE Working Paper No. 03-009.
• Sorensen, C.K., Werner, T., 2006. Bank interest rate pass-through in the Euro area: A cross country comparison. ECB Working Paper No. 580.
• Toolsema, L. A., Sturm, J.-E. and de Haan J. (2002) Convergence of pass-through from money market to lending rates in EMU countries: new
evidence. CCSO Working Papers 200206.
• Vajanne L., 2007, “Integration in euro area retail banking markets – convergence of credit interest rates”, Research Discussion Papers
27/2007, Bank of Finland.
20

21. Impulse responses of latent factors on the cost of borrowing to NFIs and households for the center and periphery
NFIs Households
21

22. Forecast error variance decompositions for the center and periphery of Eurozone
22

23. 23
Bayesian estimation of vector auto-regression

24. • Regarding the prior beliefs and how these can be incorporated in the model, Sims and Zha (1998) give a set of hyper parameters that control for these beliefs. To be
more precise, the authors, based among others on Litterman (1986), give certain directions that make the estimation of the model more consistent and efficient.
The parameterizations done, have mainly to do with beliefs and assumptions common in the economics literature, like existence of unit root, common trends,
cointegration, behaviour of coefficients across consecutive lags, existence of constant term in the model, relation of the coefficients across the equations of the
system etc. Moreover, the initial parameterization is not only related to certain aspects regarding these characteristics, but also to the degree of confidence for
these settings.
• Coming to our model, the prior beliefs we set are: Existence of unit root, constant is zero, existence of common trends, the coefficients of consecutive lags are
gradually decreased and finally that there is not distinction between the coefficients of a factor in the equation of that factor. The parameters regarding the level of
confidence on these prior beliefs, are:
• λ0, that expresses the total dispersion on prior beliefs on the existence of random walk.
• λ1 that expresses the dispersion on prior beliefs of unit root.
• λ3 that controls for the decrease of the variance on the estimates of the lags’ coefficients.
• λ4 that controls for the dispersion of beliefs on the prior assumption that constants are zero.
• As λs decrease, the prior beliefs become stronger. As they increase, the confidence is being relaxed. Besides parameters related to the prior confidence, there are
also parameters in the form of dummy variables regarding the long term behaviour of the system:
• μ1, which controls for the parameterization of the system in first differences (the greater its value, the closer to the first differences form). μ1 defines the sum of
the lags’ coefficients.
• μ2, which expresses beliefs on the existence of cointegration and common trends. As μ2 is being increased, the series have common trends, while the constants are
zero.
• In the present study, the following values have been set regarding the parameters described:
•
• These values are not restrictive and someone could make an excaustive research trying different sets of values in a hierarchical structure of models, in order to
determine the best set in terms of explanatory and predictive power. However, a task like this is above the purposes of this study, which is based on the set of
parameters given by Sims and Zha (1998), Robertson and Tallman (1999) and Sims and Zha (2006). Morever, as the factors in the system were obtained from
principal component analysis and are uncorrelated, a small dispersion is chosen regarding the prior beliefs on the zero constant term in the equations.
24
Bayesian estimation of vector auto-regression
(λ0, λ1, λ3, λ4, μ5, μ6)= (0,6,0,1,1,2,0,01,5,5)